{"id":65300,"date":"2026-01-24T19:32:38","date_gmt":"2026-01-24T11:32:38","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/65300.html"},"modified":"2026-01-24T19:32:38","modified_gmt":"2026-01-24T11:32:38","slug":"python%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%ef%bc%9a%e4%bb%8e%e5%85%a5%e9%97%a8%e5%88%b0%e5%ae%9e%e6%88%98","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/65300.html","title":{"rendered":"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218"},"content":{"rendered":"<h2 style=\"background-color:transparent\">\u76ee\u5f55<\/h2>\n<h3>\u7b2c\u4e00\u90e8\u5206&#xff1a;\u57fa\u7840\u7bc7 \u2014\u2014 \u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3<\/h3>\n<h4>\u7b2c1\u7ae0&#xff1a;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u4e4b\u65c5<\/h4>\n<ul>\n<li>1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb\u3002<\/li>\n<li>1.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174\u3002<\/li>\n<li>1.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u751f\u6001\u3001\u793e\u533a\u4e0e\u54f2\u5b66\u7684\u7edf\u4e00\u3002<\/li>\n<li>1.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66\u3002<\/li>\n<li>1.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90\u3002<\/li>\n<\/ul>\n<h4>\u7b2c2\u7ae0&#xff1a;\u6570\u5b66\u4e0e\u7f16\u7a0b\u57fa\u7840 \u2014\u2014 \u5185\u529f\u5fc3\u6cd5<\/h4>\n<ul>\n<li>2.1\u00a0\u7ebf\u6027\u4ee3\u6570&#xff1a;\u5411\u91cf\u3001\u77e9\u9635\u3001\u5f20\u91cf\u53ca\u5176\u8fd0\u7b97&#xff08;\u4e0d\u4ec5\u662f\u8ba1\u7b97&#xff0c;\u66f4\u662f\u7a7a\u95f4\u7684\u53d8\u6362&#xff09;\u3002<\/li>\n<li>2.2\u00a0\u5fae\u79ef\u5206&#xff1a;\u5bfc\u6570\u3001\u504f\u5bfc\u6570\u3001\u94fe\u5f0f\u6cd5\u5219\u4e0e\u68af\u5ea6&#xff08;\u7406\u89e3\u53d8\u5316\u4e0e\u4f18\u5316\u7684\u8bed\u8a00&#xff09;\u3002<\/li>\n<li>2.3\u00a0\u6982\u7387\u8bba\u4e0e\u4fe1\u606f\u8bba&#xff1a;\u6982\u7387\u5206\u5e03\u3001\u671f\u671b\u3001\u71b5\u4e0e\u4ea4\u53c9\u71b5&#xff08;\u8861\u91cf\u4e0d\u786e\u5b9a\u6027\u4e0e\u4fe1\u606f&#xff09;\u3002<\/li>\n<li>2.4\u00a0NumPy&#xff1a;\u7cbe\u901a\u591a\u7ef4\u6570\u7ec4\u64cd\u4f5c&#xff0c;\u4e3a\u6570\u636e\u5904\u7406\u63d0\u901f\u3002<\/li>\n<li>2.5\u00a0Pandas&#xff1a;\u7ed3\u6784\u5316\u6570\u636e\u7684\u63a2\u67e5\u3001\u6e05\u6d17\u4e0e\u9884\u5904\u7406\u3002<\/li>\n<li>2.6 Matplotlib &amp; Seaborn&#xff1a;\u6570\u636e\u7684\u53ef\u89c6\u5316&#xff0c;\u8ba9\u6d1e\u5bdf\u76f4\u89c2\u5448\u73b0\u3002\u00a0<\/li>\n<\/ul>\n<h4>\u7b2c3\u7ae0&#xff1a;\u673a\u5668\u5b66\u4e60\u7ecf\u5178\u56de\u987e \u2014\u2014 \u6e29\u6545\u800c\u77e5\u65b0<\/h4>\n<ul>\n<li>3.1 \u673a\u5668\u5b66\u4e60\u7684\u4e09\u5927\u8303\u5f0f&#xff1a;\u95ee\u9053\u4e8e\u5929\u5730\u3002<\/li>\n<li>3.2 \u7ecf\u5178\u6a21\u578b\u5256\u6790&#xff1a;\u4e00\u82b1\u4e00\u4e16\u754c&#xff0c;\u4e00\u53f6\u4e00\u83e9\u63d0\u3002<\/li>\n<li>3.3 \u6a21\u578b\u7684\u8bc4\u4f30\u4e0e\u4f18\u5316&#xff1a;\u77e5\u5176\u7136&#xff0c;\u66f4\u8981\u77e5\u5176\u6240\u4ee5\u7136\u3002<\/li>\n<\/ul>\n<h3>\u7b2c\u4e8c\u90e8\u5206&#xff1a;\u6838\u5fc3\u7bc7 \u2014\u2014 \u6df1\u5165\u795e\u7ecf\u7f51\u7edc\u7684\u6bbf\u5802<\/h3>\n<h4>\u7b2c4\u7ae0&#xff1a;\u795e\u7ecf\u7f51\u7edc\u57fa\u7840<\/h4>\n<ul>\n<li>4.1 \u4ece\u751f\u7269\u795e\u7ecf\u5143\u5230\u4eba\u5de5\u795e\u7ecf\u5143&#xff1a;\u6a21\u578b\u7684\u7075\u611f\u6765\u6e90\u3002<\/li>\n<li>4.2 \u611f\u77e5\u673a\u6a21\u578b&#xff1a;\u6700\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc\u53ca\u5176\u5c40\u9650\u6027\u3002<\/li>\n<li>4.3 \u591a\u5c42\u611f\u77e5\u673a&#xff08;MLP&#xff09;&#xff1a;\u6784\u5efa\u6df1\u5ea6\u7f51\u7edc\u7684\u7b2c\u4e00\u6b65\u3002<\/li>\n<li>4.4 \u6fc0\u6d3b\u51fd\u6570\u5927\u5168&#xff1a;Sigmoid\u3001Tanh\u3001ReLU\u3001Leaky ReLU\u3001ELU\u7b49&#xff08;\u4e3a\u4f55\u9700\u8981\u975e\u7ebf\u6027&#xff09;\u3002<\/li>\n<li>4.5 \u635f\u5931\u51fd\u6570&#xff1a;MSE\u3001\u4ea4\u53c9\u71b5\u7b49&#xff08;\u8861\u91cf\u201c\u7406\u60f3\u201d\u4e0e\u201c\u73b0\u5b9e\u201d\u7684\u5dee\u8ddd&#xff09;\u3002<\/li>\n<li>4.6 \u53cd\u5411\u4f20\u64ad\u7b97\u6cd5&#xff1a;\u68af\u5ea6\u4e0b\u964d\u4e0e\u94fe\u5f0f\u6cd5\u5219\u7684\u5b8c\u7f8e\u7ed3\u5408&#xff08;\u7f51\u7edc\u5982\u4f55\u5b66\u4e60&#xff09;\u3002<\/li>\n<\/ul>\n<h4>\u7b2c5\u7ae0&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u5165\u95e8\u4e0e\u5b9e\u6218<\/h4>\n<ul>\n<li>5.1 TensorFlow 2.x \u4e0e Keras&#xff1a;Google\u7684\u5de5\u4e1a\u7ea7\u89e3\u51b3\u65b9\u6848\u3002<\/li>\n<li>5.2 PyTorch&#xff1a;Facebook\u7684\u52a8\u6001\u56fe\u4e0e\u7814\u7a76\u8005\u9996\u9009\u3002<\/li>\n<li>5.3 \u73af\u5883\u642d\u5efa&#xff1a;Conda\u3001Jupyter Notebook\u4e0eGPU\u914d\u7f6e\u6307\u5357\u3002<\/li>\n<\/ul>\n<h4>\u7b2c6\u7ae0&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u70bc\u4e39\u672f\u201d \u2014\u2014 \u8bad\u7ec3\u4e0e\u4f18\u5316<\/h4>\n<ul>\n<li>6.1 \u4f18\u5316\u5668\u8be6\u89e3&#xff1a;SGD\u3001Momentum\u3001Adagrad\u3001RMSprop\u3001Adam\u3002<\/li>\n<li>6.2 \u6b63\u5219\u5316\u6280\u672f&#xff1a;L1\/L2\u6b63\u5219\u5316\u3001Dropout\u3001\u65e9\u505c&#xff08;\u9632\u6b62\u6a21\u578b\u201c\u8d70\u706b\u5165\u9b54\u201d&#xff09;\u3002<\/li>\n<li>6.3 \u6279\u5f52\u4e00\u5316&#xff08;Batch Normalization&#xff09;&#xff1a;\u52a0\u901f\u6536\u655b\u7684\u5229\u5668\u3002<\/li>\n<li>6.4 \u8d85\u53c2\u6570\u8c03\u4f18&#xff1a;\u7f51\u683c\u641c\u7d22\u3001\u968f\u673a\u641c\u7d22\u4e0e\u8d1d\u53f6\u65af\u4f18\u5316\u3002<\/li>\n<li>6.5 \u6743\u91cd\u521d\u59cb\u5316\u7b56\u7565&#xff1a;Xavier\u3001He\u521d\u59cb\u5316\u7b49\u3002<\/li>\n<\/ul>\n<h3>\u7b2c\u4e09\u90e8\u5206&#xff1a;\u8fdb\u9636\u7bc7 \u2014\u2014 \u638c\u63e1\u6838\u5fc3\u7f51\u7edc\u67b6\u6784<\/h3>\n<h4>\u7b2c7\u7ae0&#xff1a;\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09; \u2014\u2014 \u6d1e\u6089\u56fe\u50cf\u7684\u5965\u79d8<\/h4>\n<ul>\n<li>7.1 CNN\u7684\u6838\u5fc3\u601d\u60f3&#xff1a;\u5c40\u90e8\u8fde\u63a5\u3001\u6743\u503c\u5171\u4eab\u4e0e\u6c60\u5316\u3002<\/li>\n<li>7.2 \u5377\u79ef\u5c42\u3001\u6c60\u5316\u5c42\u4e0e\u5168\u8fde\u63a5\u5c42\u8be6\u89e3\u3002<\/li>\n<li>7.3 \u7ecf\u5178CNN\u67b6\u6784\u6f14\u8fdb&#xff1a;LeNet-5, AlexNet, VGG, GoogLeNet , ResNet\u3002<\/li>\n<li>7.4 \u8fc1\u79fb\u5b66\u4e60\u4e0e\u5fae\u8c03&#xff08;Fine-tuning&#xff09;&#xff1a;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a\u3002<\/li>\n<li>7.5 CNN\u5e94\u7528&#xff1a;\u56fe\u50cf\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b&#xff08;YOLO, Faster R-CNN&#xff09;\u3001\u56fe\u50cf\u5206\u5272\u3002<\/li>\n<\/ul>\n<h4>\u7b2c8\u7ae0&#xff1a;\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09; \u2014\u2014 \u7406\u89e3\u5e8f\u5217\u7684\u667a\u6167<\/h4>\n<ul>\n<li>8.1 RNN\u7684\u7ed3\u6784\u4e0e\u6311\u6218&#xff1a;\u77ed\u671f\u8bb0\u5fc6\u4e0e\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8\u95ee\u9898\u3002<\/li>\n<li>8.2 \u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;&#xff1a;\u8bb0\u5fc6\u95e8\u7684\u8bbe\u8ba1\u54f2\u5b66\u3002<\/li>\n<li>8.3 \u95e8\u63a7\u5faa\u73af\u5355\u5143&#xff08;GRU&#xff09;&#xff1a;LSTM\u7684\u7b80\u5316\u4e0e\u53d8\u4f53\u3002<\/li>\n<li>8.4 \u53cc\u5411RNN\u4e0e\u6df1\u5ea6RNN\u3002<\/li>\n<li>8.5 RNN\u5e94\u7528&#xff1a;\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;\u6587\u672c\u5206\u7c7b\u3001\u60c5\u611f\u5206\u6790&#xff09;\u3001\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u3002<\/li>\n<\/ul>\n<h4>\u7b2c9\u7ae0&#xff1a;\u6ce8\u610f\u529b\u673a\u5236\u4e0eTransformer \u2014\u2014 \u73b0\u4ee3NLP\u7684\u57fa\u77f3<\/h4>\n<ul>\n<li>9.1 \u6ce8\u610f\u529b&#xff08;Attention&#xff09;\u673a\u5236\u7684\u539f\u7406\u4e0e\u9b45\u529b\u3002<\/li>\n<li>9.2 Transformer\u67b6\u6784\u8be6\u89e3&#xff1a;\u81ea\u6ce8\u610f\u529b\u3001\u591a\u5934\u6ce8\u610f\u529b\u3001\u4f4d\u7f6e\u7f16\u7801\u3002<\/li>\n<li>9.3 BERT\u3001GPT\u53ca\u5176\u4ed6\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b&#xff1a;\u65b0\u8303\u5f0f\u7684\u5d1b\u8d77\u3002<\/li>\n<li>9.4 Transformer\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u7684\u5e94\u7528&#xff08;Vision Transformer, ViT&#xff09;\u3002<\/li>\n<\/ul>\n<h4>\u7b2c10\u7ae0&#xff1a;\u751f\u6210\u5f0f\u6a21\u578b \u2014\u2014 \u521b\u9020\u4e0e\u60f3\u8c61<\/h4>\n<ul>\n<li>10.1 \u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;&#xff1a;\u751f\u6210\u5668\u4e0e\u5224\u522b\u5668\u7684\u535a\u5f08\u3002<\/li>\n<li>10.2 \u53d8\u5206\u81ea\u7f16\u7801\u5668&#xff08;VAE&#xff09;&#xff1a;\u6982\u7387\u751f\u6210\u7684\u7f8e\u5b66\u3002<\/li>\n<li>10.3 \u6269\u6563\u6a21\u578b&#xff08;Diffusion Models&#xff09;&#xff1a;\u4ece\u566a\u58f0\u4e2d\u751f\u6210\u9ad8\u6e05\u56fe\u50cf\u7684\u827a\u672f\u3002<\/li>\n<li>10.4 \u5e94\u7528&#xff1a;\u56fe\u50cf\u751f\u6210\u3001\u98ce\u683c\u8fc1\u79fb\u3001\u6570\u636e\u589e\u5f3a\u3002<\/li>\n<\/ul>\n<h3>\u7b2c\u56db\u90e8\u5206&#xff1a;\u5b9e\u6218\u7bc7 \u2014\u2014 \u4ece\u7406\u8bba\u5230\u4ef7\u503c\u7684\u8f6c\u5316<\/h3>\n<h4>\u7b2c11\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u8ba1\u7b97\u673a\u89c6\u89c9<\/h4>\n<ul>\n<li>11.1 \u56fe\u50cf\u5206\u7c7b&#xff1a;\u6784\u5efa\u4e00\u4e2a\u5783\u573e\u5206\u7c7b\u7cfb\u7edf\u3002<\/li>\n<li>11.2 \u76ee\u6807\u68c0\u6d4b&#xff1a;\u5b9e\u73b0\u4e00\u4e2a\u5b9e\u65f6\u4eba\u8138\u6216\u8f66\u8f86\u68c0\u6d4b\u5668\u3002<\/li>\n<li>11.3 \u56fe\u50cf\u98ce\u683c\u8fc1\u79fb&#xff1a;\u5c06\u7167\u7247\u53d8\u6210\u68b5\u9ad8\u98ce\u683c\u7684\u6cb9\u753b\u3002<\/li>\n<\/ul>\n<h4>\u7b2c12\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u81ea\u7136\u8bed\u8a00\u5904\u7406<\/h4>\n<ul>\n<li>12.1 \u6587\u672c\u60c5\u611f\u5206\u6790&#xff1a;\u5206\u6790\u7535\u5f71\u8bc4\u8bba\u7684\u60c5\u611f\u503e\u5411\u3002<\/li>\n<li>12.2 \u673a\u5668\u7ffb\u8bd1&#xff1a;\u6784\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u4e2d\u82f1\u7ffb\u8bd1\u6a21\u578b\u3002<\/li>\n<li>12.3 \u667a\u80fd\u95ee\u7b54\u673a\u5668\u4eba&#xff1a;\u57fa\u4e8e\u77e5\u8bc6\u5e93\u7684\u95ee\u7b54\u7cfb\u7edf\u3002<\/li>\n<\/ul>\n<h4>\u7b2c13\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u5176\u4ed6\u9886\u57df<\/h4>\n<ul>\n<li>13.1 \u65f6\u95f4\u5e8f\u5217\u9884\u6d4b&#xff1a;\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u6216\u5929\u6c14\u53d8\u5316\u3002<\/li>\n<li>13.2 \u63a8\u8350\u7cfb\u7edf&#xff1a;\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6784\u5efa\u7535\u5f71\u6216\u5546\u54c1\u63a8\u8350\u5f15\u64ce\u3002<\/li>\n<li>13.3 \u5f3a\u5316\u5b66\u4e60\u5165\u95e8&#xff1a;\u4f7f\u7528\u6df1\u5ea6Q\u7f51\u7edc&#xff08;DQN&#xff09;\u73a9\u8f6c\u7b80\u5355\u6e38\u620f\u3002<\/li>\n<\/ul>\n<h4>\u7b2c14\u7ae0&#xff1a;\u6a21\u578b\u90e8\u7f72\u4e0e\u5de5\u7a0b\u5316<\/h4>\n<ul>\n<li>14.1 \u6a21\u578b\u8f7b\u91cf\u5316&#xff1a;\u526a\u679d\u3001\u91cf\u5316\u4e0e\u77e5\u8bc6\u84b8\u998f\u3002<\/li>\n<li>14.2 \u6a21\u578b\u90e8\u7f72&#xff1a;ONNX\u3001TensorFlow Serving\u3001TorchServe\u3002<\/li>\n<li>14.3 \u5c06\u6a21\u578b\u5c01\u88c5\u4e3aAPI\u670d\u52a1&#xff08;\u4f7f\u7528Flask\u6216FastAPI&#xff09;\u3002<\/li>\n<li>14.4 MLOps\u7b80\u4ecb&#xff1a;\u6570\u636e\u3001\u6a21\u578b\u4e0e\u4ee3\u7801\u7684\u7248\u672c\u63a7\u5236\u3002<\/li>\n<\/ul>\n<h3>\u7b2c\u4e94\u90e8\u5206&#xff1a;\u5c55\u671b\u7bc7 \u2014\u2014 \u63a2\u7d22\u672a\u6765\u7684\u8fb9\u754c<\/h3>\n<h4>\u7b2c15\u7ae0&#xff1a;\u524d\u6cbf\u4e13\u9898\u4e0e\u672a\u6765\u8d8b\u52bf<\/h4>\n<ul>\n<li>15.1 \u56fe\u795e\u7ecf\u7f51\u7edc&#xff08;GNN&#xff09;&#xff1a;\u5904\u7406\u975e\u6b27\u51e0\u91cc\u5f97\u6570\u636e\u3002<\/li>\n<li>15.2 \u8054\u90a6\u5b66\u4e60&#xff1a;\u9690\u79c1\u4fdd\u62a4\u4e0b\u7684\u5206\u5e03\u5f0f\u5b66\u4e60\u3002<\/li>\n<li>15.3 \u53ef\u89e3\u91ca\u6027AI&#xff08;XAI&#xff09;&#xff1a;\u6253\u5f00\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u9ed1\u7bb1\u201d\u3002<\/li>\n<li>15.4 \u591a\u6a21\u6001\u5b66\u4e60&#xff1a;\u878d\u5408\u6587\u672c\u3001\u56fe\u50cf\u4e0e\u58f0\u97f3\u3002<\/li>\n<li>15.5 AI\u4f26\u7406\u4e0e\u793e\u4f1a\u8d23\u4efb&#xff1a;\u4f5c\u4e3a\u6280\u672f\u521b\u9020\u8005\u7684\u601d\u8003\u3002<\/li>\n<\/ul>\n<h4>\u9644\u5f55&#xff1a;\u884c\u8005\u7684\u201c\u5b9d\u5e93\u201d\u4e0e\u201c\u8def\u4e66\u201d<\/h4>\n<ul>\n<li>A&#xff1a;\u5e38\u7528\u6570\u636e\u96c6\u4e0e\u8d44\u6e90\u94fe\u63a5\u3002<\/li>\n<li>B&#xff1a;\u6570\u5b66\u7b26\u53f7\u8868\u3002<\/li>\n<li>C&#xff1a;\u5e38\u89c1\u95ee\u9898\u4e0e\u6392\u9519\u6307\u5357\u3002<\/li>\n<li>D&#xff1a;\u8fdb\u4e00\u6b65\u9605\u8bfb\u5efa\u8bae\u3002<\/li>\n<\/ul>\n<hr \/>\n<h2>\u7b2c\u4e00\u90e8\u5206&#xff1a;\u57fa\u7840\u7bc7 \u2014\u2014 \u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3<\/h2>\n<hr \/>\n<h3>\u7b2c\u4e00\u7ae0&#xff1a;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u4e4b\u65c5<\/h3>\n<ul>\n<li>1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb\u3002<\/li>\n<li>1.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174\u3002<\/li>\n<li>1.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u751f\u6001\u3001\u793e\u533a\u4e0e\u54f2\u5b66\u7684\u7edf\u4e00\u3002<\/li>\n<li>1.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66\u3002<\/li>\n<li>1.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90\u3002<\/li>\n<\/ul>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u5f53\u60a8\u7ffb\u5f00\u8fd9\u4e00\u9875&#xff0c;\u60a8\u6b63\u7ad9\u5728\u4e00\u4e2a\u65f6\u4ee3\u7684\u5165\u53e3\u3002\u8fd9\u4e2a\u65f6\u4ee3\u7531\u6570\u636e\u9a71\u52a8&#xff0c;\u7531\u7b97\u6cd5\u8d4b\u80fd&#xff0c;\u5b83\u7684\u6838\u5fc3\u5f15\u64ce&#xff0c;\u4fbf\u662f\u6211\u4eec\u5373\u5c06\u6df1\u5165\u63a2\u7d22\u7684\u201c\u6df1\u5ea6\u5b66\u4e60\u201d\u3002\u5b83\u4e0d\u4ec5\u662f\u5de5\u7a0b\u5e08\u548c\u79d1\u5b66\u5bb6\u7684\u5de5\u5177&#xff0c;\u66f4\u662f\u4e00\u79cd\u5168\u65b0\u7684\u601d\u7ef4\u65b9\u5f0f&#xff0c;\u6b63\u5728\u4ee5\u78c5\u7934\u4e4b\u52bf\u91cd\u5851\u6211\u4eec\u6240\u77e5\u7684\u4e16\u754c\u3002\u672c\u7ae0\u5c06\u4f5c\u4e3a\u60a8\u7684\u5411\u5bfc&#xff0c;\u4e3a\u60a8\u63cf\u7ed8\u4e00\u5e45\u5b8f\u5927\u7684\u5730\u56fe&#xff0c;\u8ba9\u60a8\u660e\u767d\u6211\u4eec\u4ece\u4f55\u800c\u6765&#xff0c;\u8eab\u5728\u4f55\u5904&#xff0c;\u53c8\u5c06\u53bb\u5411\u4f55\u65b9\u3002<\/p>\n<h4>1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb<\/h4>\n<p>\u5728\u8e0f\u4e0a\u6df1\u5ea6\u5b66\u4e60\u7684\u5947\u5999\u65c5\u7a0b\u4e4b\u524d&#xff0c;\u6211\u4eec\u5fc5\u987b\u5148\u6821\u51c6\u7f57\u76d8&#xff0c;\u660e\u786e\u51e0\u4e2a\u6700\u57fa\u672c\u5374\u4e5f\u6700\u5bb9\u6613\u6df7\u6dc6\u7684\u6982\u5ff5&#xff1a;\u4eba\u5de5\u667a\u80fd&#xff08;Artificial Intelligence, AI&#xff09;\u3001\u673a\u5668\u5b66\u4e60&#xff08;Machine Learning, ML&#xff09;\u548c\u6df1\u5ea6\u5b66\u4e60&#xff08;Deep Learning, DL&#xff09;\u3002\u5b83\u4eec\u5e76\u975e\u540c\u4e49\u8bcd&#xff0c;\u800c\u662f\u4e00\u4e2a\u5c42\u5c42\u9012\u8fdb\u3001\u76f8\u4e92\u5305\u542b\u7684\u6709\u673a\u6574\u4f53\u3002\u7406\u89e3\u5b83\u4eec\u4e4b\u95f4\u7684\u7cbe\u5999\u5173\u7cfb&#xff0c;\u662f\u60a8\u6784\u5efa\u6e05\u6670\u8ba4\u77e5\u6846\u67b6\u7684\u7b2c\u4e00\u6b65\u3002<\/p>\n<h5>1.1.1 \u4eba\u5de5\u667a\u80fd&#xff08;AI&#xff09;&#xff1a;\u6700\u521d\u7684\u68a6\u60f3\u4e0e\u5b8f\u5927\u613f\u666f<\/h5>\n<p>\u4eba\u5de5\u667a\u80fd&#xff0c;\u662f\u8fd9\u573a\u4f1f\u5927\u63a2\u7d22\u7684\u7ec8\u6781\u68a6\u60f3\u3002\u5b83\u7684\u6838\u5fc3\u76ee\u6807&#xff0c;\u662f\u521b\u9020\u51fa\u80fd\u591f\u50cf\u4eba\u7c7b\u4e00\u6837\u601d\u8003\u3001\u5b66\u4e60\u3001\u611f\u77e5\u548c\u884c\u52a8\u7684\u667a\u80fd\u673a\u5668\u3002\u8fd9\u4e2a\u68a6\u60f3&#xff0c;\u5982\u540c\u4e00\u9897\u53e4\u8001\u7684\u661f\u8fb0&#xff0c;\u5728\u4eba\u7c7b\u6587\u660e\u7684\u591c\u7a7a\u4e2d\u95ea\u8000\u4e86\u6570\u5343\u5e74&#xff0c;\u4ece\u53e4\u5e0c\u814a\u795e\u8bdd\u4e2d\u7684\u81ea\u52a8\u673a\u68b0&#xff0c;\u5230\u4e2d\u4e16\u7eaa\u70bc\u91d1\u672f\u58eb\u7684\u201c\u4eba\u9020\u4eba\u201d\u4f20\u8bf4&#xff0c;\u65e0\u4e0d\u5bc4\u6258\u7740\u4eba\u7c7b\u5bf9\u521b\u9020\u667a\u6167\u751f\u547d\u7684\u65e0\u9650\u9050\u60f3\u3002<\/p>\n<p>\u4ec0\u4e48\u662f\u4eba\u5de5\u667a\u80fd&#xff1f;<\/p>\n<p>\u5728\u73b0\u4ee3\u79d1\u5b66\u8bed\u5883\u4e0b&#xff0c;\u4eba\u5de5\u667a\u80fd\u7684\u6b63\u5f0f\u8bde\u751f&#xff0c;\u901a\u5e38\u88ab\u8ffd\u6eaf\u52301956\u5e74\u7684\u8fbe\u7279\u8305\u65af\u4f1a\u8bae\u3002\u4e00\u7fa4\u5bcc\u6709\u8fdc\u89c1\u7684\u79d1\u5b66\u5bb6\u9f50\u805a\u4e00\u5802&#xff0c;\u9996\u6b21\u63d0\u51fa\u4e86\u201c\u4eba\u5de5\u667a\u80fd\u201d\u8fd9\u4e00\u672f\u8bed&#xff0c;\u5e76\u5bf9\u5176\u8fdb\u884c\u4e86\u521d\u6b65\u5b9a\u4e49&#xff1a;\u8ba9\u673a\u5668\u80fd\u591f\u6267\u884c\u901a\u5e38\u9700\u8981\u4eba\u7c7b\u667a\u80fd\u624d\u80fd\u5b8c\u6210\u7684\u4efb\u52a1\u3002\u5176\u4e2d&#xff0c;\u82f1\u56fd\u6570\u5b66\u5bb6\u3001\u903b\u8f91\u5b66\u5bb6\u827e\u4f26\u00b7\u56fe\u7075\u57281950\u5e74\u63d0\u51fa\u7684\u8457\u540d\u201c\u56fe\u7075\u6d4b\u8bd5\u201d&#xff0c;\u81f3\u4eca\u4ecd\u662f\u8861\u91cf\u673a\u5668\u662f\u5426\u5177\u5907\u667a\u80fd\u7684\u7ecf\u5178\u6807\u51c6&#xff1a;\u5982\u679c\u4e00\u53f0\u673a\u5668\u80fd\u591f\u4e0e\u4eba\u7c7b\u8fdb\u884c\u5bf9\u8bdd&#xff0c;\u800c\u4eba\u7c7b\u65e0\u6cd5\u5206\u8fa8\u51fa\u5bf9\u65b9\u662f\u673a\u5668\u8fd8\u662f\u4eba&#xff0c;\u90a3\u4e48\u6211\u4eec\u5c31\u53ef\u4ee5\u8ba4\u4e3a\u8fd9\u53f0\u673a\u5668\u5177\u5907\u4e86\u667a\u80fd\u3002<\/p>\n<p>\u4eba\u5de5\u667a\u80fd\u7684\u5b8f\u5927\u613f\u666f&#xff0c;\u53c8\u53ef\u7ec6\u5206\u4e3a\u4e24\u4e2a\u5c42\u6b21&#xff1a;<\/p>\n<ul>\n<li>\u5f3a\u4eba\u5de5\u667a\u80fd&#xff08;Strong AI&#xff09;&#xff1a; \u6307\u7684\u662f\u80fd\u591f\u771f\u6b63\u62e5\u6709\u81ea\u6211\u610f\u8bc6\u3001\u5177\u5907\u4e0e\u4eba\u7c7b\u540c\u7b49\u751a\u81f3\u8d85\u8d8a\u4eba\u7c7b\u667a\u6167\u7684\u901a\u7528\u667a\u80fd\u4f53\u3002\u5b83\u80fd\u591f\u50cf\u4eba\u7c7b\u4e00\u6837\u8fdb\u884c\u62bd\u8c61\u601d\u7ef4\u3001\u89e3\u51b3\u901a\u7528\u95ee\u9898\u3001\u751a\u81f3\u62e5\u6709\u60c5\u611f\u548c\u521b\u9020\u529b\u3002\u8fd9\u662f\u79d1\u5e7b\u7535\u5f71\u4e2d\u5e38\u89c1\u7684\u9898\u6750&#xff0c;\u4e5f\u662f\u6211\u4eec\u8ffd\u6c42\u7684\u661f\u8fb0\u5927\u6d77&#xff0c;\u4f46\u76ee\u524d\u4ecd\u5904\u4e8e\u7406\u8bba\u63a2\u7d22\u548c\u9065\u8fdc\u672a\u6765\u7684\u9636\u6bb5\u3002<\/li>\n<li>\u5f31\u4eba\u5de5\u667a\u80fd&#xff08;Weak AI&#xff09;\u6216\u79f0\u5e94\u7528\u578b\u4eba\u5de5\u667a\u80fd&#xff1a; \u6307\u7684\u662f\u4e13\u6ce8\u4e8e\u5728\u7279\u5b9a\u4efb\u52a1\u4e0a\u6a21\u62df\u751a\u81f3\u8d85\u8d8a\u4eba\u7c7b\u8868\u73b0\u7684\u667a\u80fd\u7cfb\u7edf\u3002\u4f8b\u5982&#xff0c;\u60a8\u7684\u624b\u673a\u8bed\u97f3\u52a9\u624b\u3001\u4e3a\u60a8\u7cbe\u51c6\u63a8\u8350\u7535\u5f71\u7684\u7b97\u6cd5\u3001\u4ee5\u53ca\u5728\u56f4\u68cb\u76d8\u4e0a\u6218\u80dc\u4e16\u754c\u51a0\u519b\u7684AlphaGo\u3002\u672c\u4e66\u6240\u8ba8\u8bba\u7684\u4e00\u5207&#xff0c;\u90fd\u5c5e\u4e8e\u5f31\u4eba\u5de5\u667a\u80fd\u7684\u8303\u7574&#xff0c;\u5b83\u4eec\u5728\u7279\u5b9a\u9886\u57df\u5c55\u73b0\u51fa\u4ee4\u4eba\u60ca\u53f9\u7684\u667a\u80fd&#xff0c;\u4f46\u5e76\u4e0d\u5177\u5907\u901a\u7528\u667a\u80fd\u548c\u81ea\u6211\u610f\u8bc6\u3002<\/li>\n<\/ul>\n<p>AI\u7684\u53d1\u5c55\u7b80\u53f2&#xff1a;<\/p>\n<p>AI\u7684\u53d1\u5c55\u5e76\u975e\u4e00\u5e06\u98ce\u987a&#xff0c;\u5b83\u5982\u540c\u4e00\u6761\u873f\u8712\u66f2\u6298\u7684\u6cb3\u6d41&#xff0c;\u7ecf\u5386\u4e86\u6570\u4e2a\u201c\u9ec4\u91d1\u65f6\u4ee3\u201d\u4e0e\u201cAI\u5bd2\u51ac\u201d\u7684\u4ea4\u66ff\u3002\u65e9\u671f&#xff0c;\u7814\u7a76\u8005\u4eec\u5bf9\u57fa\u4e8e\u903b\u8f91\u63a8\u7406\u548c\u7b26\u53f7\u64cd\u4f5c\u7684\u201c\u4e13\u5bb6\u7cfb\u7edf\u201d\u5bc4\u4e88\u539a\u671b&#xff0c;\u53d6\u5f97\u4e86\u4e00\u4e9b\u521d\u6b65\u6210\u529f\u3002\u7136\u800c&#xff0c;\u5f53\u8fd9\u4e9b\u7cfb\u7edf\u9762\u5bf9\u73b0\u5b9e\u4e16\u754c\u4e2d\u6d77\u91cf\u3001\u6a21\u7cca\u3001\u4e0d\u786e\u5b9a\u7684\u6570\u636e\u65f6&#xff0c;\u5176\u5c40\u9650\u6027\u4fbf\u66b4\u9732\u65e0\u9057\u3002\u5f53\u671f\u671b\u65e0\u6cd5\u5151\u73b0&#xff0c;\u8d44\u91d1\u548c\u70ed\u60c5\u4fbf\u4f1a\u9000\u6f6e&#xff0c;AI\u7814\u7a76\u4fbf\u8fdb\u5165\u201c\u5bd2\u51ac\u201d\u3002\u7136\u800c&#xff0c;\u6b63\u662f\u8fd9\u4e9b\u8d77\u4f0f&#xff0c;\u4fc3\u4f7f\u7814\u7a76\u8005\u4eec\u4e0d\u65ad\u53cd\u601d&#xff0c;\u6700\u7ec8\u627e\u5230\u4e86\u4e00\u6761\u66f4\u5177\u6f5c\u529b\u7684\u9053\u8def\u2014\u2014\u8ba9\u673a\u5668\u81ea\u5df1\u53bb\u5b66\u4e60&#xff0c;\u4ece\u6570\u636e\u4e2d\u53d1\u73b0\u89c4\u5f8b&#xff0c;\u800c\u975e\u4f9d\u8d56\u4eba\u7c7b\u9884\u8bbe\u7684\u50f5\u786c\u89c4\u5219\u3002\u8fd9\u4e3a\u673a\u5668\u5b66\u4e60\u7684\u5174\u8d77\u57cb\u4e0b\u4e86\u4f0f\u7b14\u3002<\/p>\n<h5>1.1.2 \u673a\u5668\u5b66\u4e60&#xff08;ML&#xff09;&#xff1a;\u5b9e\u73b0\u4eba\u5de5\u667a\u80fd\u7684\u6838\u5fc3\u9014\u5f84<\/h5>\n<p>\u5982\u679c\u8bf4\u4eba\u5de5\u667a\u80fd\u662f\u201c\u9020\u51fa\u806a\u660e\u7684\u673a\u5668\u201d\u8fd9\u4e2a\u5b8f\u4f1f\u76ee\u6807&#xff0c;\u90a3\u4e48\u673a\u5668\u5b66\u4e60\u5c31\u662f\u5b9e\u73b0\u8fd9\u4e2a\u76ee\u6807\u6700\u4e3b\u6d41\u3001\u6700\u6709\u6548\u7684\u4e00\u6761\u8def\u5f84\u3002\u5b83\u7684\u6838\u5fc3\u601d\u60f3&#xff0c;\u6b63\u5982\u5176\u540d&#xff0c;\u662f\u8ba9\u673a\u5668\u62e5\u6709\u5b66\u4e60\u7684\u80fd\u529b&#xff0c;\u4ece\u7ecf\u9a8c\u4e2d\u81ea\u52a8\u6539\u8fdb\u5176\u6027\u80fd\u3002<\/p>\n<p>\u4ec0\u4e48\u662f\u673a\u5668\u5b66\u4e60&#xff1f;<\/p>\n<p>\u8ba1\u7b97\u673a\u79d1\u5b66\u5bb6\u4e9a\u745f\u00b7\u8428\u7f2a\u5c14&#xff08;Arthur Samuel&#xff09;\u57281959\u5e74\u7ed9\u51fa\u4e86\u4e00\u4e2a\u7ecf\u5178\u4e14\u7cbe\u8f9f\u7684\u5b9a\u4e49&#xff1a;\u201c\u673a\u5668\u5b66\u4e60\u662f\u5728\u6ca1\u6709\u88ab\u660e\u786e\u7f16\u7a0b\u7684\u60c5\u51b5\u4e0b&#xff0c;\u7ed9\u4e88\u8ba1\u7b97\u673a\u5b66\u4e60\u80fd\u529b\u7684\u7814\u7a76\u9886\u57df\u3002\u201d<\/p>\n<p>\u8fd9\u610f\u5473\u7740&#xff0c;\u6211\u4eec\u4e0d\u518d\u50cf\u4f20\u7edf\u7f16\u7a0b\u90a3\u6837&#xff0c;\u4e3a\u673a\u5668\u7f16\u5199\u8be6\u5c3d\u7684\u6307\u4ee4&#xff08;\u4f8b\u5982&#xff1a;\u201c\u5982\u679c\u8f93\u5165\u662f\u732b\u7684\u56fe\u7247&#xff0c;\u90a3\u4e48\u5224\u65ad\u4e3a\u732b\u201d&#xff09;\u3002\u53d6\u800c\u4ee3\u4e4b\u7684\u662f&#xff0c;\u6211\u4eec\u7ed9\u673a\u5668\u63d0\u4f9b\u5927\u91cf\u7684\u6570\u636e&#xff08;\u4f8b\u5982&#xff0c;\u6210\u5343\u4e0a\u4e07\u5f20\u732b\u7684\u56fe\u7247&#xff09;&#xff0c;\u5e76\u8bbe\u8ba1\u4e00\u4e2a\u7b97\u6cd5&#xff0c;\u8ba9\u5b83\u4ece\u8fd9\u4e9b\u6570\u636e\u4e2d\u81ea\u52a8\u53d1\u73b0\u89c4\u5f8b\u3001\u6a21\u5f0f&#xff0c;\u751a\u81f3\u751f\u6210\u77e5\u8bc6\u3002\u8fd9\u5c31\u50cf\u6559\u4e00\u4e2a\u5b69\u5b50\u8ba4\u8bc6\u732b&#xff1a;\u60a8\u4e0d\u662f\u544a\u8bc9\u4ed6\u201c\u732b\u6709\u4e24\u53ea\u8033\u6735\u3001\u56db\u6761\u817f\u3001\u4e00\u6761\u5c3e\u5df4\u2026\u2026\u201d&#xff0c;\u800c\u662f\u4e0d\u65ad\u5730\u6307\u7740\u5404\u79cd\u5404\u6837\u7684\u732b\u544a\u8bc9\u4ed6\u201c\u8fd9\u662f\u732b\u201d&#xff0c;\u4e45\u800c\u4e45\u4e4b&#xff0c;\u4ed6\u81ea\u7136\u5c31\u5b66\u4f1a\u4e86\u8bc6\u522b\u732b&#xff0c;\u5373\u4f7f\u9047\u5230\u524d\u6240\u672a\u89c1\u7684\u54c1\u79cd&#xff0c;\u4e5f\u80fd\u4e3e\u4e00\u53cd\u4e09\u3002<\/p>\n<p>\u673a\u5668\u5b66\u4e60\u7684\u8303\u7574&#xff1a;<\/p>\n<p>\u6839\u636e\u5b66\u4e60\u65b9\u5f0f\u548c\u6570\u636e\u7279\u6027\u7684\u4e0d\u540c&#xff0c;\u673a\u5668\u5b66\u4e60\u4e3b\u8981\u5206\u4e3a\u4e09\u5927\u8303\u5f0f&#xff0c;\u5b83\u4eec\u6784\u6210\u4e86\u673a\u5668\u5b66\u4e60\u9886\u57df\u7684\u6838\u5fc3\u9aa8\u67b6&#xff1a;<\/p>\n<ul>\n<li>\u76d1\u7763\u5b66\u4e60&#xff08;Supervised Learning&#xff09;&#xff1a; \u8fd9\u662f\u6700\u5e38\u89c1\u7684\u4e00\u79cd\u5b66\u4e60\u8303\u5f0f\u3002\u6211\u4eec\u63d0\u4f9b\u7ed9\u673a\u5668\u7684\u6570\u636e\u662f\u201c\u5e26\u6807\u7b7e\u201d\u7684&#xff0c;\u5c31\u50cf\u6709\u6807\u51c6\u7b54\u6848\u7684\u7ec3\u4e60\u518c\u3002\u6a21\u578b\u901a\u8fc7\u5b66\u4e60\u8f93\u5165\u6570\u636e&#xff08;\u7279\u5f81&#xff09;\u4e0e\u5bf9\u5e94\u8f93\u51fa&#xff08;\u6807\u7b7e&#xff09;\u4e4b\u95f4\u7684\u6620\u5c04\u5173\u7cfb\u6765\u505a\u51fa\u9884\u6d4b\u3002\u4f8b\u5982&#xff0c;\u7ed9\u6a21\u578b\u6210\u5343\u4e0a\u4e07\u5c01\u90ae\u4ef6&#xff0c;\u5e76\u660e\u786e\u6807\u660e\u54ea\u4e9b\u662f\u201c\u5783\u573e\u90ae\u4ef6\u201d&#xff0c;\u54ea\u4e9b\u662f\u201c\u6b63\u5e38\u90ae\u4ef6\u201d&#xff0c;\u8ba9\u5b83\u5b66\u4f1a\u5206\u7c7b&#xff1b;\u6216\u8005\u6839\u636e\u623f\u5c4b\u9762\u79ef\u3001\u5367\u5ba4\u6570\u91cf\u7b49\u7279\u5f81&#xff0c;\u9884\u6d4b\u623f\u5c4b\u4ef7\u683c\u3002<\/li>\n<li>\u65e0\u76d1\u7763\u5b66\u4e60&#xff08;Unsupervised Learning&#xff09;&#xff1a; \u4e0e\u76d1\u7763\u5b66\u4e60\u4e0d\u540c&#xff0c;\u65e0\u76d1\u7763\u5b66\u4e60\u63d0\u4f9b\u7ed9\u673a\u5668\u7684\u6570\u636e\u662f\u6ca1\u6709\u6807\u7b7e\u7684&#xff0c;\u5c31\u50cf\u4e00\u672c\u6ca1\u6709\u7b54\u6848\u7684\u4e60\u9898\u96c6\u3002\u673a\u5668\u9700\u8981\u81ea\u5df1\u53bb\u53d1\u73b0\u6570\u636e\u4e2d\u9690\u85cf\u7684\u7ed3\u6784\u3001\u6a21\u5f0f\u6216\u5173\u8054\u6027\u3002\u4f8b\u5982&#xff0c;\u5c06\u6d77\u91cf\u7528\u6237\u6570\u636e\u8fdb\u884c\u805a\u7c7b&#xff0c;\u627e\u51fa\u5177\u6709\u76f8\u4f3c\u5174\u8da3\u7684\u7fa4\u4f53&#xff1b;\u6216\u8005\u901a\u8fc7\u964d\u7ef4\u6280\u672f&#xff0c;\u53d1\u73b0\u6570\u636e\u4e2d\u6700\u91cd\u8981\u7684\u6f5c\u5728\u7279\u5f81\u3002<\/li>\n<li>\u5f3a\u5316\u5b66\u4e60&#xff08;Reinforcement Learning&#xff09;&#xff1a; \u8fd9\u662f\u4e00\u79cd\u901a\u8fc7\u201c\u8bd5\u9519\u201d\u6765\u5b66\u4e60\u7684\u8303\u5f0f\u3002\u6211\u4eec\u4e0d\u7ed9\u673a\u5668\u660e\u786e\u7684\u8f93\u5165-\u8f93\u51fa\u5bf9&#xff0c;\u800c\u662f\u7ed9\u5b83\u4e00\u4e2a\u76ee\u6807\u548c\u4e00\u4e2a\u5956\u60e9\u7cfb\u7edf\u3002\u673a\u5668&#xff08;\u4ee3\u7406&#xff09;\u901a\u8fc7\u4e0d\u65ad\u5730\u4e0e\u73af\u5883\u8fdb\u884c\u4ea4\u4e92&#xff0c;\u6839\u636e\u73af\u5883\u7684\u53cd\u9988&#xff08;\u5956\u52b1\u6216\u60e9\u7f5a&#xff09;\u6765\u5b66\u4e60\u5982\u4f55\u91c7\u53d6\u884c\u52a8\u4ee5\u83b7\u5f97\u6700\u5927\u7684\u7d2f\u79ef\u5956\u52b1\u3002AlphaGo\u4e0b\u56f4\u68cb\u3001\u673a\u5668\u4eba\u5b66\u4e60\u884c\u8d70&#xff0c;\u90fd\u662f\u5f3a\u5316\u5b66\u4e60\u6700\u5178\u578b\u7684\u5e94\u7528\u3002\u5b83\u8ba9\u673a\u5668\u5b66\u4f1a\u4e86\u5982\u4f55\u5728\u590d\u6742\u52a8\u6001\u7684\u73af\u5883\u4e2d\u505a\u51fa\u6700\u4f18\u51b3\u7b56\u3002<\/li>\n<\/ul>\n<h5>1.1.3 \u6df1\u5ea6\u5b66\u4e60&#xff08;DL&#xff09;&#xff1a;\u673a\u5668\u5b66\u4e60\u7684\u5f3a\u5927\u5206\u652f<\/h5>\n<p>\u73b0\u5728&#xff0c;\u6211\u4eec\u6765\u5230\u4e86\u8fd9\u573a\u63a2\u7d22\u7684\u6838\u5fc3\u2014\u2014\u6df1\u5ea6\u5b66\u4e60\u3002\u5982\u679c\u8bf4\u673a\u5668\u5b66\u4e60\u662f\u8ba9\u4eba\u5de5\u667a\u80fd\u5f97\u4ee5\u5b9e\u73b0\u7684\u5eb7\u5e84\u5927\u9053&#xff0c;\u90a3\u4e48\u6df1\u5ea6\u5b66\u4e60\u5c31\u662f\u8fd9\u6761\u5927\u9053\u4e0a\u76ee\u524d\u6700\u5feb\u3001\u6700\u5f3a\u5927\u7684\u8d85\u7ea7\u8dd1\u8f66\u3002\u5b83\u4ee5\u5176\u5353\u8d8a\u7684\u6027\u80fd\u548c\u5e7f\u6cdb\u7684\u5e94\u7528&#xff0c;\u5c06\u4eba\u5de5\u667a\u80fd\u63a8\u5411\u4e86\u524d\u6240\u672a\u6709\u7684\u9ad8\u5ea6\u3002<\/p>\n<p>\u4ec0\u4e48\u662f\u6df1\u5ea6\u5b66\u4e60&#xff1f;<\/p>\n<p>\u4ece\u6280\u672f\u4e0a\u8bb2&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u672c\u8d28\u662f\u4f7f\u7528\u201c\u6df1\u5c42\u201d\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc&#xff08;Deep Neural Networks, DNN&#xff09;\u7684\u673a\u5668\u5b66\u4e60\u6280\u672f\u3002\u8fd9\u91cc\u7684\u201c\u6df1\u201d&#xff08;Deep&#xff09;&#xff0c;\u662f\u5176\u6700\u5173\u952e\u7684\u7279\u5f81&#xff0c;\u6307\u7684\u662f\u795e\u7ecf\u7f51\u7edc\u7684\u5c42\u6570\u975e\u5e38\u591a&#xff0c;\u6709\u65f6\u53ef\u8fbe\u6570\u5341\u751a\u81f3\u4e0a\u5343\u5c42\u3002\u8fd9\u4e9b\u5c42\u7ea7\u5e76\u975e\u7b80\u5355\u5806\u53e0&#xff0c;\u800c\u662f\u901a\u8fc7\u590d\u6742\u7684\u8fde\u63a5\u548c\u975e\u7ebf\u6027\u53d8\u6362&#xff0c;\u5171\u540c\u534f\u4f5c&#xff0c;\u4ece\u539f\u59cb\u6570\u636e\u4e2d\u9010\u5c42\u63d0\u53d6\u548c\u62bd\u8c61\u51fa\u8d8a\u6765\u8d8a\u9ad8\u7ea7\u7684\u7279\u5f81\u3002<\/p>\n<p>\u8fd9\u4e2a\u601d\u60f3\u7684\u7075\u611f&#xff0c;\u7cbe\u5999\u5730\u6e90\u81ea\u4e8e\u6211\u4eec\u4eba\u7c7b\u81ea\u5df1\u7684\u5927\u8111\u89c6\u89c9\u76ae\u5c42\u3002\u5f53\u6211\u4eec\u770b\u5230\u4e00\u53ea\u732b\u65f6&#xff0c;\u60a8\u7684\u5927\u8111\u5e76\u975e\u201c\u4e00\u6b65\u5230\u4f4d\u201d\u5c31\u5b8c\u6210\u4e86\u8bc6\u522b\u3002\u771f\u5b9e\u7684\u8fc7\u7a0b\u662f\u4e00\u4e2a\u9ad8\u5ea6\u5c42\u6b21\u5316\u7684\u4fe1\u606f\u5904\u7406\u6d41\u7a0b&#xff1a;<\/p>\n<li>\u5e95\u5c42\u795e\u7ecf\u5143\u53ef\u80fd\u53ea\u5bf9\u89c6\u91ce\u4e2d\u975e\u5e38\u7b80\u5355\u7684\u6a21\u5f0f\u505a\u51fa\u53cd\u5e94&#xff0c;\u6bd4\u5982\u5149\u70b9\u3001\u8fb9\u7f18\u3001\u7279\u5b9a\u7684\u989c\u8272\u6216\u89d2\u5ea6\u3002<\/li>\n<li>\u8fd9\u4e9b\u5e95\u5c42\u4fe1\u606f\u88ab\u6c47\u96c6\u5230\u4e2d\u5c42\u795e\u7ecf\u5143&#xff0c;\u5b83\u4eec\u5c06\u7b80\u5355\u7684\u6a21\u5f0f\u7ec4\u5408\u6210\u7a0d\u5fae\u590d\u6742\u7684\u5c40\u90e8\u7279\u5f81&#xff0c;\u6bd4\u5982\u773c\u775b\u7684\u8f6e\u5ed3\u3001\u80e1\u987b\u7684\u7ebf\u6761\u3001\u8033\u6735\u7684\u4e09\u89d2\u5f62\u72b6\u3002<\/li>\n<li>\u4fe1\u606f\u7ee7\u7eed\u5411\u4e0a\u4f20\u9012&#xff0c;\u66f4\u9ad8\u5c42\u7684\u795e\u7ecf\u5143\u518d\u5c06\u8fd9\u4e9b\u5c40\u90e8\u7279\u5f81\u8fdb\u884c\u7ec4\u5408&#xff0c;\u6700\u7ec8\u5f62\u6210\u201c\u732b\u201d\u8fd9\u4e2a\u5b8c\u6574\u7684\u3001\u62bd\u8c61\u7684\u6982\u5ff5\u3002<\/li>\n<p>\u6df1\u5ea6\u5b66\u4e60\u6b63\u662f\u6a21\u62df\u4e86\u8fd9\u79cd\u5c42\u6b21\u5316\u7684\u7279\u5f81\u5b66\u4e60&#xff08;Hierarchical Feature Learning&#xff09;\u673a\u5236\u3002\u901a\u8fc7\u6784\u5efa\u4e00\u4e2a\u591a\u5c42\u7684\u7f51\u7edc\u7ed3\u6784&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u80fd\u591f\u81ea\u52a8\u5730\u4ece\u6700\u539f\u59cb\u7684\u8f93\u5165\u6570\u636e\u4e2d&#xff08;\u4f8b\u5982&#xff0c;\u56fe\u50cf\u7684\u50cf\u7d20\u70b9&#xff09;&#xff0c;\u9010\u5c42\u5b66\u4e60\u5230\u4ece\u7b80\u5355\u5230\u590d\u6742\u7684\u7279\u5f81&#xff0c;\u6700\u7ec8\u5b8c\u6210\u8bc6\u522b\u3001\u5206\u7c7b\u6216\u751f\u6210\u7b49\u9ad8\u7ea7\u4efb\u52a1\u3002\u8fd9\u79cd\u7aef\u5230\u7aef&#xff08;End-to-End&#xff09;\u7684\u81ea\u52a8\u5b66\u4e60\u7279\u5f81\u7684\u80fd\u529b&#xff0c;\u662f\u5b83\u76f8\u8f83\u4e8e\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u6700\u5f3a\u5927\u3001\u6700\u5177\u98a0\u8986\u6027\u7684\u4f18\u52bf\u4e4b\u4e00&#xff0c;\u6781\u5927\u5730\u7b80\u5316\u4e86\u7279\u5f81\u5de5\u7a0b\u7684\u590d\u6742\u6027\u3002<\/p>\n<p>\u4e09\u8005\u5173\u7cfb\u7684\u53ef\u89c6\u5316&#xff1a;<\/p>\n<p>\u4e3a\u4e86\u8ba9\u60a8\u5bf9\u8fd9\u4e09\u8005\u7684\u5173\u7cfb\u6709\u4e00\u4e2a\u76f4\u89c2\u4e14\u7262\u56fa\u7684\u5370\u8c61&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528\u4e00\u4e2a\u5173\u7cfb\u56fe\u6765\u63cf\u7ed8\u5b83\u4eec&#xff1a;<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"235\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/01\/20260124113235-6974add34cef2.png\" width=\"80\" \/><\/p>\n<ul>\n<li>\u4eba\u5de5\u667a\u80fd&#xff08;AI&#xff09; \u662f\u6700\u4e0a\u5c42\u7684\u751f\u6001&#xff0c;\u4ee3\u8868\u7740\u6211\u4eec\u521b\u9020\u667a\u80fd\u673a\u5668\u7684\u5b8f\u5927\u613f\u666f\u548c\u6574\u4e2a\u7814\u7a76\u9886\u57df\u3002<\/li>\n<li>\u673a\u5668\u5b66\u4e60&#xff08;ML&#xff09; \u662f\u4e2d\u95f4\u5c42\u7684\u67b6\u6784&#xff0c;\u5b83\u662f\u5b9e\u73b0\u4eba\u5de5\u667a\u80fd\u7684\u4e00\u79cd\u6838\u5fc3\u65b9\u6cd5\u8bba&#xff0c;\u5176\u6838\u5fc3\u662f\u8ba9\u673a\u5668\u4ece\u6570\u636e\u4e2d\u5b66\u4e60\u3002<\/li>\n<li>\u6df1\u5ea6\u5b66\u4e60&#xff08;DL&#xff09; \u662f\u6700\u5185\u5c42\u7684\u6838\u5fc3&#xff0c;\u5b83\u662f\u673a\u5668\u5b66\u4e60\u6280\u672f\u4e2d\u7684\u4e00\u4e2a\u6781\u5176\u5f3a\u5927\u4e14\u6210\u679c\u6590\u7136\u7684\u5206\u652f&#xff0c;\u5b83\u4ee5\u6df1\u5c42\u795e\u7ecf\u7f51\u7edc\u4e3a\u4e3b\u8981\u5de5\u5177&#xff0c;\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u5c42\u6b21\u5316\u7279\u5f81\u5b66\u4e60\u3002<\/li>\n<\/ul>\n<p>\u73b0\u5728&#xff0c;\u60a8\u53ef\u4ee5\u81ea\u4fe1\u5730\u5411\u4ed6\u4eba\u89e3\u91ca&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u662f\u5b9e\u73b0\u673a\u5668\u5b66\u4e60\u7684\u4e00\u79cd\u6d41\u884c\u4e14\u5f3a\u5927\u7684\u6280\u672f&#xff0c;\u800c\u673a\u5668\u5b66\u4e60\u662f\u901a\u5f80\u4eba\u5de5\u667a\u80fd\u7684\u4e00\u6761\u6838\u5fc3\u8def\u5f84\u3002 \u6211\u4eec\u5b66\u4e60\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u6b63\u662f\u4e3a\u4e86\u638c\u63e1\u5f53\u524d\u5b9e\u73b0\u4eba\u5de5\u667a\u80fd\u6700\u5f3a\u5927\u7684\u90a3\u628a\u94a5\u5319&#xff0c;\u5f00\u542f\u901a\u5f80\u672a\u6765\u667a\u80fd\u4e16\u754c\u7684\u5927\u95e8\u3002<\/p>\n<h4>1.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174<\/h4>\n<p>\u4efb\u4f55\u4e00\u95e8\u4f1f\u5927\u7684\u6280\u672f&#xff0c;\u90fd\u4e0d\u662f\u51ed\u7a7a\u51fa\u73b0\u7684\u3002\u6df1\u5ea6\u5b66\u4e60\u7684\u53d1\u5c55\u53f2&#xff0c;\u5982\u540c\u4e00\u90e8\u6ce2\u6f9c\u58ee\u9614\u7684\u53f2\u8bd7&#xff0c;\u5145\u6ee1\u4e86\u5929\u624d\u7684\u6d1e\u89c1\u3001\u957f\u4e45\u7684\u6c89\u5bc2\u3001\u4e0d\u61c8\u7684\u575a\u5b88\u4e0e\u6700\u7ec8\u7684\u8f89\u714c\u3002\u4e86\u89e3\u8fd9\u6bb5\u5386\u53f2&#xff0c;\u80fd\u8ba9\u6211\u4eec\u66f4\u6df1\u523b\u5730\u7406\u89e3\u5176\u6838\u5fc3\u601d\u60f3\u7684\u6f14\u8fdb&#xff0c;\u4ece\u800c\u66f4\u597d\u5730\u628a\u63e1\u5176\u672a\u6765\u8d70\u5411\u3002<\/p>\n<h5>1.2.1 \u601d\u60f3\u7684\u840c\u82bd&#xff1a;\u7406\u8bba\u5960\u57fa\u65f6\u4ee3&#xff08;1940s-1960s&#xff09;<\/h5>\n<p>\u6df1\u5ea6\u5b66\u4e60\u7684\u6839\u6e90&#xff0c;\u53ef\u4ee5\u8ffd\u6eaf\u5230\u4e0a\u4e16\u7eaa\u4e2d\u53f6\u5bf9\u4eba\u8111\u795e\u7ecf\u5143\u5de5\u4f5c\u673a\u5236\u7684\u521d\u6b65\u63a2\u7d22\u3002<\/p>\n<p>\u8d6b\u5e03\u7406\u8bba\u4e0e\u611f\u77e5\u673a&#xff1a; 1949\u5e74&#xff0c;\u52a0\u62ff\u5927\u5fc3\u7406\u5b66\u5bb6\u5510\u7eb3\u5fb7\u00b7\u8d6b\u5e03&#xff08;Donald Hebb&#xff09;\u63d0\u51fa\u4e86\u8457\u540d\u7684\u8d6b\u5e03\u7406\u8bba&#xff08;Hebb&#039;s Rule&#xff09;&#xff0c;\u5176\u6838\u5fc3\u601d\u60f3\u662f\u201c\u4e00\u8d77\u6fc0\u53d1\u7684\u795e\u7ecf\u5143\u4f1a\u8fde\u63a5\u5728\u4e00\u8d77\u201d&#xff08;Cells that fire together, wire together&#xff09;\u3002\u8fd9\u4e3a\u795e\u7ecf\u7f51\u7edc\u7684\u5b66\u4e60\u673a\u5236\u63d0\u4f9b\u4e86\u751f\u7269\u5b66\u4e0a\u7684\u542f\u53d1\u548c\u7406\u8bba\u57fa\u7840\u3002\u57fa\u4e8e\u6b64&#xff0c;1958\u5e74&#xff0c;\u7f8e\u56fd\u5fc3\u7406\u5b66\u5bb6\u5f17\u5170\u514b\u00b7\u7f57\u68ee\u5e03\u62c9\u7279&#xff08;Frank Rosenblatt&#xff09;\u53d1\u660e\u4e86\u611f\u77e5\u673a&#xff08;Perceptron&#xff09;&#xff0c;\u8fd9\u662f\u7b2c\u4e00\u4e2a\u80fd\u591f\u901a\u8fc7\u5b66\u4e60\u6765\u89e3\u51b3\u95ee\u9898\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u3002\u5b83\u80fd\u591f\u5bf9\u8f93\u5165\u8fdb\u884c\u4e8c\u5143\u5206\u7c7b&#xff0c;\u5728\u5f53\u65f6\u5f15\u8d77\u4e86\u5de8\u5927\u7684\u8f70\u52a8&#xff0c;\u88ab\u89c6\u4e3a\u4eba\u5de5\u667a\u80fd\u7684\u66d9\u5149\u3002<\/p>\n<p>\u201c\u5f02\u6216\u201d\u95ee\u9898\u4e0e\u7b2c\u4e00\u6b21\u5bd2\u51ac&#xff1a; \u7136\u800c&#xff0c;\u521d\u751f\u7684\u559c\u60a6\u662f\u77ed\u6682\u7684\u30021969\u5e74&#xff0c;\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u5148\u9a71\u9a6c\u6587\u00b7\u660e\u65af\u57fa&#xff08;Marvin Minsky&#xff09;\u548c\u897f\u6469\u5c14\u00b7\u4f69\u666e\u7279&#xff08;Seymour Papert&#xff09;\u5728\u5176\u8457\u4f5c\u300a\u611f\u77e5\u673a\u300b&#xff08;Perceptrons&#xff09;\u4e2d&#xff0c;\u4ece\u6570\u5b66\u4e0a\u4e25\u683c\u8bc1\u660e\u4e86\u5355\u5c42\u611f\u77e5\u673a\u65e0\u6cd5\u89e3\u51b3\u7ebf\u6027\u4e0d\u53ef\u5206\u95ee\u9898&#xff0c;\u5176\u4e2d\u6700\u8457\u540d\u7684\u4f8b\u5b50\u5c31\u662f\u201c\u5f02\u6216\u201d&#xff08;XOR&#xff09;\u95ee\u9898\u3002\u8fd9\u4e2a\u7ed3\u8bba\u7ed9\u5f53\u65f6\u8fc7\u4e8e\u4e50\u89c2\u7684\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u6cfc\u4e86\u4e00\u76c6\u51b7\u6c34&#xff0c;\u5bfc\u81f4\u4e86\u8be5\u9886\u57df\u7684\u8d44\u91d1\u548c\u7814\u7a76\u5174\u8da3\u6025\u5267\u4e0b\u964d&#xff0c;\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u8fdb\u5165\u4e86\u957f\u8fbe\u8fd120\u5e74\u7684\u7b2c\u4e00\u6b21\u201cAI\u5bd2\u51ac\u201d\u3002\u8bb8\u591a\u7814\u7a76\u8005\u8f6c\u5411\u4e86\u57fa\u4e8e\u903b\u8f91\u548c\u7b26\u53f7\u63a8\u7406\u7684AI\u65b9\u5411\u3002<\/p>\n<h5>1.2.2 \u6c89\u5bc2\u4e2d\u7684\u575a\u5b88&#xff1a;\u8fde\u63a5\u4e3b\u4e49\u7684\u590d\u5174&#xff08;1980s-1990s&#xff09;<\/h5>\n<p>\u5c3d\u7ba1\u5904\u4e8e\u5bd2\u51ac&#xff0c;\u4f46\u4ecd\u6709\u4e00\u6279\u575a\u5b9a\u7684\u7814\u7a76\u8005&#xff08;\u88ab\u79f0\u4e3a\u201c\u8fde\u63a5\u4e3b\u4e49\u8005\u201d&#xff09;\u5728\u9ed8\u9ed8\u575a\u5b88&#xff0c;\u4ed6\u4eec\u76f8\u4fe1\u795e\u7ecf\u7f51\u7edc\u7684\u6f5c\u529b\u3002\u8f6c\u673a\u51fa\u73b0\u572820\u4e16\u7eaa80\u5e74\u4ee3\u3002<\/p>\n<p>\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u7684\u666e\u53ca&#xff1a; 1986\u5e74&#xff0c;\u7531\u6234\u7ef4\u00b7\u9c81\u59c6\u54c8\u7279&#xff08;David Rumelhart&#xff09;\u3001\u6770\u5f17\u91cc\u00b7\u8f9b\u987f&#xff08;Geoffrey Hinton&#xff09;\u548c\u7f57\u7eb3\u5fb7\u00b7\u5a01\u5ec9\u59c6\u65af&#xff08;Ronald Williams&#xff09;\u7b49\u4eba\u63a8\u5e7f\u7684\u53cd\u5411\u4f20\u64ad&#xff08;Backpropagation&#xff09;\u7b97\u6cd5&#xff0c;\u6709\u6548\u5730\u89e3\u51b3\u4e86\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u96be\u9898\u3002\u5b83\u5229\u7528\u5fae\u79ef\u5206\u4e2d\u7684\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u80fd\u591f\u9ad8\u6548\u5730\u8ba1\u7b97\u51fa\u635f\u5931\u51fd\u6570\u5bf9\u7f51\u7edc\u4e2d\u6bcf\u4e00\u5c42\u53c2\u6570\u7684\u68af\u5ea6&#xff0c;\u4ece\u800c\u5bf9\u7f51\u7edc\u8fdb\u884c\u8fed\u4ee3\u66f4\u65b0\u3002\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u7684\u6210\u719f&#xff0c;\u662f\u795e\u7ecf\u7f51\u7edc\u80fd\u591f\u5904\u7406\u590d\u6742\u4efb\u52a1\u7684\u5173\u952e\u4e00\u6b65&#xff0c;\u81f3\u4eca\u4ecd\u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u6838\u5fc3\u7b97\u6cd5\u3002<\/p>\n<p>\u7ecf\u5178\u6a21\u578b\u7684\u8bde\u751f&#xff1a; \u6709\u4e86\u8bad\u7ec3\u65b9\u6cd5&#xff0c;\u591a\u5c42\u7f51\u7edc\u7684\u5a01\u529b\u5f00\u59cb\u663e\u73b0\u30021998\u5e74&#xff0c;\u6768\u7acb\u6606&#xff08;Yann LeCun&#xff09;\u7b49\u4eba\u63d0\u51fa\u7684LeNet-5\u6a21\u578b&#xff0c;\u4e00\u4e2a\u7ecf\u5178\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;Convolutional Neural Network, CNN&#xff09;&#xff0c;\u5728\u624b\u5199\u6570\u5b57\u8bc6\u522b\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u5de8\u5927\u6210\u529f&#xff0c;\u5e76\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u94f6\u884c\u7684\u652f\u7968\u8bc6\u522b\u7cfb\u7edf\u4e2d\u3002\u8fd9\u8bc1\u660e\u4e86\u6df1\u5c42\u7f51\u7edc\u5728\u771f\u5b9e\u4e16\u754c\u95ee\u9898\u4e2d\u7684\u5e94\u7528\u6f5c\u529b&#xff0c;\u5c24\u5176\u662f\u5728\u56fe\u50cf\u5904\u7406\u9886\u57df\u3002<\/p>\n<h5>1.2.3 \u7206\u53d1\u7684\u524d\u591c&#xff1a;\u74f6\u9888\u4e0e\u7a81\u7834&#xff08;2000s-2010s&#xff09;<\/h5>\n<p>\u8fdb\u516521\u4e16\u7eaa\u521d&#xff0c;\u5c3d\u7ba1\u6709\u4e86\u53cd\u5411\u4f20\u64ad\u548cLeNet-5\u7684\u6210\u529f&#xff0c;\u4f46\u6df1\u5ea6\u5b66\u4e60\u7684\u53d1\u5c55\u518d\u6b21\u9677\u5165\u74f6\u9888&#xff0c;\u88ab\u79f0\u4e3a\u7b2c\u4e8c\u6b21\u201cAI\u5bd2\u51ac\u201d\u3002\u8fd9\u4e3b\u8981\u6e90\u4e8e\u4e09\u5927\u6311\u6218&#xff0c;\u5b83\u4eec\u5982\u540c\u4e09\u5ea7\u5927\u5c71&#xff0c;\u963b\u788d\u7740\u6df1\u5ea6\u5b66\u4e60\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55&#xff1a;<\/p>\n<li>\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8\u95ee\u9898&#xff1a; \u5728\u6df1\u5c42\u7f51\u7edc\u4e2d&#xff0c;\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u65f6\u4f1a\u9010\u5c42\u8870\u51cf&#xff08;\u6d88\u5931&#xff09;\u6216\u6025\u5267\u589e\u5927&#xff08;\u7206\u70b8&#xff09;&#xff0c;\u5bfc\u81f4\u9760\u8fd1\u8f93\u5165\u5c42\u7684\u7f51\u7edc\u65e0\u6cd5\u5f97\u5230\u6709\u6548\u8bad\u7ec3&#xff0c;\u6a21\u578b\u96be\u4ee5\u6536\u655b\u3002<\/li>\n<li>\u7b97\u529b\u4e0d\u8db3&#xff1a; \u8bad\u7ec3\u6df1\u5ea6\u7f51\u7edc\u9700\u8981\u5de8\u5927\u7684\u8ba1\u7b97\u91cf&#xff0c;\u5f53\u65f6\u7684CPU\u6027\u80fd\u8fdc\u4e0d\u80fd\u6ee1\u8db3\u9700\u6c42&#xff0c;\u4f7f\u5f97\u8bad\u7ec3\u65f6\u95f4\u8fc7\u957f&#xff0c;\u96be\u4ee5\u8fdb\u884c\u5927\u89c4\u6a21\u5b9e\u9a8c\u3002<\/li>\n<li>\u6570\u636e\u532e\u4e4f&#xff1a; \u6df1\u5ea6\u5b66\u4e60\u662f\u6570\u636e\u9a71\u52a8\u7684&#xff0c;\u9700\u8981\u6d77\u91cf\u6807\u6ce8\u6570\u636e\u624d\u80fd\u53d1\u6325\u5176\u4f18\u52bf\u3002\u800c\u5f53\u65f6\u7f3a\u4e4f\u5927\u89c4\u6a21\u3001\u9ad8\u8d28\u91cf\u3001\u5e26\u6807\u7b7e\u7684\u516c\u5f00\u6570\u636e\u96c6&#xff0c;\u9650\u5236\u4e86\u6a21\u578b\u7684\u8bad\u7ec3\u89c4\u6a21\u548c\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<p>\u7136\u800c&#xff0c;\u6b63\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898&#xff0c;\u4e09\u5927\u5173\u952e\u7684\u50ac\u5316\u5242\u5e94\u8fd0\u800c\u751f&#xff0c;\u5b83\u4eec\u5982\u540c\u4e09\u628a\u706b\u70ac&#xff0c;\u5171\u540c\u70b9\u71c3\u4e86\u6df1\u5ea6\u5b66\u4e60\u590d\u5174\u7684\u706b\u7130&#xff1a;<\/p>\n<li>\u7b97\u6cd5\u7a81\u7834&#xff1a; \u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u65b0\u7684\u6fc0\u6d3b\u51fd\u6570&#xff08;\u5982ReLU&#xff0c;\u4fee\u6b63\u7ebf\u6027\u5355\u5143&#xff09;\u6765\u6709\u6548\u7f13\u89e3\u68af\u5ea6\u6d88\u5931\u95ee\u9898&#xff0c;\u4ee5\u53ca\u65b0\u7684\u6b63\u5219\u5316\u6280\u672f&#xff08;\u5982Dropout&#xff09;\u6765\u9632\u6b62\u8fc7\u62df\u5408&#xff0c;\u63d0\u9ad8\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002\u6b64\u5916&#xff0c;\u9884\u8bad\u7ec3&#xff08;Pre-training&#xff09;\u548c\u5fae\u8c03&#xff08;Fine-tuning&#xff09;\u7b49\u6280\u672f\u4e5f\u5f00\u59cb\u5d2d\u9732\u5934\u89d2\u3002<\/li>\n<li>\u7b97\u529b\u98de\u8dc3&#xff1a; \u4eba\u4eec\u53d1\u73b0&#xff0c;\u7528\u4e8e\u56fe\u5f62\u6e32\u67d3\u7684GPU&#xff08;\u56fe\u5f62\u5904\u7406\u5668&#xff09;\u5176\u5927\u89c4\u6a21\u5e76\u884c\u8ba1\u7b97\u7684\u67b6\u6784&#xff0c;\u5929\u7136\u9002\u5408\u795e\u7ecf\u7f51\u7edc\u7684\u77e9\u9635\u8fd0\u7b97\u3002NVIDIA\u7b49\u516c\u53f8\u5f00\u59cb\u63d0\u4f9b\u7528\u4e8e\u901a\u7528\u8ba1\u7b97\u7684GPU&#xff08;GPGPU&#xff09;&#xff0c;\u4f7f\u5f97\u8bad\u7ec3\u901f\u5ea6\u63d0\u5347\u4e86\u6570\u5341\u751a\u81f3\u4e0a\u767e\u500d&#xff0c;\u4e3a\u6df1\u5ea6\u5b66\u4e60\u7684\u5927\u89c4\u6a21\u5b9e\u9a8c\u63d0\u4f9b\u4e86\u786c\u4ef6\u57fa\u7840\u3002<\/li>\n<li>\u6570\u636e\u7206\u70b8&#xff1a; \u4e92\u8054\u7f51\u7684\u666e\u53ca\u5e26\u6765\u4e86\u6d77\u91cf\u6570\u636e\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u50cf\u7531\u674e\u98de\u98de\u6559\u6388\u56e2\u961f\u521b\u5efa\u7684ImageNet\u8fd9\u6837\u5305\u542b\u6570\u767e\u4e07\u5f20\u5e26\u6807\u7b7e\u56fe\u50cf\u7684\u5927\u578b\u516c\u5f00\u6570\u636e\u96c6\u7684\u51fa\u73b0&#xff0c;\u4e3a\u8bad\u7ec3\u548c\u8bc4\u6d4b\u6df1\u5ea6\u6a21\u578b\u63d0\u4f9b\u4e86\u524d\u6240\u672a\u6709\u7684\u201c\u71c3\u6599\u201d&#xff0c;\u4f7f\u5f97\u6a21\u578b\u80fd\u591f\u4ece\u6d77\u91cf\u771f\u5b9e\u6570\u636e\u4e2d\u5b66\u4e60\u5230\u66f4\u4e30\u5bcc\u3001\u66f4\u9c81\u68d2\u7684\u7279\u5f81\u3002<\/li>\n<h5>1.2.4 \u9ec4\u91d1\u65f6\u4ee3&#xff1a;AlexNet\u81f3\u4eca&#xff08;2012-\u81f3\u4eca&#xff09;<\/h5>\n<p>\u4e09\u5927\u50ac\u5316\u5242\u7684\u6c47\u805a&#xff0c;\u6700\u7ec8\u57282012\u5e74\u5f15\u7206\u4e86\u6df1\u5ea6\u5b66\u4e60\u7684\u9ec4\u91d1\u65f6\u4ee3\u3002<\/p>\n<p>AlexNet\u7684\u91cc\u7a0b\u7891\u610f\u4e49&#xff1a; 2012\u5e74&#xff0c;\u5728ImageNet\u56fe\u50cf\u8bc6\u522b\u6311\u6218\u8d5b\u4e0a&#xff0c;\u7531\u6770\u5f17\u91cc\u00b7\u8f9b\u987f&#xff08;Geoffrey Hinton&#xff09;\u7684\u5b66\u751f\u4e9a\u5386\u514b\u65af\u00b7\u514b\u91cc\u70ed\u592b\u65af\u57fa&#xff08;Alex Krizhevsky&#xff09;\u6240\u8bbe\u8ba1\u7684AlexNet\u6a21\u578b&#xff0c;\u4ee5\u8fdc\u8d85\u7b2c\u4e8c\u540d&#xff08;\u8d85\u8fc710\u4e2a\u767e\u5206\u70b9&#xff09;\u7684\u60ca\u4eba\u51c6\u786e\u7387\u593a\u5f97\u51a0\u519b\u3002\u5b83\u6210\u529f\u5730\u5c06ReLU\u3001Dropout\u3001GPU\u52a0\u901f\u7b49\u6280\u672f\u878d\u4e8e\u4e00\u4e2a\u6df1\u5c42\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u4e2d&#xff0c;\u5176\u538b\u5012\u6027\u7684\u80dc\u5229&#xff0c;\u65e0\u53ef\u8fa9\u9a73\u5730\u5411\u5168\u4e16\u754c\u5ba3\u544a&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u7684\u65f6\u4ee3&#xff0c;\u6b63\u5f0f\u5230\u6765\u3002 \u8fd9\u4e5f\u6807\u5fd7\u7740\u4eba\u5de5\u667a\u80fd\u9886\u57df\u4ece\u201c\u7b26\u53f7\u4e3b\u4e49\u201d\u5411\u201c\u8fde\u63a5\u4e3b\u4e49\u201d\u7684\u8303\u5f0f\u8f6c\u79fb\u3002<\/p>\n<p>\u767e\u82b1\u9f50\u653e&#xff1a; AlexNet\u4e4b\u540e&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u8fdb\u5165\u4e86\u7206\u70b8\u5f0f\u53d1\u5c55\u7684\u9ec4\u91d1\u65f6\u4ee3\u3002VGG\u3001GoogLeNet\u3001ResNet\u7b49\u66f4\u6df1\u3001\u66f4\u7cbe\u5de7\u7684CNN\u67b6\u6784\u4e0d\u65ad\u5237\u65b0\u7740\u56fe\u50cf\u8bc6\u522b\u7684\u8bb0\u5f55&#xff0c;\u63a8\u52a8\u4e86\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u7684\u98de\u901f\u53d1\u5c55&#xff1b;\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09;\u53ca\u5176\u53d8\u4f53\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;\u5728\u5904\u7406\u5e8f\u5217\u6570\u636e&#xff08;\u5982\u6587\u672c\u548c\u8bed\u97f3&#xff09;\u4e0a\u5927\u653e\u5f02\u5f69&#xff0c;\u4e3a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u548c\u8bed\u97f3\u8bc6\u522b\u5e26\u6765\u4e86\u9769\u547d&#xff1b;\u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;\u8ba9\u6211\u4eec\u770b\u5230\u4e86\u673a\u5668\u521b\u9020\u7684\u53ef\u80fd&#xff0c;\u751f\u6210\u903c\u771f\u7684\u56fe\u50cf\u548c\u827a\u672f\u4f5c\u54c1&#xff1b;\u800c\u8fd1\u5e74\u6765&#xff0c;Transformer\u67b6\u6784\u66f4\u662f\u98a0\u8986\u4e86\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df&#xff0c;\u50ac\u751f\u4e86BERT\u3001GPT\u7b49\u4e00\u7cfb\u5217\u9884\u8bad\u7ec3\u5927\u6a21\u578b&#xff0c;\u5e76\u5f00\u59cb\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u7b49\u591a\u4e2a\u9886\u57df\u5c55\u73b0\u5176\u7edf\u6cbb\u529b\u3002<\/p>\n<p>\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u5df2\u7136\u6210\u4e3a\u4eba\u5de5\u667a\u80fd\u9886\u57df\u6700\u8000\u773c\u7684\u4e3b\u65cb\u5f8b&#xff0c;\u5176\u5e94\u7528\u5df2\u6e17\u900f\u5230\u6211\u4eec\u751f\u6d3b\u7684\u65b9\u65b9\u9762\u9762&#xff0c;\u4ece\u667a\u80fd\u624b\u673a\u5230\u81ea\u52a8\u9a7e\u9a76&#xff0c;\u4ece\u533b\u7597\u8bca\u65ad\u5230\u79d1\u5b66\u7814\u7a76&#xff0c;\u65e0\u4e0d\u95ea\u8000\u7740\u5176\u667a\u6167\u7684\u5149\u8292\u3002<\/p>\n<h4>1.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u5e94\u8bb8\u4e4b\u5730\u201d<\/h4>\n<p>\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u5b8f\u5927\u53d9\u4e8b\u4e2d&#xff0c;\u7f16\u7a0b\u8bed\u8a00\u662f\u6211\u4eec\u8d56\u4ee5\u5c06\u601d\u60f3\u84dd\u56fe\u6784\u7b51\u4e3a\u73b0\u5b9e\u5947\u89c2\u7684\u57fa\u77f3\u3002\u800c\u5728\u4f17\u591a\u4f18\u79c0\u7684\u7f16\u7a0b\u8bed\u8a00\u4e2d&#xff0c;Python\u4ee5\u4e00\u79cd\u8fd1\u4e4e\u201c\u5929\u9009\u201d\u7684\u59ff\u6001&#xff0c;\u6210\u4e3a\u4e86\u6df1\u5ea6\u5b66\u4e60\u4e43\u81f3\u6574\u4e2a\u6570\u636e\u79d1\u5b66\u9886\u57df\u7684\u901a\u7528\u8bed&#xff08;Lingua Franca&#xff09;\u3002<\/p>\n<p>\u8fd9\u79cd\u538b\u5012\u6027\u7684\u4f18\u52bf\u5e76\u975e\u5076\u7136&#xff0c;\u800c\u662f\u6e90\u4e8e\u5176\u5185\u5728\u54f2\u5b66\u3001\u5916\u5728\u751f\u6001\u4e0e\u793e\u533a\u6587\u5316\u4e09\u8005\u7684\u5b8c\u7f8e\u5171\u9e23&#xff0c;\u5171\u540c\u5c06Python\u5851\u9020\u6210\u4e86\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u5e94\u8bb8\u4e4b\u5730\u201d\u3002\u5bf9\u4e8e\u6709\u5fd7\u4e8e\u6b64\u7684\u63a2\u7d22\u8005\u800c\u8a00&#xff0c;\u7406\u89e3Python\u4e3a\u4f55\u80dc\u51fa&#xff0c;\u5c31\u5982\u540c\u5251\u5ba2\u61c2\u5f97\u81ea\u5df1\u624b\u4e2d\u4e4b\u5251\u7684\u7279\u6027&#xff0c;\u662f\u8e0f\u4e0a\u5f81\u9014\u524d\u7684\u91cd\u8981\u4e00\u8bfe\u3002<\/p>\n<h5>1.3.1 \u201c\u5f00\u53d1\u6548\u7387\u4f18\u5148\u201d\u7684\u5185\u5728\u54f2\u5b66<\/h5>\n<p>Python\u7684\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66&#xff0c;\u53ef\u4ee5\u7528\u201cThe Zen of Python\u201d&#xff08;Python\u4e4b\u7985&#xff09;\u4e2d\u7684\u51e0\u53e5\u8bdd\u6765\u6982\u62ec&#xff1a;\u201c\u4f18\u7f8e\u80dc\u4e8e\u4e11\u964b&#xff0c;\u660e\u4e86\u80dc\u4e8e\u6666\u6da9&#xff0c;\u7b80\u6d01\u80dc\u4e8e\u590d\u6742\u201d\u3002\u8fd9\u79cd\u5bf9\u53ef\u8bfb\u6027\u548c\u7b80\u6d01\u6027\u7684\u6781\u81f4\u8ffd\u6c42&#xff0c;\u76f4\u63a5\u8f6c\u5316\u4e3a\u65e0\u4e0e\u4f26\u6bd4\u7684\u5f00\u53d1\u6548\u7387\u3002<\/p>\n<p>\u4f4e\u8ba4\u77e5\u8d1f\u8377&#xff1a; Python\u7684\u8bed\u6cd5\u7ed3\u6784\u6e05\u6670&#xff0c;\u63a5\u8fd1\u81ea\u7136\u8bed\u8a00&#xff0c;\u8ba9\u5f00\u53d1\u8005\u53ef\u4ee5\u5c06\u5b9d\u8d35\u7684\u8111\u529b\u8d44\u6e90\u96c6\u4e2d\u4e8e\u89e3\u51b3\u95ee\u9898\u672c\u8eab\u2014\u2014\u5373\u6a21\u578b\u7684\u8bbe\u8ba1\u3001\u7b97\u6cd5\u7684\u4f18\u5316\u548c\u5b9e\u9a8c\u7684\u8fed\u4ee3\u2014\u2014\u800c\u4e0d\u662f\u8017\u8d39\u5728\u5904\u7406\u7e41\u7410\u7684\u8bed\u6cd5\u7ec6\u8282\u3001\u6307\u9488\u7ba1\u7406\u6216\u5185\u5b58\u5206\u914d\u4e0a\u3002\u5728\u9700\u8981\u5feb\u901f\u8fed\u4ee3\u548c\u5b9e\u9a8c\u7684\u6df1\u5ea6\u5b66\u4e60\u7814\u7a76\u4e2d&#xff0c;\u8fd9\u79cd\u201c\u6240\u601d\u5373\u6240\u5f97\u201d\u7684\u8868\u8fbe\u80fd\u529b\u81f3\u5173\u91cd\u8981\u3002<\/p>\n<p>\u5feb\u901f\u539f\u578b\u9a8c\u8bc1&#xff1a; \u4ece\u4e00\u4e2a\u7b97\u6cd5\u6784\u601d\u5230\u53ef\u8fd0\u884c\u7684\u539f\u578b\u4ee3\u7801&#xff0c;Python\u7684\u5f00\u53d1\u5468\u671f\u6781\u77ed\u3002\u8fd9\u4f7f\u5f97\u7814\u7a76\u8005\u548c\u5de5\u7a0b\u5e08\u80fd\u591f\u8fc5\u901f\u68c0\u9a8c\u65b0\u60f3\u6cd5\u7684\u6709\u6548\u6027&#xff0c;\u5927\u5927\u52a0\u901f\u4e86\u6574\u4e2a\u9886\u57df\u7684\u521b\u65b0\u6b65\u4f10\u3002<\/p>\n<h5>1.3.2 \u201c\u80f6\u6c34\u7279\u6027\u201d\u4e0e\u201c\u5de8\u4eba\u80a9\u8180\u201d\u7684\u5916\u5728\u751f\u6001<\/h5>\n<p>\u5982\u679c\u8bf4\u7b80\u6d01\u7684\u54f2\u5b66\u662fPython\u7684\u7075\u9b42&#xff0c;\u90a3\u4e48\u5176\u5f3a\u5927\u7684\u79d1\u5b66\u8ba1\u7b97\u751f\u6001\u5c31\u662f\u5b83\u575a\u5b9e\u7684\u8eaf\u4f53\u3002Python\u672c\u8eab\u5e76\u975e\u4ee5\u6781\u81f4\u7684\u8fd0\u884c\u901f\u5ea6\u89c1\u957f&#xff0c;\u4f46\u5b83\u62e5\u6709\u5f3a\u5927\u7684\u201c\u80f6\u6c34\u7279\u6027\u201d&#xff0c;\u80fd\u591f\u9ad8\u6548\u5730\u7c98\u5408\u7531C\u3001C&#043;&#043;\u6216Fortran\u7b49\u9ad8\u6027\u80fd\u8bed\u8a00\u7f16\u5199\u7684\u5e95\u5c42\u8ba1\u7b97\u5e93\u3002\u8fd9\u8ba9Python\u5f97\u4ee5\u201c\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a\u201d&#xff0c;\u517c\u5177\u5f00\u53d1\u6548\u7387\u4e0e\u8fd0\u884c\u6027\u80fd\u3002<\/p>\n<p>\u79d1\u5b66\u8ba1\u7b97\u7684\u201c\u4e09\u9a7e\u9a6c\u8f66\u201d&#xff1a;<\/p>\n<ul>\n<li>NumPy: \u63d0\u4f9b\u4e86\u9ad8\u6027\u80fd\u7684\u591a\u7ef4\u6570\u7ec4\u5bf9\u8c61ndarray\u548c\u4e30\u5bcc\u7684\u6570\u5b66\u51fd\u6570&#xff0c;\u662f\u51e0\u4e4e\u6240\u6709\u6570\u636e\u79d1\u5b66\u5de5\u5177\u7684\u5e95\u5c42\u4f9d\u8d56&#xff0c;\u4e3aPython\u6ce8\u5165\u4e86\u5f3a\u5927\u7684\u6570\u503c\u8ba1\u7b97\u80fd\u529b\u3002<\/li>\n<li>Pandas: \u63d0\u4f9b\u4e86DataFrame\u8fd9\u4e00\u7075\u6d3b\u3001\u5f3a\u5927\u7684\u6570\u636e\u7ed3\u6784&#xff0c;\u8ba9\u7ed3\u6784\u5316\u6570\u636e\u7684\u6e05\u6d17\u3001\u5904\u7406\u3001\u5206\u6790\u53d8\u5f97\u5f02\u5e38\u8f7b\u677e\u3002<\/li>\n<li>Matplotlib\/Seaborn: \u63d0\u4f9b\u4e86\u4ece\u57fa\u7840\u5230\u9ad8\u7ea7\u7684\u6570\u636e\u53ef\u89c6\u5316\u80fd\u529b&#xff0c;\u662f\u63a2\u7d22\u6570\u636e\u3001\u5448\u73b0\u7ed3\u679c\u4e0d\u53ef\u6216\u7f3a\u7684\u5de5\u5177\u3002<\/li>\n<\/ul>\n<p>\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u7684\u201c\u4e00\u81f4\u9009\u62e9\u201d&#xff1a; \u5f53\u4eca\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u4e24\u5927\u5de8\u5934\u2014\u2014Google\u652f\u6301\u7684TensorFlow\u548cFacebook&#xff08;Meta&#xff09;\u652f\u6301\u7684PyTorch\u2014\u2014\u90fd\u5c06Python\u4f5c\u4e3a\u5176\u6700\u4e3b\u8981\u3001\u6700\u4f18\u5148\u652f\u6301\u7684\u524d\u7aef\u5f00\u53d1\u63a5\u53e3\u3002\u5b83\u4eec\u901a\u8fc7Python\u5c01\u88c5\u4e86\u5e95\u5c42\u7684CUDA\u6838\u5fc3&#xff0c;\u8ba9\u5f00\u53d1\u8005\u53ef\u4ee5\u7528\u7b80\u6d01\u7684Python\u4ee3\u7801&#xff0c;\u9a71\u52a8GPU\u8fdb\u884c\u5927\u89c4\u6a21\u5e76\u884c\u8ba1\u7b97\u3002\u8fd9\u79cd\u4e1a\u754c\u5de8\u5934\u7684\u4e00\u81f4\u9009\u62e9&#xff0c;\u53cd\u8fc7\u6765\u4e5f\u6781\u5927\u5730\u5de9\u56fa\u4e86Python\u5728\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u7edf\u6cbb\u5730\u4f4d\u3002<\/p>\n<h5>1.3.3 \u201c\u5171\u5efa\u5171\u4eab\u201d\u7684\u793e\u533a\u6587\u5316<\/h5>\n<p>\u6280\u672f\u7684\u53d1\u5c55&#xff0c;\u5f52\u6839\u7ed3\u5e95\u662f\u4eba\u7684\u53d1\u5c55\u3002Python\u62e5\u6709\u4e00\u4e2a\u5168\u7403\u8303\u56f4\u5185\u89c4\u6a21\u6700\u5927\u3001\u6700\u6d3b\u8dc3\u3001\u6700\u5177\u534f\u4f5c\u7cbe\u795e\u7684\u5f00\u6e90\u793e\u533a\u3002\u8fd9\u79cd\u793e\u533a\u6587\u5316\u662fPython\u751f\u6001\u80fd\u591f\u6301\u7eed\u7e41\u8363\u7684\u6d3b\u6c34\u6e90\u5934\u3002<\/p>\n<ul>\n<li>\u77e5\u8bc6\u7684\u6c11\u4e3b\u5316&#xff1a; \u65e0\u8bba\u662f\u521d\u5b66\u8005\u9047\u5230\u7684\u4e00\u4e2a\u7b80\u5355\u62a5\u9519&#xff0c;\u8fd8\u662f\u524d\u6cbf\u7814\u7a76\u6240\u9700\u7684\u590d\u6742\u7b97\u6cd5\u5b9e\u73b0&#xff0c;\u5728Python\u793e\u533a\u4e2d&#xff08;\u5982Stack Overflow\u3001GitHub\u3001\u5404\u7c7b\u8bba\u575b&#xff09;&#xff0c;\u51e0\u4e4e\u603b\u80fd\u627e\u5230\u7b54\u6848\u3001\u6559\u7a0b\u6216\u73b0\u6210\u7684\u5f00\u6e90\u4ee3\u7801\u3002\u8fd9\u6781\u5927\u5730\u964d\u4f4e\u4e86\u5b66\u4e60\u548c\u8fdb\u5165\u8be5\u9886\u57df\u7684\u95e8\u69db\u3002<\/li>\n<li>\u521b\u65b0\u7684\u52a0\u901f\u5668&#xff1a; \u4e00\u7bc7\u91cd\u8981\u7684\u6df1\u5ea6\u5b66\u4e60\u8bba\u6587\u53d1\u8868\u540e&#xff0c;\u5f80\u5f80\u5728\u51e0\u5929\u751a\u81f3\u51e0\u5c0f\u65f6\u5185&#xff0c;\u793e\u533a\u4e2d\u5c31\u4f1a\u51fa\u73b0\u5176PyTorch\u6216TensorFlow\u7684\u5f00\u6e90\u590d\u73b0\u4ee3\u7801\u3002\u8fd9\u79cd\u201c\u601d\u60f3-\u4ee3\u7801-\u5e94\u7528\u201d\u7684\u5feb\u901f\u8f6c\u5316\u94fe\u6761&#xff0c;\u5f97\u76ca\u4e8e\u5f00\u6e90\u793e\u533a\u7684\u201c\u5171\u5efa\u5171\u4eab\u201d\u6587\u5316&#xff0c;\u5b83\u4ee5\u524d\u6240\u672a\u6709\u7684\u901f\u5ea6\u63a8\u52a8\u7740\u6574\u4e2a\u9886\u57df\u7684\u8fb9\u754c\u5411\u524d\u62d3\u5c55\u3002<\/li>\n<li>\u4eba\u624d\u7684\u84c4\u6c34\u6c60&#xff1a; \u5e9e\u5927\u7684\u793e\u533a\u4e5f\u610f\u5473\u7740\u5e9e\u5927\u7684\u4eba\u624d\u5e93\u3002\u5bf9\u4e8e\u4f01\u4e1a\u800c\u8a00&#xff0c;\u9009\u62e9Python\u6280\u672f\u6808\u610f\u5473\u7740\u66f4\u5bb9\u6613\u62db\u8058\u5230\u5408\u9002\u7684\u5f00\u53d1\u4eba\u624d&#xff1b;\u5bf9\u4e8e\u4e2a\u4eba\u800c\u8a00&#xff0c;\u638c\u63e1Python\u5219\u610f\u5473\u7740\u62e5\u6709\u4e86\u901a\u5f80\u6570\u636e\u79d1\u5b66\u548c\u4eba\u5de5\u667a\u80fd\u9886\u57df\u6700\u91cd\u8981\u7684\u4e00\u5f20\u201c\u901a\u884c\u8bc1\u201d\u3002<\/li>\n<\/ul>\n<p>\u7efc\u4e0a\u6240\u8ff0&#xff0c;\u9009\u62e9Python\u4f5c\u4e3a\u5b66\u4e60\u6df1\u5ea6\u5b66\u4e60\u7684\u8bed\u8a00&#xff0c;\u5e76\u975e\u4ec5\u4ec5\u662f\u9009\u62e9\u4e00\u4e2a\u5de5\u5177&#xff0c;\u66f4\u662f\u9009\u62e9\u878d\u5165\u4e00\u4e2a\u9ad8\u6548\u7684\u5f00\u53d1\u8303\u5f0f\u3001\u4e00\u4e2a\u6210\u719f\u5f3a\u5927\u7684\u6280\u672f\u751f\u6001&#xff0c;\u4ee5\u53ca\u4e00\u4e2a\u5145\u6ee1\u6d3b\u529b\u4e0e\u667a\u6167\u7684\u5168\u7403\u6027\u793e\u533a\u3002\u5b83\u5c06\u662f\u60a8\u5728\u8fd9\u8d9f\u63a2\u7d22\u4e4b\u65c5\u4e2d\u6700\u503c\u5f97\u4fe1\u8d56\u7684\u4f19\u4f34\u3002<\/p>\n<h4>1.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66<\/h4>\n<p>\u4e3a\u4e86\u5f15\u5bfc\u60a8\u7cfb\u7edf\u5730\u3001\u7531\u6d45\u5165\u6df1\u5730\u638c\u63e1\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u672c\u4e66\u7cbe\u5fc3\u8bbe\u8ba1\u4e86\u4e94\u4e2a\u5faa\u5e8f\u6e10\u8fdb\u7684\u7bc7\u7ae0&#xff0c;\u5b83\u4eec\u5171\u540c\u6784\u6210\u4e86\u4e00\u5f20\u4ece\u201c\u77e5\u201d\u5230\u201c\u884c\u201d\u518d\u5230\u201c\u601d\u201d\u7684\u5b8c\u6574\u667a\u6167\u5730\u56fe\u3002<\/p>\n<h5>1.4.1 \u5168\u4e66\u4e94\u5927\u7bc7\u7ae0\u6982\u89c8<\/h5>\n<p>1.\u57fa\u7840\u7bc7&#xff08;\u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3&#xff09;&#xff1a; \u672c\u7bc7\u7ae0\u5c06\u4e3a\u60a8\u6253\u4e0b\u575a\u5b9e\u7684\u7406\u8bba\u4e0e\u5b9e\u8df5\u57fa\u7840\u3002\u60a8\u5c06\u5398\u6e05\u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60\u7684\u5185\u5728\u903b\u8f91&#xff0c;\u56de\u987e\u5176\u6ce2\u6f9c\u58ee\u9614\u7684\u53d1\u5c55\u5386\u7a0b&#xff0c;\u5e76\u638c\u63e1\u6df1\u5ea6\u5b66\u4e60\u6240\u5fc5\u9700\u7684\u6570\u5b66\u8bed\u8a00\u548cPython\u7f16\u7a0b\u5de5\u5177\u3002\u540c\u65f6&#xff0c;\u901a\u8fc7\u5bf9\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u7684\u6e29\u6545\u77e5\u65b0&#xff0c;\u60a8\u5c06\u66f4\u597d\u5730\u7406\u89e3\u6df1\u5ea6\u5b66\u4e60\u7684\u7a81\u7834\u4e0e\u521b\u65b0\u3002<\/p>\n<p>2.\u6838\u5fc3\u7bc7&#xff08;\u6df1\u5165\u795e\u7ecf\u7f51\u7edc\u7684\u6bbf\u5802&#xff09;&#xff1a; \u5728\u672c\u7bc7\u7ae0\u4e2d&#xff0c;\u60a8\u5c06\u6b63\u5f0f\u8e0f\u5165\u795e\u7ecf\u7f51\u7edc\u7684\u6838\u5fc3\u9886\u57df\u3002\u4ece\u6700\u57fa\u672c\u7684\u795e\u7ecf\u5143\u5230\u591a\u5c42\u611f\u77e5\u673a&#xff0c;\u4ece\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u5230\u4e3b\u6d41\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6&#xff08;TensorFlow\/PyTorch&#xff09;\u7684\u5b9e\u6218\u8fd0\u7528&#xff0c;\u60a8\u5c06\u7cfb\u7edf\u5b66\u4e60\u795e\u7ecf\u7f51\u7edc\u7684\u6784\u5efa\u3001\u8bad\u7ec3\u4e0e\u4f18\u5316\u6280\u5de7&#xff0c;\u638c\u63e1\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u70bc\u4e39\u672f\u201d\u3002<\/p>\n<p>3.\u8fdb\u9636\u7bc7&#xff08;\u638c\u63e1\u6838\u5fc3\u7f51\u7edc\u67b6\u6784&#xff09;&#xff1a; \u672c\u7bc7\u7ae0\u5c06\u5e26\u60a8\u6df1\u5165\u63a2\u7d22\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u6700\u91cd\u8981\u3001\u6700\u5177\u5f71\u54cd\u529b\u7684\u51e0\u7c7b\u7f51\u7edc\u67b6\u6784\u3002\u60a8\u5c06\u5b66\u4e60\u7528\u4e8e\u56fe\u50cf\u5904\u7406\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u3001\u7528\u4e8e\u5e8f\u5217\u6570\u636e\u5206\u6790\u7684\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09;&#xff0c;\u4ee5\u53ca\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\u6380\u8d77\u9769\u547d\u7684Transformer\u67b6\u6784\u3002\u6b64\u5916&#xff0c;\u8fd8\u5c06\u4ecb\u7ecd\u751f\u6210\u5f0f\u6a21\u578b&#xff0c;\u5f00\u542f\u60a8\u7684\u521b\u9020\u4e4b\u65c5\u3002<\/p>\n<p>4.\u5b9e\u6218\u7bc7&#xff08;\u4ece\u7406\u8bba\u5230\u4ef7\u503c\u7684\u8f6c\u5316&#xff09;&#xff1a; \u7406\u8bba\u7684\u4ef7\u503c\u5728\u4e8e\u6307\u5bfc\u5b9e\u8df5&#xff0c;\u800c\u5b9e\u8df5\u7684\u610f\u4e49\u5728\u4e8e\u521b\u9020\u4ef7\u503c\u3002\u672c\u7bc7\u7ae0\u5c06\u5e26\u60a8\u8d70\u51fa\u8c61\u7259\u5854&#xff0c;\u7528\u6240\u5b66\u77e5\u8bc6\u89e3\u51b3\u771f\u5b9e\u4e16\u754c\u7684\u95ee\u9898\u3002\u901a\u8fc7\u591a\u4e2a\u5b8c\u6574\u7684\u9879\u76ee\u5b9e\u6218&#xff0c;\u6db5\u76d6\u8ba1\u7b97\u673a\u89c6\u89c9\u3001\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7b49\u70ed\u95e8\u9886\u57df&#xff0c;\u60a8\u5c06\u4f53\u9a8c\u4ece\u6570\u636e\u5904\u7406\u3001\u6a21\u578b\u8bad\u7ec3\u3001\u8bc4\u4f30\u5230\u6700\u7ec8\u90e8\u7f72\u7684\u5b8c\u6574\u6d41\u7a0b&#xff0c;\u5177\u5907\u5c06AI\u6280\u672f\u4ea7\u4e1a\u5316\u7684\u80fd\u529b\u3002<\/p>\n<p>5.\u5c55\u671b\u7bc7&#xff08;\u63a2\u7d22\u672a\u6765\u7684\u8fb9\u754c&#xff09;&#xff1a; \u6280\u672f\u7684\u53d1\u5c55\u6c38\u65e0\u6b62\u5883&#xff0c;\u667a\u6167\u7684\u63a2\u7d22\u4e5f\u6c38\u4e0d\u505c\u6b47\u3002\u672c\u7bc7\u7ae0\u5c06\u5e26\u60a8\u7ad9\u5728\u65f6\u4ee3\u7684\u524d\u6cbf&#xff0c;\u773a\u671b\u6df1\u5ea6\u5b66\u4e60\u7684\u672a\u6765\u53d1\u5c55\u65b9\u5411\u3002\u60a8\u5c06\u4e86\u89e3\u56fe\u795e\u7ecf\u7f51\u7edc\u3001\u8054\u90a6\u5b66\u4e60\u3001\u53ef\u89e3\u91caAI\u3001\u591a\u6a21\u6001\u5b66\u4e60\u7b49\u524d\u6cbf\u6280\u672f&#xff0c;\u5e76\u6df1\u5165\u63a2\u8ba8AI\u4f26\u7406\u4e0e\u793e\u4f1a\u8d23\u4efb\u7b49\u91cd\u8981\u8bae\u9898&#xff0c;\u6210\u4e3a\u4e00\u4e2a\u6709\u8fdc\u89c1\u3001\u6709\u8d23\u4efb\u611f\u7684AI\u4ece\u4e1a\u8005\u3002<\/p>\n<h5>1.4.2 \u5efa\u8bae\u7684\u5b66\u4e60\u8def\u5f84<\/h5>\n<p>\u5b66\u4e60\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u5982\u540c\u6500\u767b\u4e00\u5ea7\u9ad8\u5cf0&#xff0c;\u9700\u8981\u5faa\u5e8f\u6e10\u8fdb&#xff0c;\u6b65\u6b65\u4e3a\u8425\u3002\u6211\u4eec\u4e3a\u60a8\u63d0\u4f9b\u4ee5\u4e0b\u4e24\u79cd\u5efa\u8bae\u7684\u5b66\u4e60\u8def\u5f84&#xff1a;<\/p>\n<ul>\n<li>\u5bf9\u4e8e\u521d\u5b66\u8005&#xff08;\u96f6\u57fa\u7840\u6216\u57fa\u7840\u8584\u5f31&#xff09;&#xff1a; \u5f3a\u70c8\u5efa\u8bae\u60a8\u6309\u90e8\u5c31\u73ed\u5730\u9605\u8bfb\u672c\u4e66\u3002\u8bf7\u52a1\u5fc5\u6295\u5165\u8db3\u591f\u7684\u65f6\u95f4\u548c\u7cbe\u529b&#xff0c;\u8ba4\u771f\u6d88\u5316\u57fa\u7840\u7bc7\u7684\u5185\u5bb9&#xff0c;\u5c24\u5176\u662f\u6570\u5b66\u4e0e\u7f16\u7a0b\u57fa\u7840\u3002\u5728\u5b66\u4e60\u8fc7\u7a0b\u4e2d&#xff0c;\u8bf7\u52a1\u5fc5\u4eb2\u624b\u6572\u5165\u5e76\u8fd0\u884c\u4e66\u4e2d\u7684\u6bcf\u4e00\u4e2a\u4ee3\u7801\u793a\u4f8b&#xff0c;\u5b8c\u6210\u6bcf\u4e00\u4e2a\u7ec3\u4e60\u3002\u4e0d\u8981\u6025\u4e8e\u6c42\u6210&#xff0c;\u6253\u7262\u57fa\u7840\u662f\u60a8\u672a\u6765\u80fd\u591f\u8d70\u5f97\u66f4\u8fdc\u3001\u6500\u5f97\u66f4\u9ad8\u7684\u5173\u952e\u3002<\/li>\n<li>\u5bf9\u4e8e\u6709\u7ecf\u9a8c\u8005&#xff08;\u6709\u4e00\u5b9a\u7f16\u7a0b\u6216\u673a\u5668\u5b66\u4e60\u57fa\u7840&#xff09;&#xff1a; \u5982\u679c\u60a8\u5df2\u7ecf\u5177\u5907\u4e00\u5b9a\u7684Python\u7f16\u7a0b\u7ecf\u9a8c\u6216\u673a\u5668\u5b66\u4e60\u57fa\u7840\u77e5\u8bc6&#xff0c;\u53ef\u4ee5\u5c06\u57fa\u7840\u7bc7\u4f5c\u4e3a\u5feb\u901f\u56de\u987e\u548c\u67e5\u6f0f\u8865\u7f3a\u7684\u5de5\u5177\u3002\u60a8\u53ef\u4ee5\u91cd\u70b9\u6295\u5165\u5230\u6838\u5fc3\u7bc7\u548c\u8fdb\u9636\u7bc7\u7684\u5b66\u4e60\u4e2d&#xff0c;\u6df1\u5165\u7406\u89e3\u5404\u79cd\u6a21\u578b\u67b6\u6784\u548c\u8bad\u7ec3\u6280\u5de7\u3002\u540c\u65f6&#xff0c;\u8bf7\u76f4\u63a5\u6311\u6218\u5b9e\u6218\u7bc7\u7684\u9879\u76ee&#xff0c;\u901a\u8fc7\u5b9e\u9645\u64cd\u4f5c\u6765\u68c0\u9a8c\u548c\u63d0\u5347\u81ea\u5df1\u7684\u80fd\u529b&#xff0c;\u5e76\u5c06\u672c\u4e66\u4f5c\u4e3a\u4e00\u672c\u5b9e\u7528\u7684\u53c2\u8003\u624b\u518c\u3002<\/li>\n<\/ul>\n<p>\u65e0\u8bba\u60a8\u9009\u62e9\u54ea\u79cd\u8def\u5f84&#xff0c;\u8bf7\u8bb0\u4f4f&#xff1a;\u5b9e\u8df5\u662f\u68c0\u9a8c\u771f\u7406\u7684\u552f\u4e00\u6807\u51c6\u3002 \u7406\u8bba\u77e5\u8bc6\u7684\u638c\u63e1\u56fa\u7136\u91cd\u8981&#xff0c;\u4f46\u53ea\u6709\u901a\u8fc7\u5927\u91cf\u7684\u52a8\u624b\u5b9e\u8df5&#xff0c;\u624d\u80fd\u771f\u6b63\u5c06\u77e5\u8bc6\u5185\u5316\u4e3a\u60a8\u7684\u80fd\u529b\u3002<\/p>\n<h5>1.4.3 \u672c\u4e66\u7279\u8272&#xff1a;\u7406\u8bba\u3001\u4ee3\u7801\u4e0e\u5fc3\u6cd5\u7684\u7ed3\u5408<\/h5>\n<p>\u4e3a\u4e86\u786e\u4fdd\u60a8\u80fd\u591f\u9ad8\u6548\u3001\u6df1\u5165\u5730\u5b66\u4e60&#xff0c;\u672c\u4e66\u5728\u5185\u5bb9\u7f16\u6392\u548c\u5448\u73b0\u65b9\u5f0f\u4e0a&#xff0c;\u529b\u6c42\u505a\u5230\u4ee5\u4e0b\u51e0\u70b9&#xff1a;<\/p>\n<ul>\n<li>\u7406\u8bba\u6df1\u5ea6\u4e0e\u5b9e\u8df5\u5e7f\u5ea6\u5e76\u91cd&#xff1a; \u6211\u4eec\u4e0d\u4ec5\u4f1a\u6df1\u5165\u5256\u6790\u7b97\u6cd5\u7684\u6570\u5b66\u539f\u7406\u548c\u8bbe\u8ba1\u601d\u60f3&#xff0c;\u66f4\u4f1a\u63d0\u4f9b\u4e30\u5bcc\u7684\u4ee3\u7801\u793a\u4f8b\u548c\u5b9e\u6218\u9879\u76ee&#xff0c;\u786e\u4fdd\u60a8\u65e2\u80fd\u201c\u77e5\u5176\u7136\u201d&#xff0c;\u66f4\u80fd\u201c\u77e5\u5176\u6240\u4ee5\u7136\u201d&#xff0c;\u5e76\u6700\u7ec8\u201c\u884c\u5176\u9053\u201d\u3002<\/li>\n<li>\u56fe\u6587\u5e76\u8302&#xff0c;\u6df1\u5165\u6d45\u51fa&#xff1a; \u590d\u6742\u7684\u6982\u5ff5\u548c\u62bd\u8c61\u7684\u539f\u7406&#xff0c;\u6211\u4eec\u5c06\u901a\u8fc7\u5927\u91cf\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u56fe\u8868\u3001\u6d41\u7a0b\u56fe\u548c\u53ef\u89c6\u5316\u6848\u4f8b\u6765\u8f85\u52a9\u89e3\u91ca&#xff0c;\u529b\u6c42\u5316\u7e41\u4e3a\u7b80&#xff0c;\u8ba9\u6666\u6da9\u96be\u61c2\u7684\u77e5\u8bc6\u53d8\u5f97\u76f4\u89c2\u6613\u61c2\u3002<\/li>\n<li>\u4ee3\u7801\u5373\u4fee\u884c&#xff0c;\u6ce8\u91cd\u5b9e\u64cd&#xff1a; \u4e66\u4e2d\u6240\u6709\u4ee3\u7801\u5747\u63d0\u4f9b\u5b8c\u6574\u6ce8\u91ca&#xff0c;\u5e76\u9f13\u52b1\u60a8\u4eb2\u624b\u6572\u5165\u3001\u8fd0\u884c\u3001\u4fee\u6539\u548c\u5b9e\u9a8c\u3002\u6211\u4eec\u76f8\u4fe1&#xff0c;\u53ea\u6709\u7ecf\u60a8\u4e4b\u624b\u7684\u4ee3\u7801&#xff0c;\u624d\u80fd\u771f\u6b63\u5185\u5316\u4e3a\u60a8\u7684\u80fd\u529b\u3002\u6240\u6709\u4ee3\u7801\u90fd\u5c06\u662f\u53ef\u8fd0\u884c\u3001\u53ef\u590d\u73b0\u7684&#xff0c;\u5e76\u4e0e\u771f\u5b9e\u9879\u76ee\u63a5\u8f68\u3002<\/li>\n<li>\u201c\u667a\u6167\u9526\u56ca\u201d&#xff0c;\u70b9\u62e8\u8ff7\u6d25&#xff1a; \u5728\u5173\u952e\u7ae0\u8282\u6216\u6613\u9519\u70b9\u5904&#xff0c;\u6211\u4eec\u4f1a\u4ee5\u201c\u667a\u6167\u9526\u56ca\u201d\u7684\u5f62\u5f0f&#xff0c;\u5206\u4eab\u4e00\u4e9b\u4e66\u672c\u4e4b\u5916\u7684\u5b9e\u8df5\u7ecf\u9a8c\u3001\u5e38\u89c1\u8bef\u533a\u3001\u8c03\u8bd5\u6280\u5de7\u548c\u601d\u7ef4\u6d1e\u89c1&#xff0c;\u5e0c\u671b\u80fd\u52a9\u60a8\u5728\u63a2\u7d22\u4e4b\u8def\u4e0a\u5c11\u8d70\u5f2f\u8def&#xff0c;\u76f4\u62b5\u6838\u5fc3\u3002<\/li>\n<\/ul>\n<h4>1.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90<\/h4>\n<p>\u5728\u7ed3\u675f\u672c\u7ae0\u4e4b\u524d&#xff0c;\u8bf7\u5141\u8bb8\u6211\u5206\u4eab\u4e09\u6761\u8d2f\u7a7f\u6574\u4e2a\u5b66\u4e60\u65c5\u7a0b\u7684\u201c\u5fc3\u6cd5\u603b\u7eb2\u201d\u3002\u5b83\u4eec\u6bd4\u4efb\u4f55\u5177\u4f53\u7684\u77e5\u8bc6\u70b9\u90fd\u66f4\u4e3a\u91cd\u8981&#xff0c;\u662f\u60a8\u5728\u6df1\u5ea6\u5b66\u4e60\u4e43\u81f3\u4efb\u4f55\u590d\u6742\u9886\u57df\u53d6\u5f97\u6210\u529f\u7684\u57fa\u77f3\u3002<\/p>\n<h5>1.5.1 \u597d\u5947\u5fc3\u662f\u7b2c\u4e00\u9a71\u52a8\u529b<\/h5>\n<p>\u8bf7\u6c38\u8fdc\u4fdd\u6301\u4e00\u9897\u5b69\u7ae5\u822c\u7684\u597d\u5947\u5fc3\u3002\u5f53\u60a8\u9762\u5bf9\u4e00\u4e2a\u65b0\u7684\u7b97\u6cd5\u3001\u4e00\u4e2a\u590d\u6742\u7684\u6a21\u578b\u65f6&#xff0c;\u4e0d\u8981\u4ec5\u4ec5\u6ee1\u8db3\u4e8e\u77e5\u9053\u5b83\u201c\u5982\u4f55\u5de5\u4f5c\u201d&#xff0c;\u66f4\u8981\u8ffd\u95ee\u201c\u4e3a\u4ec0\u4e48\u8fd9\u6837\u8bbe\u8ba1&#xff1f;\u201d\u3001\u201c\u5b83\u89e3\u51b3\u4e86\u4ec0\u4e48\u95ee\u9898&#xff1f;\u201d\u3001\u201c\u6709\u6ca1\u6709\u5176\u4ed6\u53ef\u80fd\u6027&#xff1f;\u201d\u3002\u8fd9\u79cd\u5bf9\u539f\u7406\u7684\u4e0d\u65ad\u8ffd\u95ee&#xff0c;\u662f\u60a8\u4ece\u77e5\u8bc6\u7684\u6d88\u8d39\u8005\u8715\u53d8\u4e3a\u521b\u9020\u8005\u7684\u4e0d\u4e8c\u6cd5\u95e8\u3002\u597d\u5947\u5fc3\u5c06\u9a71\u52a8\u60a8\u4e0d\u65ad\u63a2\u7d22&#xff0c;\u53d1\u73b0\u77e5\u8bc6\u7684\u8fb9\u754c&#xff0c;\u5e76\u6700\u7ec8\u8d85\u8d8a\u5b83\u3002<\/p>\n<h5>1.5.2 \u5b9e\u8df5\u662f\u68c0\u9a8c\u771f\u7406\u7684\u552f\u4e00\u6807\u51c6<\/h5>\n<p>\u8ba1\u7b97\u673a\u79d1\u5b66\u9886\u57df\u6d41\u4f20\u7740\u4e00\u53e5\u540d\u8a00&#xff1a;\u201cTalk is cheap, show me the code.\u201d&#xff08;\u7a7a\u8c08\u65e0\u76ca&#xff0c;\u4ee3\u7801\u4e3a\u8bc1&#xff09;\u3002\u6df1\u5ea6\u5b66\u4e60\u662f\u4e00\u95e8\u9ad8\u5ea6\u5b9e\u8df5\u6027\u7684\u5b66\u79d1\u3002\u5b66\u4e60\u6e38\u6cf3\u6700\u597d\u7684\u65b9\u5f0f\u5c31\u662f\u8df3\u4e0b\u6c34\u3002\u8bf7\u4e0d\u8981\u53ea\u505c\u7559\u5728\u9605\u8bfb\u548c\u601d\u8003&#xff0c;\u4e00\u5b9a\u8981\u52a8\u624b\u7f16\u7a0b\u3001\u590d\u73b0\u8bba\u6587\u4e2d\u7684\u6a21\u578b\u3001\u53c2\u52a0Kaggle\u7b49\u6570\u636e\u79d1\u5b66\u7ade\u8d5b\u3001\u7528\u5b66\u5230\u7684\u77e5\u8bc6\u53bb\u89e3\u51b3\u4e00\u4e2a\u60a8\u81ea\u5df1\u611f\u5174\u8da3\u7684\u771f\u5b9e\u95ee\u9898\u3002\u5728\u5b9e\u8df5\u4e2d\u9047\u5230\u7684\u95ee\u9898\u548c\u83b7\u5f97\u7684\u6210\u529f&#xff0c;\u5c06\u662f\u60a8\u6700\u5b9d\u8d35\u7684\u8d22\u5bcc&#xff0c;\u4e5f\u662f\u60a8\u80fd\u529b\u63d0\u5347\u6700\u5feb\u7684\u9014\u5f84\u3002<\/p>\n<h5>1.5.3 \u62e5\u62b1\u5f00\u6e90&#xff0c;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a<\/h5>\n<p>\u73b0\u4ee3\u79d1\u6280\u7684\u98de\u901f\u53d1\u5c55&#xff0c;\u79bb\u4e0d\u5f00\u5f00\u6e90\u7cbe\u795e\u3002\u8bf7\u79ef\u6781\u5730\u5b66\u4e60\u3001\u4f7f\u7528\u548c\u8d21\u732e\u4e8e\u5f00\u6e90\u793e\u533a\u3002\u53bbGitHub\u4e0a\u9605\u8bfb\u4f18\u79c0\u9879\u76ee\u7684\u6e90\u7801&#xff0c;\u5b66\u4e60\u4ed6\u4eba\u7684\u4ee3\u7801\u98ce\u683c\u548c\u67b6\u6784\u601d\u60f3&#xff1b;\u5f53\u60a8\u4ece\u793e\u533a\u83b7\u76ca\u540e&#xff0c;\u4e5f\u8bf7\u5c1d\u8bd5\u901a\u8fc7\u63d0\u4ea4\u95ee\u9898\u3001\u6539\u8fdb\u6587\u6863\u751a\u81f3\u8d21\u732e\u4ee3\u7801\u7684\u65b9\u5f0f\u56de\u9988\u793e\u533a\u3002\u8fd9\u79cd\u5f00\u653e\u3001\u534f\u4f5c\u7684\u6587\u5316&#xff0c;\u6781\u5927\u5730\u7f29\u77ed\u4e86\u524d\u6cbf\u7b97\u6cd5\u4ece\u8bba\u6587\u5230\u5de5\u4e1a\u5e94\u7528\u7684\u5468\u671f&#xff0c;\u63a8\u52a8\u4e86\u6574\u4e2a\u9886\u57df\u7684\u5feb\u901f\u53d1\u5c55\u3002\u878d\u5165\u5f00\u6e90&#xff0c;\u60a8\u5c06\u4e0d\u518d\u662f\u5b64\u519b\u594b\u6218&#xff0c;\u800c\u662f\u4e0e\u5168\u7403\u9876\u5c16\u7684\u5f00\u53d1\u8005\u5171\u540c\u6210\u957f\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u4e8c\u7ae0&#xff1a;\u6570\u5b66\u4e0e\u7f16\u7a0b\u57fa\u7840 \u2014\u2014 \u5185\u529f\u5fc3\u6cd5<\/h3>\n<ul>\n<li>2.1\u00a0\u7ebf\u6027\u4ee3\u6570&#xff1a;\u5411\u91cf\u3001\u77e9\u9635\u3001\u5f20\u91cf\u53ca\u5176\u8fd0\u7b97&#xff08;\u4e0d\u4ec5\u662f\u8ba1\u7b97&#xff0c;\u66f4\u662f\u7a7a\u95f4\u7684\u53d8\u6362&#xff09;\u3002<\/li>\n<li>2.2\u00a0\u5fae\u79ef\u5206&#xff1a;\u5bfc\u6570\u3001\u504f\u5bfc\u6570\u3001\u94fe\u5f0f\u6cd5\u5219\u4e0e\u68af\u5ea6&#xff08;\u7406\u89e3\u53d8\u5316\u4e0e\u4f18\u5316\u7684\u8bed\u8a00&#xff09;\u3002<\/li>\n<li>2.3\u00a0\u6982\u7387\u8bba\u4e0e\u4fe1\u606f\u8bba&#xff1a;\u6982\u7387\u5206\u5e03\u3001\u671f\u671b\u3001\u71b5\u4e0e\u4ea4\u53c9\u71b5&#xff08;\u8861\u91cf\u4e0d\u786e\u5b9a\u6027\u4e0e\u4fe1\u606f&#xff09;\u3002<\/li>\n<li>2.4\u00a0NumPy&#xff1a;\u7cbe\u901a\u591a\u7ef4\u6570\u7ec4\u64cd\u4f5c&#xff0c;\u4e3a\u6570\u636e\u5904\u7406\u63d0\u901f\u3002<\/li>\n<li>2.5\u00a0Pandas&#xff1a;\u7ed3\u6784\u5316\u6570\u636e\u7684\u63a2\u67e5\u3001\u6e05\u6d17\u4e0e\u9884\u5904\u7406\u3002<\/li>\n<li>2.6 Matplotlib &amp; Seaborn&#xff1a;\u6570\u636e\u7684\u53ef\u89c6\u5316&#xff0c;\u8ba9\u6d1e\u5bdf\u76f4\u89c2\u5448\u73b0\u3002\u00a0<\/li>\n<\/ul>\n<p>\u683c\u7269\u81f4\u77e5 \u2014\u2014 \u6df1\u5ea6\u5b66\u4e60\u80cc\u540e\u7684\u6570\u5b66\u4e0e\u4ee3\u7801\u4e4b\u9053<\/p>\n<p>\u6b22\u8fce\u6765\u5230\u672c\u4e66\u6700\u5177\u5960\u57fa\u610f\u4e49\u7684\u7ae0\u8282\u3002\u5728\u8bfb\u8005\u4eec\u5373\u5c06\u542f\u7a0b&#xff0c;\u63a2\u7d22\u6df1\u5ea6\u5b66\u4e60\u8fd9\u4e2a\u7531\u7b97\u6cd5\u3001\u6a21\u578b\u4e0e\u6570\u636e\u6784\u6210\u7684\u5947\u5999\u65b0\u4e16\u754c\u4e4b\u524d&#xff0c;\u6211\u4eec\u5fc5\u987b\u9996\u5148\u4fee\u70bc\u4e00\u5957\u201c\u5185\u529f\u5fc3\u6cd5\u201d\u3002\u8fd9\u5957\u5fc3\u6cd5&#xff0c;\u7531\u4e24\u90e8\u5206\u7ec4\u6210&#xff1a;\u4e00\u90e8\u5206\u662f\u6d1e\u5bdf\u4e07\u7269\u89c4\u5f8b\u7684\u6570\u5b66\u4e4b\u9053&#xff0c;\u53e6\u4e00\u90e8\u5206\u5219\u662f\u5c06\u601d\u60f3\u5316\u4e3a\u73b0\u5b9e\u7684\u7f16\u7a0b\u4e4b\u672f\u3002<\/p>\n<p>\u6216\u8bb8&#xff0c;\u90e8\u5206\u8bfb\u8005\u4f1a\u5bf9\u6807\u9898\u4e2d\u7684\u201c\u6570\u5b66\u201d\u4e8c\u5b57\u5fc3\u751f\u754f\u60e7&#xff0c;\u8054\u60f3\u5230\u67af\u71e5\u7684\u516c\u5f0f\u63a8\u5bfc\u4e0e\u7e41\u590d\u7684\u8ba1\u7b97\u3002\u8bf7\u653e\u5fc3&#xff0c;\u8fd9\u6070\u6070\u662f\u672c\u7ae0\u81f4\u529b\u4e8e\u7834\u9664\u7684\u8ff7\u601d\u3002\u6211\u4eec\u4e0d\u4f1a\u5c06\u91cd\u70b9\u653e\u5728\u201c\u5982\u4f55\u8ba1\u7b97\u201d\u4e0a&#xff0c;\u800c\u662f\u805a\u7126\u4e8e\u201c\u4e3a\u4f55\u5982\u6b64\u201d\u2014\u2014\u6211\u4eec\u5c06\u4e00\u540c\u5efa\u7acb\u8d77\u5bf9\u5173\u952e\u6570\u5b66\u6982\u5ff5\u76f4\u89c2\u7684\u51e0\u4f55\u60f3\u8c61&#xff0c;\u7406\u89e3\u5176\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u7684\u7269\u7406\u610f\u4e49&#xff0c;\u751a\u81f3\u54c1\u5473\u5176\u80cc\u540e\u8574\u542b\u7684\u54f2\u5b66\u601d\u8fa8\u3002\u60a8\u4f1a\u53d1\u73b0&#xff0c;\u7ebf\u6027\u4ee3\u6570\u662f\u63cf\u8ff0\u7ed3\u6784\u4e0e\u53d8\u6362\u7684\u8bed\u8a00&#xff0c;\u5fae\u79ef\u5206\u662f\u8c31\u5199\u53d8\u5316\u4e0e\u4f18\u5316\u7684\u4e50\u7ae0&#xff0c;\u800c\u6982\u7387\u8bba\u5219\u662f\u8861\u91cf\u4e0d\u786e\u5b9a\u6027\u7684\u6807\u5c3a\u3002\u5b83\u4eec\u662f\u7406\u89e3\u7b97\u6cd5\u672c\u8d28\u7684\u201c\u4e16\u754c\u89c2\u201d\u3002<\/p>\n<p>\u4e0e\u6b64\u76f8\u8f85\u76f8\u6210\u7684&#xff0c;\u662f\u7f16\u7a0b\u7684\u201c\u65b9\u6cd5\u8bba\u201d\u3002\u5982\u679c\u8bf4\u6570\u5b66\u601d\u60f3\u662f\u8bbe\u8ba1\u5e08\u7684\u84dd\u56fe&#xff0c;\u90a3\u4e48Python\u53ca\u5176\u5f3a\u5927\u7684\u79d1\u5b66\u8ba1\u7b97\u5e93&#xff0c;\u5c31\u662f\u6211\u4eec\u624b\u4e2d\u5de7\u593a\u5929\u5de5\u7684\u5de5\u5177\u3002\u672c\u7ae0\u5c06\u5f15\u5bfc\u8bfb\u8005\u7cbe\u901aNumPy\u3001Pandas\u3001Matplotlib\u8fd9\u4e09\u67c4\u201c\u795e\u5175\u5229\u5668\u201d&#xff0c;\u5b83\u4eec\u662f\u6570\u636e\u5904\u7406\u3001\u5206\u6790\u4e0e\u53ef\u89c6\u5316\u7684\u57fa\u77f3&#xff0c;\u662f\u9a71\u52a8\u6df1\u5ea6\u5b66\u4e60\u8fd9\u9a7e\u8c6a\u534e\u9a6c\u8f66\u7684\u53cc\u8f6e\u4e2d&#xff0c;\u575a\u5b9e\u6709\u529b\u7684\u90a3\u4e00\u73af\u3002<\/p>\n<p>\u672c\u7ae0\u7684\u4fee\u884c\u8def\u5f84\u6e05\u6670\u800c\u575a\u5b9a&#xff1a;\u5148\u901a\u201c\u7406\u201d&#xff0c;\u518d\u5229\u201c\u5668\u201d\u3002\u6211\u4eec\u5c06\u9996\u5148\u6f2b\u6b65\u4e8e\u6570\u5b66\u7684\u6bbf\u5802&#xff0c;\u800c\u540e\u5728\u7f16\u7a0b\u7684\u5de5\u574a\u91cc\u6325\u6d12\u5b9e\u8df5\u3002\u552f\u6709\u5185\u5916\u517c\u4fee&#xff0c;\u7406\u672f\u5408\u4e00&#xff0c;\u65b9\u80fd\u5728\u540e\u7eed\u7684\u6df1\u5ea6\u5b66\u4e60\u5b9e\u6218\u4e2d&#xff0c;\u505a\u5230\u77e5\u5176\u7136&#xff0c;\u66f4\u77e5\u5176\u6240\u4ee5\u7136&#xff0c;\u4ece\u800c\u6e38\u5203\u6709\u4f59&#xff0c;\u76f4\u62b5\u7cbe\u901a\u3002<\/p>\n<p>\u73b0\u5728&#xff0c;\u8bf7\u629b\u5f00\u7591\u8651&#xff0c;\u4fdd\u6301\u4e00\u9897\u597d\u5947\u4e0e\u5f00\u653e\u7684\u5fc3&#xff0c;\u4e0e\u6211\u4eec\u4e00\u540c\u5f00\u59cb\u8fd9\u573a\u201c\u683c\u7269\u81f4\u77e5\u201d\u7684\u65c5\u7a0b\u3002<\/p>\n<h4>2.1 \u7ebf\u6027\u4ee3\u6570&#xff1a;\u63cf\u7ed8\u4e16\u754c\u7684\u7ed3\u6784\u4e0e\u53d8\u6362<\/h4>\n<p>\u7ebf\u6027\u4ee3\u6570&#xff0c;\u662f\u73b0\u4ee3\u6570\u5b66\u7684\u652f\u67f1\u4e4b\u4e00&#xff0c;\u66f4\u662f\u6df1\u5ea6\u5b66\u4e60\u4e0d\u53ef\u6216\u7f3a\u7684\u8bed\u8a00\u3002\u5b83\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u4e00\u5957\u5f3a\u5927\u7684\u5de5\u5177\u548c\u89c6\u89d2&#xff0c;\u6765\u7406\u89e3\u548c\u64cd\u4f5c\u9ad8\u7ef4\u6570\u636e\u3002\u672c\u8d28\u4e0a&#xff0c;\u7ebf\u6027\u4ee3\u6570\u7814\u7a76\u7684\u662f\u5411\u91cf\u3001\u5411\u91cf\u7a7a\u95f4\u4ee5\u53ca\u7ebf\u6027\u53d8\u6362\u3002\u5728\u672c\u8282\u4e2d&#xff0c;\u6211\u4eec\u5c06\u6452\u5f03\u7e41\u6742\u7684\u8bc1\u660e&#xff0c;\u4e13\u6ce8\u4e8e\u5efa\u7acb\u8fd9\u4e9b\u6982\u5ff5\u7684\u76f4\u89c2\u7406\u89e3&#xff0c;\u770b\u5b83\u4eec\u5982\u4f55\u63cf\u7ed8\u51fa\u4e16\u754c\u7684\u7ed3\u6784&#xff0c;\u5e76\u6267\u884c\u4f18\u96c5\u7684\u53d8\u6362\u3002<\/p>\n<h5>2.1.1 \u5411\u91cf&#xff08;Vector&#xff09;&#xff1a;\u5b9a\u4e49\u7a7a\u95f4\u4e2d\u7684\u201c\u5b58\u5728\u201d<\/h5>\n<p>\u5411\u91cf\u7684\u51e0\u4f55\u4e0e\u7269\u7406\u610f\u4e49<\/p>\n<ul>\n<li>\n<p>\u4f55\u4e3a\u5411\u91cf \u5728\u6700\u6d45\u7684\u5c42\u9762&#xff0c;\u4e00\u4e2a\u5411\u91cf\u53ef\u4ee5\u770b\u4f5c\u662f\u5217\u8868\u4e2d\u7684\u4e00\u4e32\u6570\u5b57&#xff0c;\u4f8b\u5982 v &#061; [3, 4]\u3002\u7136\u800c&#xff0c;\u5b83\u7684\u5185\u6db5\u8fdc\u6bd4\u8fd9\u4e30\u5bcc\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u8bed\u5883\u4e2d&#xff0c;\u4e00\u4e2a\u5411\u91cf\u662f\u5bf9\u73b0\u5b9e\u4e16\u754c\u67d0\u4e2a\u5bf9\u8c61\u5173\u952e\u7279\u5f81\u7684\u6570\u5b57\u5316\u62bd\u8c61\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u5b83\u5728\u51e0\u4f55\u4e0a\u4ee3\u8868\u4e86\u591a\u7ef4\u7a7a\u95f4\u4e2d\u7684\u4e00\u4e2a\u70b9&#xff0c;\u6216\u4e00\u4e2a\u4ece\u539f\u70b9\u51fa\u53d1\u3001\u5e26\u6709\u65b9\u5411\u548c\u957f\u5ea6\u7684\u7bad\u5934\u3002<\/p>\n<p>\u4e00\u4e2a\u76f4\u89c2\u7684\u4f8b\u5b50&#xff1a; \u5047\u8bbe\u6211\u4eec\u60f3\u63cf\u8ff0\u4e00\u4e2a\u4eba\u7684\u5065\u5eb7\u72b6\u51b5&#xff0c;\u6211\u4eec\u9009\u62e9\u4e86\u4e24\u4e2a\u7279\u5f81&#xff1a;\u8eab\u9ad8&#xff08;\u5398\u7c73&#xff09;\u548c\u4f53\u91cd&#xff08;\u516c\u65a4&#xff09;\u3002\u5982\u679c\u67d0\u4eba\u7684\u8eab\u9ad8\u662f175cm&#xff0c;\u4f53\u91cd\u662f70kg&#xff0c;\u6211\u4eec\u5c31\u53ef\u4ee5\u7528\u4e00\u4e2a\u4e8c\u7ef4\u5411\u91cf p &#061; [175, 70] \u6765\u8868\u793a\u4ed6\u3002\u8fd9\u4e2a\u5411\u91cf&#xff0c;\u5c31\u662f\u201c\u5065\u5eb7\u7279\u5f81\u7a7a\u95f4\u201d\u8fd9\u4e2a\u4e8c\u7ef4\u5e73\u9762\u4e0a\u7684\u4e00\u4e2a\u786e\u5b9a\u7684\u70b9\u3002<\/p>\n<p>\u4ece\u6587\u672c\u3001\u56fe\u50cf\u5230\u58f0\u97f3&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u7b2c\u4e00\u6b65&#xff0c;\u5f80\u5f80\u5c31\u662f\u5c06\u8fd9\u4e9b\u590d\u6742\u7684\u3001\u975e\u7ed3\u6784\u5316\u7684\u4fe1\u606f&#xff0c;\u8f6c\u5316\u4e3a\u4e00\u4e2a\u4e2a\u9ad8\u7ef4\u7a7a\u95f4\u4e2d\u7684\u5411\u91cf&#xff08;\u8fd9\u4e2a\u8fc7\u7a0b\u79f0\u4e3a\u201c\u5d4c\u5165\u201d\u6216\u201c\u7279\u5f81\u63d0\u53d6\u201d&#xff09;\u3002\u56e0\u6b64&#xff0c;\u7406\u89e3\u5411\u91cf&#xff0c;\u5c31\u662f\u7406\u89e3\u6df1\u5ea6\u5b66\u4e60\u5982\u4f55\u201c\u770b\u5f85\u201d\u4e16\u754c\u4e07\u7269\u3002<\/p>\n<\/li>\n<li>\n<p>\u5411\u91cf\u7684\u6a21\u4e0e\u65b9\u5411 \u6bcf\u4e00\u4e2a\u5411\u91cf\u90fd\u5305\u542b\u4e24\u4e2a\u6838\u5fc3\u4fe1\u606f&#xff1a;<\/p>\n<ul>\n<li>\u6a21&#xff08;Magnitude\/Norm&#xff09;&#xff1a;\u4ee3\u8868\u5411\u91cf\u7684\u201c\u957f\u5ea6\u201d\u3002\u5728\u51e0\u4f55\u4e0a&#xff0c;\u5b83\u662f\u4ece\u539f\u70b9\u5230\u5411\u91cf\u6240\u6307\u5411\u7684\u70b9\u7684\u8ddd\u79bb\u3002\u5728\u7269\u7406\u4e0a&#xff0c;\u5b83\u901a\u5e38\u4ee3\u8868\u4e86\u7279\u5f81\u7684\u5f3a\u5ea6\u6216\u4e8b\u7269\u7684\u5927\u5c0f\u3002\u4f8b\u5982&#xff0c;\u5728\u63a8\u8350\u7cfb\u7edf\u4e2d&#xff0c;\u4e00\u4e2a\u7528\u6237\u5411\u91cf\u7684\u6a21\u53ef\u80fd\u4ee3\u8868\u5176\u5174\u8da3\u7684\u5e7f\u6cdb\u7a0b\u5ea6\u3002\u5411\u91cf\u00a0v &#061; [3, 4]\u00a0\u7684\u6a21&#xff08;L2\u8303\u6570&#xff09;\u4e3a\u00a0\u221a(3\u00b2 &#043; 4\u00b2) &#061; 5\u3002<\/li>\n<li>\u65b9\u5411&#xff08;Direction&#xff09;&#xff1a;\u4ee3\u8868\u5411\u91cf\u5728\u7a7a\u95f4\u4e2d\u6240\u6307\u7684\u671d\u5411\u3002\u5b83\u901a\u5e38\u4ee3\u8868\u4e86\u7279\u5f81\u7684\u6027\u8d28\u6216\u4e8b\u7269\u7684\u7c7b\u522b\u5f52\u5c5e\u3002\u4e24\u4e2a\u65b9\u5411\u76f8\u8fd1\u7684\u5411\u91cf&#xff0c;\u901a\u5e38\u610f\u5473\u7740\u5b83\u4eec\u6240\u4ee3\u8868\u7684\u5bf9\u8c61\u5728\u672c\u8d28\u4e0a\u662f\u76f8\u4f3c\u7684\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u5411\u91cf\u7684\u8fd0\u7b97\u4e0e\u5e94\u7528<\/p>\n<ul>\n<li>\n<p>\u70b9\u79ef&#xff08;Dot Product&#xff09; \u5411\u91cf\u70b9\u79ef\u662f\u7ebf\u6027\u4ee3\u6570\u4e2d\u4e00\u4e2a\u6781\u5176\u6709\u7528\u7684\u8fd0\u7b97\u3002\u5bf9\u4e8e\u4e24\u4e2a\u5411\u91cf a &#061; [a\u2081, a\u2082] \u548c b &#061; [b\u2081, b\u2082]&#xff0c;\u5b83\u4eec\u7684\u70b9\u79ef\u662f a \u00b7 b &#061; a\u2081b\u2081 &#043; a\u2082b\u2082\u3002\u8fd9\u4e2a\u7b80\u5355\u7684\u8ba1\u7b97\u80cc\u540e&#xff0c;\u8574\u542b\u7740\u6df1\u523b\u7684\u51e0\u4f55\u610f\u4e49&#xff1a; a \u00b7 b &#061; |a| * |b| * cos(\u03b8) \u5176\u4e2d |a| \u548c |b| \u662f\u4e24\u4e2a\u5411\u91cf\u7684\u6a21&#xff0c;\u03b8 \u662f\u5b83\u4eec\u7684\u5939\u89d2\u3002\u8fd9\u4e2a\u516c\u5f0f\u544a\u8bc9\u6211\u4eec&#xff0c;\u70b9\u79ef\u7684\u7ed3\u679c\u878d\u5408\u4e86\u4e24\u4e2a\u5411\u91cf\u7684\u957f\u5ea6\u548c\u5b83\u4eec\u65b9\u5411\u4e0a\u7684\u4e00\u81f4\u6027\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b9e\u6218\u5e94\u7528&#xff1a;\u4f59\u5f26\u76f8\u4f3c\u5ea6 \u5728\u5f88\u591a\u5e94\u7528\u4e2d&#xff0c;\u6211\u4eec\u53ea\u5173\u5fc3\u4e24\u4e2a\u5411\u91cf\u7684\u65b9\u5411\u662f\u5426\u4e00\u81f4&#xff0c;\u800c\u4e0d\u60f3\u53d7\u5230\u5b83\u4eec\u957f\u5ea6\u7684\u5f71\u54cd\u3002\u4f8b\u5982&#xff0c;\u5728\u5224\u65ad\u4e24\u7bc7\u6587\u7ae0\u7684\u4e3b\u9898\u662f\u5426\u76f8\u4f3c\u65f6&#xff0c;\u4e00\u7bc7\u957f\u6587\u7ae0\u548c\u4e00\u7bc7\u77ed\u6587\u7ae0\u53ea\u8981\u4e3b\u9898\u76f8\u540c&#xff0c;\u5b83\u4eec\u7684\u5411\u91cf\u65b9\u5411\u5c31\u5e94\u8be5\u63a5\u8fd1&#xff0c;\u6211\u4eec\u4e0d\u5e0c\u671b\u56e0\u4e3a\u6587\u7ae0\u957f\u5ea6&#xff08;\u5411\u91cf\u7684\u6a21&#xff09;\u4e0d\u540c\u800c\u5f71\u54cd\u5224\u65ad\u3002<\/p>\n<p>\u4e3a\u6b64&#xff0c;\u6211\u4eec\u4ece\u70b9\u79ef\u516c\u5f0f\u4e2d\u63a8\u5bfc\u51fa\u4f59\u5f26\u76f8\u4f3c\u5ea6&#xff08;Cosine Similarity&#xff09;&#xff1a; Similarity &#061; cos(\u03b8) &#061; (a \u00b7 b) \/ (|a| * |b|) \u4f59\u5f26\u76f8\u4f3c\u5ea6\u7684\u503c\u57df\u5728 [-1, 1] \u4e4b\u95f4&#xff1a;<\/p>\n<p>\u8fd9\u4e2a\u7b80\u5355\u7684\u6307\u6807&#xff0c;\u662f\u73b0\u4ee3\u63a8\u8350\u7cfb\u7edf\u3001\u641c\u7d22\u5f15\u64ce\u548c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e2d\u8861\u91cf\u201c\u76f8\u4f3c\u6027\u201d\u7684\u57fa\u77f3\u3002<\/p>\n<ul>\n<li>\u5f53\u503c\u4e3a\u00a01\u00a0\u65f6&#xff0c;\u8868\u793a\u4e24\u4e2a\u5411\u91cf\u65b9\u5411\u5b8c\u5168\u76f8\u540c&#xff0c;\u4ee3\u8868\u7684\u5bf9\u8c61\u6700\u76f8\u4f3c\u3002<\/li>\n<li>\u5f53\u503c\u4e3a\u00a00\u00a0\u65f6&#xff0c;\u8868\u793a\u4e24\u4e2a\u5411\u91cf\u65b9\u5411\u6b63\u4ea4&#xff08;\u5782\u76f4&#xff09;&#xff0c;\u4ee3\u8868\u7684\u5bf9\u8c61\u4e0d\u76f8\u5173\u3002<\/li>\n<li>\u5f53\u503c\u4e3a\u00a0-1\u00a0\u65f6&#xff0c;\u8868\u793a\u4e24\u4e2a\u5411\u91cf\u65b9\u5411\u5b8c\u5168\u76f8\u53cd\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>2.1.2 \u77e9\u9635&#xff08;Matrix&#xff09;&#xff1a;\u7ec4\u7ec7\u201c\u5173\u7cfb\u201d\u4e0e\u6267\u884c\u201c\u53d8\u6362\u201d<\/h5>\n<p>\u5982\u679c\u8bf4\u5411\u91cf\u662f\u5355\u4e2a\u201c\u5b58\u5728\u201d&#xff0c;\u90a3\u4e48\u77e9\u9635\u5c31\u662f\u8fd9\u4e9b\u201c\u5b58\u5728\u201d\u7684\u6709\u5e8f\u96c6\u5408&#xff0c;\u662f\u63cf\u7ed8\u7fa4\u4f53\u5173\u7cfb\u4e0e\u6267\u884c\u7cfb\u7edf\u6027\u53d8\u5316\u7684\u5b8f\u5927\u753b\u5377\u3002<\/p>\n<p>\u77e9\u9635\u7684\u672c\u8d28<\/p>\n<ul>\n<li>\u4f55\u4e3a\u77e9\u9635\u00a0\u4e00\u4e2a\u77e9\u9635&#xff0c;\u5c31\u662f\u4e00\u4e2a\u4e8c\u7ef4\u7684\u6570\u5b57\u9635\u5217\u3002\u5b83\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u4e00\u7ec4\u884c\u5411\u91cf\u6216\u4e00\u7ec4\u5217\u5411\u91cf\u7684\u6709\u5e8f\u96c6\u5408\u3002\u77e9\u9635\u662f\u4e8c\u7ef4\u5173\u7cfb\u7684\u5929\u7136\u8f7d\u4f53\u3002\n<ul>\n<li>\u4e00\u5f20\u7070\u5ea6\u56fe\u7247&#xff0c;\u5176\u6bcf\u4e2a\u50cf\u7d20\u7684\u7070\u5ea6\u503c\u53ef\u4ee5\u6784\u6210\u4e00\u4e2a\u77e9\u9635\u3002<\/li>\n<li>\u4e00\u4e2a\u5305\u542b\u591a\u4e2a\u5b66\u751f\u3001\u591a\u95e8\u8bfe\u7a0b\u6210\u7ee9\u7684\u6570\u636e\u96c6&#xff0c;\u672c\u8eab\u5c31\u662f\u4e00\u4e2a\u201c\u5b66\u751f-\u8bfe\u7a0b\u201d\u77e9\u9635\u3002<\/li>\n<li>\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d&#xff0c;\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\u5c42\u7684\u6743\u91cd&#xff0c;\u901a\u5e38\u5c31\u5b58\u50a8\u5728\u4e00\u4e2a\u77e9\u9635\u4e2d\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u77e9\u9635\u4e58\u6cd5\u7684\u6838\u5fc3\u2014\u2014\u7a7a\u95f4\u53d8\u6362<\/p>\n<ul>\n<li>\n<p>\u77e9\u9635\u4e0e\u5411\u91cf\u76f8\u4e58 \u8fd9\u662f\u7ebf\u6027\u4ee3\u6570\u4e2d\u6700\u6838\u5fc3\u3001\u6700\u795e\u5947\u7684\u64cd\u4f5c\u3002\u5f53\u4e00\u4e2a\u77e9\u9635 M \u4e0e\u4e00\u4e2a\u5411\u91cf v \u76f8\u4e58\u5f97\u5230\u4e00\u4e2a\u65b0\u7684\u5411\u91cf v&#039; \u65f6&#xff08;v&#039; &#061; Mv&#xff09;&#xff0c;\u8fd9\u7edd\u4e0d\u4ec5\u4ec5\u662f\u4e00\u6b21\u7b97\u672f\u8ba1\u7b97\u3002\u4ece\u51e0\u4f55\u7684\u89d2\u5ea6\u770b&#xff0c;\u8fd9\u662f\u4e00\u6b21\u7531\u77e9\u9635 M \u6240\u5b9a\u4e49\u7684\u5bf9\u6574\u4e2a\u7a7a\u95f4\u7684\u7ebf\u6027\u53d8\u6362\u3002<\/p>\n<p>\u4e00\u4e2a\u751f\u52a8\u7684\u60f3\u8c61&#xff1a; \u60f3\u8c61\u4e00\u5f20\u753b\u7740\u8bb8\u591a\u70b9\u7684\u3001\u65e0\u9650\u5927\u7684\u6a61\u80f6\u819c&#xff08;\u4ee3\u8868\u6211\u4eec\u7684\u4e8c\u7ef4\u7a7a\u95f4&#xff09;\u3002\u4e00\u4e2a 2&#215;2 \u7684\u77e9\u9635&#xff0c;\u5c31\u662f\u4e00\u5957\u5bf9\u8fd9\u5f20\u819c\u8fdb\u884c\u64cd\u4f5c\u7684\u6307\u4ee4\u96c6\u3002\u5f53\u8fd9\u4e2a\u77e9\u9635\u4f5c\u7528\u4e8e\u7a7a\u95f4\u4e2d\u7684\u6bcf\u4e00\u4e2a\u70b9&#xff08;\u5411\u91cf&#xff09;\u65f6&#xff0c;\u6574\u5f20\u6a61\u80f6\u819c\u53ef\u80fd\u4f1a\u88ab\u65cb\u8f6c\u4e00\u4e2a\u89d2\u5ea6&#xff0c;\u6216\u8005\u5728\u67d0\u4e2a\u65b9\u5411\u4e0a\u88ab\u62c9\u4f38\/\u538b\u7f29&#xff08;\u7f29\u653e&#xff09;&#xff0c;\u4e5f\u53ef\u80fd\u88ab\u626d\u66f2&#xff08;\u526a\u5207&#xff09;\u3002<\/p>\n<p>\u539f\u59cb\u7a7a\u95f4\u4e2d\u6240\u6709\u5e73\u884c\u4e14\u7b49\u8ddd\u7684\u7f51\u683c\u7ebf&#xff0c;\u5728\u53d8\u6362\u540e\u4f9d\u7136\u4fdd\u6301\u5e73\u884c\u4e14\u7b49\u8ddd&#xff0c;\u8fd9\u5c31\u662f\u201c\u7ebf\u6027\u201d\u53d8\u6362\u7684\u542b\u4e49\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u5b9e\u6218\u54f2\u601d&#xff1a;\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u9b54\u6cd5\u201d \u73b0\u5728&#xff0c;\u6211\u4eec\u53ef\u4ee5\u63ed\u793a\u795e\u7ecf\u7f51\u7edc\u201c\u9b54\u6cd5\u201d\u7684\u7b2c\u4e00\u5c42\u5965\u79d8\u4e86\u3002\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\u7531\u8bb8\u591a\u5c42\u5806\u53e0\u800c\u6210&#xff0c;\u800c\u6bcf\u4e00\u5c42\u7684\u6838\u5fc3\u8ba1\u7b97&#xff0c;\u672c\u8d28\u4e0a\u5c31\u662f\u4e00\u6b21\u7531\u8be5\u5c42\u7684\u6743\u91cd\u77e9\u9635&#xff08;W&#xff09;\u6240\u4e3b\u5bfc\u7684\u7a7a\u95f4\u53d8\u6362&#xff0c;\u518d\u53e0\u52a0\u4e0a\u4e00\u4e2a\u504f\u7f6e\u5411\u91cf&#xff08;b&#xff09;\u7684\u5e73\u79fb&#xff0c;\u6700\u540e\u901a\u8fc7\u4e00\u4e2a\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570&#xff08;\u4e0b\u4e00\u7ae0\u8be6\u8ff0&#xff09;\u8fdb\u884c\u201c\u5f2f\u66f2\u201d\u3002<\/p>\n<p>output &#061; activation(W * input &#043; b)<\/p>\n<p>\u7f51\u7edc\u7684\u201c\u5b66\u4e60\u201d\u6216\u201c\u8bad\u7ec3\u201d\u8fc7\u7a0b&#xff0c;\u6b63\u662f\u5728\u4e0d\u65ad\u5730\u3001\u7cbe\u5de7\u5730\u8c03\u6574\u8fd9\u4e9b\u6743\u91cd\u77e9\u9635 W&#xff0c;\u8bd5\u56fe\u627e\u5230\u4e00\u7cfb\u5217\u6700\u4f73\u7684\u7a7a\u95f4\u53d8\u6362\u7ec4\u5408\u3002\u5176\u6700\u7ec8\u76ee\u7684&#xff0c;\u662f\u5e0c\u671b\u5c06\u539f\u59cb\u7279\u5f81\u7a7a\u95f4\u4e2d\u90a3\u4e9b\u6df7\u6742\u5728\u4e00\u8d77\u3001\u96be\u4ee5\u533a\u5206\u7684\u6570\u636e\u70b9&#xff08;\u4f8b\u5982&#xff0c;\u4ee3\u8868\u201c\u732b\u201d\u7684\u56fe\u7247\u5411\u91cf\u548c\u4ee3\u8868\u201c\u72d7\u201d\u7684\u56fe\u7247\u5411\u91cf&#xff09;&#xff0c;\u901a\u8fc7\u5c42\u5c42\u53d8\u6362&#xff0c;\u626d\u8f6c\u5230\u4e00\u4e2a\u65b0\u7684\u3001\u9ad8\u7ef4\u7684\u7a7a\u95f4\u4e2d\u53bb\u3002\u5728\u8fd9\u4e2a\u65b0\u7684\u7a7a\u95f4\u91cc&#xff0c;\u539f\u672c\u7ea0\u7f20\u7684\u6570\u636e\u88ab\u201c\u89e3\u5f00\u201d&#xff0c;\u4e0d\u540c\u7c7b\u522b\u7684\u6570\u636e\u88ab\u6e05\u6670\u5730\u63a8\u5411\u4e86\u4e0d\u540c\u7684\u533a\u57df&#xff0c;\u4ece\u800c\u53ef\u4ee5\u88ab\u4e00\u4e2a\u7b80\u5355\u7684\u5e73\u9762\u8f7b\u6613\u5730\u5206\u5f00\u3002<\/p>\n<p>\u56e0\u6b64&#xff0c;\u7406\u89e3\u4e86\u77e9\u9635\u662f\u7a7a\u95f4\u53d8\u6362\u7684\u6267\u884c\u8005&#xff0c;\u4e5f\u5c31\u7406\u89e3\u4e86\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u7279\u5f81\u5b66\u4e60\u4e0e\u5206\u7c7b\u7684\u6838\u5fc3\u673a\u5236\u3002<\/p>\n<h5>2.1.3 \u5f20\u91cf&#xff08;Tensor&#xff09;&#xff1a;\u9ad8\u7ef4\u6570\u636e\u7684\u6807\u51c6\u201c\u5bb9\u5668\u201d<\/h5>\n<p>\u5728\u638c\u63e1\u4e86\u5411\u91cf\u548c\u77e9\u9635\u4e4b\u540e&#xff0c;\u6211\u4eec\u81ea\u7136\u5730\u8fc8\u5411\u4e00\u4e2a\u66f4\u5e7f\u4e49\u3001\u66f4\u5f3a\u5927\u7684\u6982\u5ff5\u2014\u2014\u5f20\u91cf\u3002<\/p>\n<p>\u4ece\u5411\u91cf\u3001\u77e9\u9635\u5230\u5f20\u91cf<\/p>\n<ul>\n<li>\u7ef4\u5ea6\u7684\u6269\u5c55\u00a0\u5f20\u91cf\u662f\u591a\u7ef4\u6570\u7ec4\u7684\u7edf\u79f0&#xff0c;\u662f\u5411\u91cf\u548c\u77e9\u9635\u5728\u66f4\u9ad8\u7ef4\u5ea6\u4e0a\u7684\u63a8\u5e7f\u3002\u6211\u4eec\u53ef\u4ee5\u8fd9\u6837\u7406\u89e3\u5b83\u4eec\u7684\u5173\u7cfb&#xff1a;\n<ul>\n<li>0\u9636\u5f20\u91cf&#xff1a;\u4e00\u4e2a\u6807\u91cf&#xff08;Scalar&#xff09;&#xff0c;\u5373\u4e00\u4e2a\u5355\u72ec\u7684\u6570\u5b57&#xff08;\u5982\u00a05&#xff09;\u3002\u5b83\u6ca1\u6709\u65b9\u5411&#xff0c;\u53ea\u6709\u4e00\u4e2a\u5927\u5c0f\u3002<\/li>\n<li>1\u9636\u5f20\u91cf&#xff1a;\u4e00\u4e2a\u5411\u91cf&#xff08;Vector&#xff09;&#xff0c;\u5373\u4e00\u7ef4\u6570\u7ec4&#xff08;\u5982\u00a0[1, 2, 3]&#xff09;\u3002<\/li>\n<li>2\u9636\u5f20\u91cf&#xff1a;\u4e00\u4e2a\u77e9\u9635&#xff08;Matrix&#xff09;&#xff0c;\u5373\u4e8c\u7ef4\u6570\u7ec4&#xff08;\u5982\u00a0[[1, 2], [3, 4]]&#xff09;\u3002<\/li>\n<li>3\u9636\u53ca\u4ee5\u4e0a\u5f20\u91cf&#xff1a;\u6211\u4eec\u901a\u5e38\u76f4\u63a5\u79f0\u4e4b\u4e3a\u5f20\u91cf&#xff08;Tensor&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u5f20\u91cf\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d\u7684\u89d2\u8272<\/p>\n<p>\u5f20\u91cf&#xff0c;\u6b63\u662f\u627f\u8f7d\u8fd9\u4e9b\u590d\u6742\u9ad8\u7ef4\u6570\u636e\u7684\u6807\u51c6\u6570\u636e\u7ed3\u6784\u3002\u6240\u6709\u4e3b\u6d41\u7684\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6&#xff0c;\u5982TensorFlow\u548cPyTorch&#xff0c;\u5176\u6838\u5fc3\u6570\u636e\u5355\u5143\u90fd\u662f\u5f20\u91cf\u3002\u56e0\u6b64&#xff0c;\u719f\u7ec3\u5730\u7406\u89e3\u548c\u64cd\u4f5c\u5f20\u91cf&#xff0c;\u662f\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60\u7f16\u7a0b\u7684\u5fc5\u5907\u6280\u80fd\u3002<\/p>\n<p>\u4e3a\u4f55\u9700\u8981\u5f20\u91cf\u00a0\u56e0\u4e3a\u6df1\u5ea6\u5b66\u4e60\u6240\u5904\u7406\u7684\u73b0\u5b9e\u4e16\u754c\u6570\u636e&#xff0c;\u5176\u7ed3\u6784\u5929\u7136\u5c31\u662f\u9ad8\u7ef4\u7684\u3002<\/p>\n<ul>\n<li>\u4e00\u5f20\u5f69\u8272\u56fe\u7247&#xff1a;\u5b83\u6709\u9ad8\u5ea6\u3001\u5bbd\u5ea6\u548c\u989c\u8272\u901a\u9053&#xff08;\u5982RGB\u4e09\u901a\u9053&#xff09;\u4e09\u4e2a\u7ef4\u5ea6&#xff0c;\u56e0\u6b64\u5b83\u662f\u4e00\u4e2a3\u9636\u5f20\u91cf&#xff0c;\u5f62\u72b6&#xff08;Shape&#xff09;\u53ef\u4ee5\u8868\u793a\u4e3a\u00a0(\u9ad8\u5ea6, \u5bbd\u5ea6, 3)\u3002<\/li>\n<li>\u4e00\u6bb5\u89c6\u9891&#xff1a;\u5b83\u662f\u7531\u8fde\u7eed\u7684\u56fe\u7247\u5e27\u7ec4\u6210\u7684&#xff0c;\u6240\u4ee5\u5728\u56fe\u7247\u7684\u57fa\u7840\u4e0a\u589e\u52a0\u4e86\u4e00\u4e2a\u65f6\u95f4&#xff08;\u6216\u5e27\u6570&#xff09;\u7684\u7ef4\u5ea6&#xff0c;\u6784\u6210\u4e00\u4e2a4\u9636\u5f20\u91cf&#xff0c;\u5f62\u72b6\u4e3a\u00a0(\u5e27\u6570, \u9ad8\u5ea6, \u5bbd\u5ea6, 3)\u3002<\/li>\n<li>\u4e00\u4e2a\u6279\u6b21&#xff08;Batch&#xff09;\u7684\u6570\u636e&#xff1a;\u5728\u8bad\u7ec3\u795e\u7ecf\u7f51\u7edc\u65f6&#xff0c;\u6211\u4eec\u901a\u5e38\u4e0d\u4f1a\u4e00\u6b21\u53ea\u5904\u7406\u4e00\u5f20\u56fe\u7247&#xff0c;\u800c\u662f\u5c06\u591a\u5f20\u56fe\u7247\u7ec4\u6210\u4e00\u4e2a\u201c\u6279\u6b21\u201d\u6765\u5e76\u884c\u5904\u7406&#xff0c;\u4ee5\u63d0\u9ad8\u6548\u7387\u3002\u8fd9\u5c31\u53c8\u589e\u52a0\u4e86\u4e00\u4e2a\u201c\u6279\u91cf\u5927\u5c0f\u201d\u7684\u7ef4\u5ea6\u3002\u4f8b\u5982&#xff0c;\u4e00\u4e2a\u5305\u542b64\u5f20\u5f69\u8272\u56fe\u7247\u7684\u6279\u6b21&#xff0c;\u5c31\u662f\u4e00\u4e2a4\u9636\u5f20\u91cf&#xff0c;\u5f62\u72b6\u4e3a\u00a0(64, \u9ad8\u5ea6, \u5bbd\u5ea6, 3)\u3002<\/li>\n<\/ul>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u5bf9\u7ebf\u6027\u4ee3\u6570\u6838\u5fc3\u6982\u5ff5\u7684\u76f4\u89c2\u89e3\u8bfb\u3002\u8bfb\u8005\u5e94\u5efa\u7acb\u8d77\u8fd9\u6837\u7684\u8ba4\u77e5&#xff1a;\u5411\u91cf\u5b9a\u4e49\u4e86\u7a7a\u95f4\u4e2d\u7684\u5b58\u5728&#xff0c;\u77e9\u9635\u6267\u884c\u4e86\u5bf9\u7a7a\u95f4\u7684\u53d8\u6362&#xff0c;\u800c\u5f20\u91cf\u5219\u662f\u5bb9\u7eb3\u8fd9\u4e00\u5207\u9ad8\u7ef4\u6570\u636e\u7684\u6807\u51c6\u5bb9\u5668\u3002\u8fd9\u5957\u8bed\u8a00&#xff0c;\u4e3a\u6211\u4eec\u540e\u7eed\u7406\u89e3\u548c\u6784\u5efa\u590d\u6742\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b&#xff0c;\u6253\u4e0b\u4e86\u575a\u5b9e\u7684\u7ed3\u6784\u6027\u57fa\u7840\u3002\u63a5\u4e0b\u6765&#xff0c;\u6211\u4eec\u5c06\u5f15\u5165\u201c\u53d8\u5316\u201d\u7684\u7ef4\u5ea6&#xff0c;\u63a2\u7d22\u5fae\u79ef\u5206\u7684\u5965\u79d8\u3002<\/p>\n<h4>2.2 \u5fae\u79ef\u5206&#xff1a;\u7406\u89e3\u53d8\u5316\u7684\u8bed\u8a00\u4e0e\u4f18\u5316\u7684\u827a\u672f<\/h4>\n<p>\u5982\u679c\u8bf4\u7ebf\u6027\u4ee3\u6570\u7ed9\u4e86\u6211\u4eec\u63cf\u8ff0\u4e16\u754c\u72b6\u6001\u7684\u201c\u540d\u8bcd\u201d\u548c\u201c\u5f62\u5bb9\u8bcd\u201d&#xff08;\u5411\u91cf\u3001\u77e9\u9635&#xff09;&#xff0c;\u90a3\u4e48\u5fae\u79ef\u5206\u5219\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u63cf\u8ff0\u4e16\u754c\u8fd0\u52a8\u4e0e\u53d8\u5316\u7684\u201c\u52a8\u8bcd\u201d\u548c\u201c\u526f\u8bcd\u201d\u3002\u5b83\u662f\u5173\u4e8e\u201c\u53d8\u5316\u7387\u201d\u7684\u6570\u5b66&#xff0c;\u662f\u7406\u89e3\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5982\u4f55\u201c\u5b66\u4e60\u201d\u548c\u201c\u4f18\u5316\u201d\u7684\u5173\u952e\u6240\u5728\u3002<\/p>\n<h5>2.2.1 \u5bfc\u6570\u4e0e\u504f\u5bfc\u6570&#xff1a;\u6355\u6349\u201c\u77ac\u65f6\u53d8\u5316\u201d\u7684\u6167\u773c<\/h5>\n<p>\u5bfc\u6570\u7684\u672c\u8d28<\/p>\n<ul>\n<li>\n<p>\u4ece\u659c\u7387\u5230\u53d8\u5316\u7387 \u8bfb\u8005\u53ef\u80fd\u8fd8\u8bb0\u5f97&#xff0c;\u5bfc\u6570\u5728\u51e0\u4f55\u4e0a\u88ab\u5b9a\u4e49\u4e3a\u51fd\u6570\u66f2\u7ebf\u4e0a\u67d0\u4e00\u70b9\u7684\u5207\u7ebf\u659c\u7387\u3002\u4f46\u8fd9\u53ea\u662f\u5176\u8868\u8c61&#xff0c;\u5bfc\u6570\u66f4\u6df1\u523b\u7684\u672c\u8d28\u662f\u51fd\u6570\u5728\u8be5\u70b9\u7684\u201c\u77ac\u65f6\u53d8\u5316\u7387\u201d\u3002\u5b83\u7cbe\u51c6\u5730\u56de\u7b54\u4e86\u8fd9\u6837\u4e00\u4e2a\u6838\u5fc3\u95ee\u9898&#xff1a;\u201c\u5f53\u51fd\u6570\u7684\u8f93\u5165 x \u53d1\u751f\u4e00\u4e2a\u6781\u5176\u5fae\u5c0f\u7684\u6270\u52a8\u65f6&#xff0c;\u5176\u8f93\u51fa y \u4f1a\u5982\u4f55\u968f\u4e4b\u54cd\u5e94&#xff1f;\u201d<\/p>\n<p>\u8fd9\u4e2a\u201c\u54cd\u5e94\u201d\u7684\u5267\u70c8\u7a0b\u5ea6&#xff0c;\u5c31\u662f\u5bfc\u6570\u7684\u503c\u3002\u5bfc\u6570\u4e3a\u6b63&#xff0c;\u8868\u793a\u51fd\u6570\u5728\u6b64\u5904\u5448\u4e0a\u5347\u8d8b\u52bf&#xff1b;\u5bfc\u6570\u4e3a\u8d1f&#xff0c;\u5219\u5448\u4e0b\u964d\u8d8b\u52bf&#xff1b;\u5bfc\u6570\u7684\u7edd\u5bf9\u503c\u8d8a\u5927&#xff0c;\u8868\u793a\u53d8\u5316\u7684\u8d8b\u52bf\u8d8a\u201c\u9661\u5ced\u201d\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u504f\u5bfc\u6570\u7684\u667a\u6167 \u73b0\u5b9e\u4e16\u754c\u7684\u95ee\u9898&#xff0c;\u5176\u7ed3\u679c\u5f80\u5f80\u7531\u591a\u4e2a\u56e0\u7d20\u5171\u540c\u51b3\u5b9a\u3002\u4f8b\u5982&#xff0c;\u4e00\u4e2a\u5546\u54c1\u7684\u9500\u91cf S&#xff0c;\u53ef\u80fd\u540c\u65f6\u53d7\u5230\u4ef7\u683c P\u3001\u5e7f\u544a\u6295\u5165 A \u548c\u5b63\u8282 T \u7684\u5f71\u54cd&#xff0c;\u5373 S &#061; f(P, A, T)\u3002\u5982\u679c\u6211\u4eec\u60f3\u77e5\u9053\u201c\u5728\u4fdd\u6301\u5e7f\u544a\u548c\u5b63\u8282\u4e0d\u53d8\u7684\u60c5\u51b5\u4e0b&#xff0c;\u4ef7\u683c\u6bcf\u4e0a\u6da8\u4e00\u5143&#xff0c;\u9500\u91cf\u4f1a\u5982\u4f55\u53d8\u5316&#xff1f;\u201d&#xff0c;\u8fd9\u65f6\u5c31\u9700\u8981\u504f\u5bfc\u6570\u3002<\/p>\n<p>\u504f\u5bfc\u6570\u7684\u667a\u6167\u5728\u4e8e&#xff0c;\u5b83\u5141\u8bb8\u6211\u4eec\u5728\u591a\u53d8\u91cf\u7684\u590d\u6742\u7cfb\u7edf\u4e2d&#xff0c;\u6682\u65f6\u201c\u51bb\u7ed3\u201d\u5176\u4ed6\u6240\u6709\u53d8\u91cf&#xff0c;\u5355\u72ec\u8003\u5bdf\u67d0\u4e00\u4e2a\u53d8\u91cf\u5bf9\u51fd\u6570\u7ed3\u679c\u7684\u5f71\u54cd\u529b\u3002\u8fd9\u662f\u4e00\u79cd\u5f3a\u5927\u7684\u201c\u5f52\u56e0\u201d\u5de5\u5177&#xff0c;\u8ba9\u6211\u4eec\u80fd\u591f\u9010\u4e00\u5206\u6790\u6bcf\u4e2a\u56e0\u7d20\u7684\u8d21\u732e\u3002\u5728\u6570\u5b66\u4e0a&#xff0c;\u51fd\u6570 f \u5bf9\u53d8\u91cf P \u7684\u504f\u5bfc\u6570\u8bb0\u4f5c \u2202f\/\u2202P\u3002<\/p>\n<h5>2.2.2 \u94fe\u5f0f\u6cd5\u5219&#xff1a;\u5c42\u5c42\u89e3\u6784\u201c\u56e0\u679c\u94fe\u6761\u201d<\/h5>\n<p>\u94fe\u5f0f\u6cd5\u5219\u7684\u6838\u5fc3\u601d\u60f3 \u8bb8\u591a\u7cfb\u7edf\u4e2d\u7684\u56e0\u679c\u5173\u7cfb\u662f\u5c42\u5c42\u5d4c\u5957\u7684\u3002\u6bd4\u5982&#xff0c;\u65bd\u80a5\u91cf x \u5f71\u54cd\u571f\u58e4\u80a5\u529b u&#xff0c;\u571f\u58e4\u80a5\u529b u \u53c8\u5f71\u54cd\u4f5c\u7269\u4ea7\u91cf y\u3002\u5982\u679c\u6211\u4eec\u60f3\u77e5\u9053\u201c\u65bd\u80a5\u91cf x \u7684\u5fae\u5c0f\u53d8\u5316&#xff0c;\u6700\u7ec8\u4f1a\u5bf9\u4ea7\u91cf y \u4ea7\u751f\u591a\u5927\u5f71\u54cd&#xff1f;\u201d&#xff0c;\u5c31\u9700\u8981\u7528\u5230\u94fe\u5f0f\u6cd5\u5219\u3002<\/p>\n<p>\u94fe\u5f0f\u6cd5\u5219\u63ed\u793a\u4e86\u5728\u4e00\u4e2a\u7531 y &#061; f(u) \u548c u &#061; g(x) \u6784\u6210\u7684\u590d\u5408\u51fd\u6570 y &#061; f(g(x)) \u4e2d&#xff0c;\u6700\u7ec8\u7ed3\u679c\u7684\u53d8\u5316\u7387\u662f\u5982\u4f55\u7531\u5404\u4e2a\u73af\u8282\u7684\u53d8\u5316\u7387\u76f8\u4e58\u4f20\u9012\u800c\u6765\u7684&#xff1a; dy\/dx &#061; dy\/du * du\/dx \u6700\u7ec8\u5f71\u54cd\u7b49\u4e8e\u4e2d\u95f4\u73af\u8282\u5f71\u54cd\u4e0e\u521d\u59cb\u73af\u8282\u5f71\u54cd\u7684\u4e58\u79ef\u3002\u8fd9\u662f\u4e00\u4e2a\u6781\u5176\u6df1\u523b\u7684\u89c4\u5f8b&#xff0c;\u5b83\u8ba9\u6211\u4eec\u80fd\u591f\u89e3\u6784\u4e00\u4e2a\u590d\u6742\u7684\u56e0\u679c\u94fe\u6761\u3002<\/p>\n<p>\u53cd\u5411\u4f20\u64ad\u7684\u7075\u9b42 \u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc&#xff0c;\u5c31\u662f\u4e00\u4e2a\u6781\u6df1\u3001\u6781\u590d\u6742\u7684\u590d\u5408\u51fd\u6570\u3002\u7f51\u7edc\u7684\u8f93\u5165 X \u7ecf\u8fc7\u7b2c\u4e00\u5c42\u53d8\u6362\u5f97\u5230 H\u2081&#xff0c;H\u2081 \u7ecf\u8fc7\u7b2c\u4e8c\u5c42\u53d8\u6362\u5f97\u5230 H\u2082&#xff0c;\u2026\u2026&#xff0c;\u6700\u7ec8\u7ecf\u8fc7\u8f93\u51fa\u5c42\u5f97\u5230\u9884\u6d4b\u7ed3\u679c \u0176&#xff0c;\u800c\u9884\u6d4b\u7ed3\u679c \u0176 \u4e0e\u771f\u5b9e\u6807\u7b7e Y \u4e4b\u95f4\u4f1a\u4ea7\u751f\u4e00\u4e2a\u8bef\u5dee&#xff08;\u6216\u635f\u5931&#xff09;L\u3002<\/p>\n<p>\u6211\u4eec\u8bad\u7ec3\u7f51\u7edc\u7684\u76ee\u6807&#xff0c;\u662f\u8c03\u6574\u7f51\u7edc\u4e2d\u6210\u5343\u4e0a\u4e07\u7684\u53c2\u6570&#xff08;\u6743\u91cd W \u548c\u504f\u7f6e b&#xff09;&#xff0c;\u4ee5\u6700\u5c0f\u5316\u8fd9\u4e2a\u6700\u7ec8\u7684\u8bef\u5dee L\u3002\u4f46\u95ee\u9898\u662f&#xff0c;\u8eab\u5904\u7f51\u7edc\u6df1\u5904\u7684\u67d0\u4e2a\u53c2\u6570 W\u1d62\u2c7c&#xff0c;\u5b83\u4e00\u4e2a\u5c0f\u5c0f\u7684\u53d8\u52a8&#xff0c;\u662f\u5982\u4f55\u901a\u8fc7\u5c42\u5c42\u4f20\u9012&#xff0c;\u6700\u7ec8\u5f71\u54cd\u5230 L \u7684\u5462&#xff1f;<\/p>\n<p>\u53cd\u5411\u4f20\u64ad&#xff08;Backpropagation&#xff09;\u7b97\u6cd5&#xff0c;\u6b63\u662f\u94fe\u5f0f\u6cd5\u5219\u5728\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u4e00\u6b21\u5b8f\u5927\u800c\u8f89\u714c\u7684\u5e94\u7528\u3002\u5b83\u4ece\u6700\u7ec8\u7684\u8bef\u5dee L \u51fa\u53d1&#xff0c;\u201c\u53cd\u5411\u201d\u5730\u3001\u9010\u5c42\u5730\u8fd0\u7528\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u8ba1\u7b97\u51fa\u8bef\u5dee L \u5bf9\u7f51\u7edc\u4e2d\u6bcf\u4e00\u4e2a\u53c2\u6570\u7684\u504f\u5bfc\u6570&#xff08;\u2202L\/\u2202W\u1d62\u2c7c&#xff09;\u3002\u8fd9\u4e2a\u504f\u5bfc\u6570&#xff0c;\u5c31\u7cbe\u786e\u5730\u8861\u91cf\u4e86\u8be5\u53c2\u6570\u5bf9\u6700\u7ec8\u8bef\u5dee\u7684\u201c\u8d21\u732e\u5ea6\u201d&#xff0c;\u5e76\u6307\u660e\u4e86\u5b83\u5e94\u8be5\u8c03\u6574\u7684\u65b9\u5411\u3002\u53ef\u4ee5\u8bf4&#xff0c;\u6ca1\u6709\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u5c31\u6ca1\u6709\u53cd\u5411\u4f20\u64ad&#xff0c;\u4e5f\u5c31\u6ca1\u6709\u73b0\u4ee3\u6df1\u5ea6\u5b66\u4e60\u7684\u6839\u57fa\u3002<\/p>\n<h5>2.2.3 \u68af\u5ea6&#xff1a;\u6307\u5411\u201c\u6700\u4f18\u201d\u7684\u7f57\u76d8<\/h5>\n<p>\u68af\u5ea6\u7684\u51e0\u4f55\u610f\u4e49<\/p>\n<p>\u5bf9\u4e8e\u4e00\u4e2a\u591a\u53d8\u91cf\u51fd\u6570&#xff0c;\u6211\u4eec\u5c06\u5b83\u5bf9\u6240\u6709\u81ea\u53d8\u91cf\u7684\u504f\u5bfc\u6570&#xff0c;\u6253\u5305\u6210\u4e00\u4e2a\u5411\u91cf&#xff0c;\u8fd9\u4e2a\u5411\u91cf\u5c31\u53eb\u505a\u68af\u5ea6&#xff08;Gradient&#xff09;\u3002\u4f8b\u5982&#xff0c;\u5bf9\u4e8e\u51fd\u6570 f(x, y)&#xff0c;\u5176\u68af\u5ea6\u4e3a \u2207f &#061; [\u2202f\/\u2202x, \u2202f\/\u2202y]\u3002<\/p>\n<p>\u68af\u5ea6\u8fd9\u4e2a\u5411\u91cf&#xff0c;\u5177\u6709\u4e00\u4e2a\u975e\u51e1\u7684\u51e0\u4f55\u610f\u4e49&#xff1a;\u5b83\u6307\u5411\u51fd\u6570\u503c\u589e\u957f\u6700\u5feb\u7684\u65b9\u5411\u3002<\/p>\n<p>\u8d1f\u68af\u5ea6&#xff1a;\u76f8\u5e94\u5730&#xff0c;\u68af\u5ea6\u7684\u53cd\u65b9\u5411&#xff08;-\u2207f&#xff09;&#xff0c;\u5c31\u6307\u5411\u4e86\u51fd\u6570\u503c\u4e0b\u964d\u6700\u5feb\u7684\u65b9\u5411\u3002<\/p>\n<p>\u68af\u5ea6\u4e0b\u964d\u7684\u4fee\u884c<\/p>\n<p>\u68af\u5ea6\u4e3a\u6211\u4eec\u6307\u660e\u4e86\u65b9\u5411&#xff0c;\u800c\u68af\u5ea6\u4e0b\u964d\u7b97\u6cd5&#xff0c;\u5c31\u662f\u6cbf\u7740\u8fd9\u4e2a\u65b9\u5411\u53bb\u5bfb\u627e\u51fd\u6570\u6700\u5c0f\u503c\u7684\u5177\u4f53\u65b9\u6cd5\u3002\u5b83\u662f\u6df1\u5ea6\u5b66\u4e60\u4e2d\u6700\u6838\u5fc3\u3001\u6700\u5e38\u7528\u7684\u4f18\u5316\u7b97\u6cd5\u3002\u6211\u4eec\u53ef\u4ee5\u7528\u4e00\u4e2a\u751f\u52a8\u7684\u6bd4\u55bb\u6765\u7406\u89e3\u5b83\u7684\u4fee\u884c\u8fc7\u7a0b&#xff1a;<\/p>\n<p>\u9ed1\u591c\u4e0b\u5c71\u7684\u6bd4\u55bb&#xff1a; \u60f3\u8c61\u4e00\u4f4d\u8bfb\u8005\u6df1\u591c\u88ab\u56f0\u5728\u4e00\u5ea7\u8fde\u7ef5\u7684\u5c71\u8109\u4e2d&#xff08;\u5c71\u8109\u7684\u8868\u9762\u5c31\u662f\u6211\u4eec\u60f3\u8981\u6700\u5c0f\u5316\u7684\u635f\u5931\u51fd\u6570&#xff09;\u3002\u76ee\u6807\u662f\u5c3d\u5feb\u8d70\u5230\u5c71\u8109\u7684\u6700\u4f4e\u70b9&#xff08;\u635f\u5931\u51fd\u6570\u7684\u6700\u5c0f\u503c&#xff09;\u3002<\/p>\n<p>\u7531\u4e8e\u5929\u9ed1&#xff0c;\u65e0\u6cd5\u770b\u6e05\u5168\u5c40\u5730\u8c8c\u3002\u4f46\u8fd9\u4f4d\u8bfb\u8005\u53ef\u4ee5\u505a\u4e00\u4ef6\u4e8b&#xff1a;\u7528\u811a\u5728\u539f\u5730\u8bd5\u63a2\u4e00\u5708&#xff0c;\u611f\u53d7\u54ea\u4e2a\u65b9\u5411\u662f\u4e0b\u5761\u6700\u9661\u5ced\u7684\u3002\u8fd9\u4e2a\u201c\u6700\u9661\u5ced\u7684\u4e0b\u5761\u65b9\u5411\u201d&#xff0c;\u6b63\u662f\u8d1f\u68af\u5ea6\u65b9\u5411\u3002<\/p>\n<p>\u4e8e\u662f&#xff0c;\u4ed6\u671d\u7740\u8fd9\u4e2a\u65b9\u5411\u8fc8\u51fa\u4e00\u6b65\u3002\u8fd9\u4e00\u6b65\u7684\u201c\u5927\u5c0f\u201d&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u5b66\u4e60\u7387\u3002<\/p>\n<p>\u5982\u679c\u6b65\u5b50\u8fc8\u5f97\u592a\u5927&#xff08;\u5b66\u4e60\u7387\u8fc7\u9ad8&#xff09;&#xff0c;\u53ef\u80fd\u4f1a\u76f4\u63a5\u8de8\u8fc7\u8c37\u5e95&#xff0c;\u751a\u81f3\u8dd1\u5230\u5bf9\u9762\u7684\u5c71\u5761\u4e0a&#xff0c;\u5bfc\u81f4\u201c\u632f\u8361\u201d\u6216\u201c\u53d1\u6563\u201d\u3002<\/p>\n<p>\u5230\u8fbe\u65b0\u4f4d\u7f6e\u540e&#xff0c;\u4ed6\u518d\u6b21\u91cd\u590d\u8fd9\u4e2a\u8fc7\u7a0b&#xff1a;\u73af\u987e\u56db\u5468&#xff0c;\u627e\u5230\u5f53\u524d\u4f4d\u7f6e\u6700\u9661\u7684\u4e0b\u5761\u65b9\u5411&#xff0c;\u518d\u8fc8\u51fa\u4e00\u6b65\u2026\u2026\u5468\u800c\u590d\u59cb&#xff0c;\u901a\u8fc7\u65e0\u6570\u6b21\u7684\u8fed\u4ee3&#xff0c;\u4ed6\u5c06\u4e00\u6b65\u6b65\u5730\u903c\u8fd1\u5c71\u8c37\u7684\u6700\u4f4e\u70b9\u3002<\/p>\n<p>\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u4e00\u573a\u5728\u4ebf\u4e07\u7ef4\u5ea6\u53c2\u6570\u7a7a\u95f4\u4e2d\u7684\u201c\u68af\u5ea6\u4e0b\u964d\u201d\u4e4b\u65c5\u3002\u901a\u8fc7\u4e0d\u65ad\u5730\u8ba1\u7b97\u635f\u5931\u51fd\u6570\u5173\u4e8e\u6240\u6709\u53c2\u6570\u7684\u68af\u5ea6&#xff0c;\u5e76\u6cbf\u7740\u8d1f\u68af\u5ea6\u65b9\u5411\u53bb\u66f4\u65b0\u8fd9\u4e9b\u53c2\u6570&#xff0c;\u6a21\u578b\u6700\u7ec8\u80fd\u627e\u5230\u4e00\u7ec4\u4f7f\u635f\u5931\u6700\u5c0f\u5316\u7684\u6700\u4f18\u53c2\u6570\u3002<\/p>\n<p>\u5982\u679c\u6b65\u5b50\u8fc8\u5f97\u592a\u5c0f&#xff08;\u5b66\u4e60\u7387\u8fc7\u4f4e&#xff09;&#xff0c;\u4e0b\u5c71\u7684\u901f\u5ea6\u4f1a\u975e\u5e38\u7f13\u6162\u3002<\/p>\n<h4>2.3 \u6982\u7387\u8bba\u4e0e\u4fe1\u606f\u8bba&#xff1a;\u8861\u91cf\u4e0d\u786e\u5b9a\u6027\u4e0e\u77e5\u8bc6<\/h4>\n<p>\u4e16\u754c\u5145\u6ee1\u4e86\u968f\u673a\u4e0e\u672a\u77e5\u3002\u6982\u7387\u8bba\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u62e5\u62b1\u548c\u63cf\u8ff0\u4e0d\u786e\u5b9a\u6027\u7684\u6570\u5b66\u8bed\u8a00&#xff0c;\u800c\u4fe1\u606f\u8bba\u5219\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u4e00\u628a\u5ea6\u91cf\u201c\u4fe1\u606f\u201d\u4e0e\u201c\u77e5\u8bc6\u201d\u7684\u5c3a\u5b50\u3002<\/p>\n<h5>2.3.1 \u6982\u7387\u5206\u5e03\u4e0e\u671f\u671b&#xff1a;\u62e5\u62b1\u4e0d\u786e\u5b9a\u6027<\/h5>\n<p>\u6982\u7387\u5206\u5e03&#xff08;Probability Distribution&#xff09; \u5b83\u662f\u4e00\u5f20\u63cf\u7ed8\u968f\u673a\u53d8\u91cf\u6240\u6709\u53ef\u80fd\u53d6\u503c\u53ca\u5176\u5bf9\u5e94\u6982\u7387\u7684\u84dd\u56fe\u3002\u4f8b\u5982&#xff0c;\u4e00\u4e2a\u516c\u5e73\u9ab0\u5b50\u7684\u6982\u7387\u5206\u5e03\u662f {1:1\/6, 2:1\/6, &#8230;, 6:1\/6}\u3002\u5728\u673a\u5668\u5b66\u4e60\u4e2d&#xff0c;\u5206\u7c7b\u6a21\u578b\u7684\u8f93\u51fa&#xff0c;\u5f80\u5f80\u5c31\u662f\u4e00\u4e2a\u6982\u7387\u5206\u5e03&#xff0c;\u5982 {\u732b: 0.8, \u72d7: 0.15, \u5176\u4ed6: 0.05}\u3002<\/p>\n<p>\u671f\u671b&#xff08;Expectation&#xff09; \u5b83\u662f\u6982\u7387\u5206\u5e03\u7684\u52a0\u6743\u5e73\u5747\u503c&#xff0c;\u662f\u5728\u4e0d\u786e\u5b9a\u6027\u4e2d\u6211\u4eec\u80fd\u505a\u51fa\u7684\u6700\u201c\u5408\u7406\u201d\u7684\u957f\u671f\u5e73\u5747\u731c\u6d4b\u3002\u63b7\u9ab0\u5b50\u7684\u671f\u671b\u70b9\u6570\u662f3.5\u3002\u867d\u7136\u5355\u6b21\u63b7\u4e0d\u51fa3.5&#xff0c;\u4f46\u5927\u91cf\u91cd\u590d\u5b9e\u9a8c\u7684\u5e73\u5747\u7ed3\u679c\u4f1a\u8d8b\u8fd1\u4e8e\u5b83\u3002<\/p>\n<h5>2.3.2 \u71b5&#xff1a;\u5ea6\u91cf\u201c\u6df7\u4e71\u201d\u4e0e\u201c\u4fe1\u606f\u201d<\/h5>\n<p>\u4fe1\u606f\u71b5&#xff08;Information Entropy&#xff09;\u00a0\u4fe1\u606f\u71b5\u662f\u4fe1\u606f\u8bba\u7684\u5960\u57fa\u6982\u5ff5&#xff0c;\u7528\u4e8e\u91cf\u5316\u4e00\u4e2a\u7cfb\u7edf\u7684\u4e0d\u786e\u5b9a\u6027\u6216\u6df7\u4e71\u7a0b\u5ea6\u3002\u4e00\u4e2a\u7cfb\u7edf\u8d8a\u6df7\u4e71\u3001\u7ed3\u679c\u8d8a\u4e0d\u53ef\u9884\u6d4b&#xff0c;\u5176\u71b5\u503c\u5c31\u8d8a\u9ad8\u3002<\/p>\n<ul>\n<li>\u4e00\u4e2a\u4f5c\u5f0a\u7684\u3001\u603b\u51fa\u201c6\u201d\u7684\u9ab0\u5b50&#xff0c;\u5176\u7ed3\u679c\u5b8c\u5168\u786e\u5b9a&#xff0c;\u71b5\u4e3a0\u3002<\/li>\n<li>\u4e00\u4e2a\u516c\u5e73\u7684\u9ab0\u5b50&#xff0c;\u5176\u7ed3\u679c\u6700\u4e0d\u786e\u5b9a&#xff0c;\u71b5\u6700\u5927\u3002 \u71b5\u8fd8\u6709\u53e6\u4e00\u5c42\u6df1\u523b\u542b\u4e49&#xff1a;\u5b83\u7b49\u4e8e\u5f7b\u5e95\u641e\u6e05\u695a\u8be5\u7cfb\u7edf\u72b6\u6001\u6240\u9700\u8981\u7684\u6700\u5c11\u4fe1\u606f\u91cf\u3002<\/li>\n<\/ul>\n<h5>2.3.3 \u4ea4\u53c9\u71b5&#xff1a;\u8861\u91cf\u201c\u8ba4\u77e5\u201d\u4e0e\u201c\u771f\u76f8\u201d\u7684\u8ddd\u79bb<\/h5>\n<p>\u4eceKL\u6563\u5ea6\u5230\u4ea4\u53c9\u71b5 \u5982\u679c\u6211\u4eec\u6709\u4e24\u4e2a\u6982\u7387\u5206\u5e03&#xff0c;\u4e00\u4e2a\u4ee3\u8868\u201c\u771f\u76f8\u201d\u7684\u5206\u5e03 P&#xff0c;\u53e6\u4e00\u4e2a\u4ee3\u8868\u6211\u4eec\u6a21\u578b\u7684\u201c\u8ba4\u77e5\u201d&#xff08;\u9884\u6d4b&#xff09;\u7684\u5206\u5e03 Q&#xff0c;\u6211\u4eec\u5982\u4f55\u8861\u91cf\u5b83\u4eec\u4e4b\u95f4\u7684\u5dee\u8ddd&#xff1f;**KL\u6563\u5ea6&#xff08;Kullback-Leibler Divergence&#xff09;**\u6b63\u662f\u4e3a\u6b64\u800c\u751f\u3002 \u5728\u673a\u5668\u5b66\u4e60\u7684\u5206\u7c7b\u4efb\u52a1\u4e2d&#xff0c;\u6211\u4eec\u66f4\u5e38\u7528\u4e00\u4e2a\u4e0eKL\u6563\u5ea6\u7d27\u5bc6\u76f8\u5173\u7684\u91cf\u2014\u2014\u4ea4\u53c9\u71b5&#xff08;Cross-Entropy&#xff09;\u6765\u4f5c\u4e3a\u635f\u5931\u51fd\u6570\u3002<\/p>\n<p>\u635f\u5931\u51fd\u6570\u7684\u54f2\u5b66 \u4ea4\u53c9\u71b5\u635f\u5931\u51fd\u6570\u8861\u91cf\u7684\u662f&#xff1a;\u7528\u6211\u4eec\u7684\u201c\u8ba4\u77e5\u201dQ \u53bb\u7f16\u7801\u6765\u81ea\u201c\u771f\u76f8\u201dP \u7684\u4e8b\u4ef6&#xff0c;\u5e73\u5747\u9700\u8981\u591a\u5c11\u4fe1\u606f\u91cf\u3002\u5982\u679c Q \u4e0e P \u5b8c\u5168\u4e00\u81f4&#xff0c;\u4ea4\u53c9\u71b5\u5c31\u8fbe\u5230\u6700\u5c0f\u503c\u3002 \u56e0\u6b64&#xff0c;\u6a21\u578b\u8bad\u7ec3\u7684\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u4e0d\u65ad\u8c03\u6574\u53c2\u6570&#xff0c;\u4ee5\u6700\u5c0f\u5316\u9884\u6d4b\u5206\u5e03\u4e0e\u771f\u5b9e\u6807\u7b7e\u5206\u5e03\u4e4b\u95f4\u7684\u4ea4\u53c9\u71b5\u3002\u8fd9\u5728\u54f2\u5b66\u4e0a&#xff0c;\u53ef\u4ee5\u770b\u4f5c\u662f\u4e00\u4e2a\u6a21\u578b\u4e0d\u65ad\u653e\u4e0b\u81ea\u5df1\u7684\u201c\u6211\u6267\u201d&#xff08;\u9519\u8bef\u7684\u8ba4\u77e5&#xff09;&#xff0c;\u52aa\u529b\u53bb\u8d8b\u8fd1\u201c\u771f\u76f8\u201d\u7684\u4fee\u884c\u8fc7\u7a0b\u3002<\/p>\n<h4>2.4 NumPy&#xff1a;\u79d1\u5b66\u8ba1\u7b97\u7684\u57fa\u77f3\u4e0e\u52a0\u901f\u5668<\/h4>\n<p>\u7406\u8bba\u5b66\u4e60\u5b8c\u6bd5&#xff0c;\u73b0\u5728\u6211\u4eec\u5f00\u59cb\u94f8\u9020\u201c\u795e\u5175\u201d\u3002NumPy&#xff08;Numerical Python&#xff09;\u662fPython\u79d1\u5b66\u8ba1\u7b97\u751f\u6001\u7684\u6838\u5fc3&#xff0c;\u5b83\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u9ad8\u6027\u80fd\u7684\u591a\u7ef4\u6570\u7ec4\u5bf9\u8c61\u53ca\u76f8\u5173\u5de5\u5177\u3002<\/p>\n<h5>2.4.1\u00a0ndarray&#xff1a;NumPy\u7684\u7075\u9b42<\/h5>\n<p>\u9ad8\u6027\u80fd\u7684\u591a\u7ef4\u6570\u7ec4\u5bf9\u8c61 NumPy\u7684\u6838\u5fc3\u662fndarray\u5bf9\u8c61\u3002\u5b83\u7684\u9ad8\u6027\u80fd\u6e90\u4e8e\u4e24\u4e2a\u5173\u952e\u8bbe\u8ba1&#xff1a;<\/p>\n<ul>\n<li>\u5185\u5b58\u5e03\u5c40&#xff1a;ndarray\u5728\u5185\u5b58\u4e2d\u662f\u4e00\u5757\u8fde\u7eed\u7684\u3001\u672a\u5206\u88c5\u7684\u5b58\u50a8\u533a\u57df&#xff0c;\u6240\u6709\u5143\u7d20\u7c7b\u578b\u76f8\u540c\u3002\u8fd9\u4f7f\u5f97\u5b83\u53ef\u4ee5\u5229\u7528\u73b0\u4ee3CPU\u7684\u5411\u91cf\u5316\u6307\u4ee4&#xff08;SIMD&#xff09;&#xff0c;\u8fdb\u884c\u5e76\u884c\u8ba1\u7b97\u3002<\/li>\n<li>C\u8bed\u8a00\u540e\u7aef&#xff1a;\u5176\u5e95\u5c42\u8fd0\u7b97\u7531\u9ad8\u5ea6\u4f18\u5316\u7684C\u8bed\u8a00\u4ee3\u7801\u6267\u884c&#xff0c;\u907f\u514d\u4e86Python\u539f\u751f\u89e3\u91ca\u5668\u7684\u6027\u80fd\u5f00\u9500\u3002<\/li>\n<\/ul>\n<p>\u6570\u7ec4\u7684\u521b\u5efa\u4e0e\u53d8\u5f62<\/p>\n<p>import numpy as np<br \/>\n# \u4ece\u5217\u8868\u521b\u5efa<br \/>\na &#061; np.array([1, 2, 3])<br \/>\n# \u521b\u5efa\u7279\u5b9a\u6570\u7ec4<br \/>\nzeros &#061; np.zeros((2, 3)) # 2&#215;3\u7684\u51680\u77e9\u9635<br \/>\nones &#061; np.ones(5)       # \u957f\u5ea6\u4e3a5\u7684\u51681\u5411\u91cf<br \/>\narange &#061; np.arange(0, 10, 2) # [0, 2, 4, 6, 8]<br \/>\n# \u6539\u53d8\u5f62\u72b6<br \/>\nb &#061; np.arange(6).reshape((2, 3)) # \u521b\u5efa\u4e00\u4e2a1&#215;6\u5411\u91cf\u5e76\u91cd\u5851\u4e3a2&#215;3\u77e9\u9635<br \/>\n# \u8f6c\u7f6e<br \/>\nc &#061; b.T<\/p>\n<h5>2.4.2 \u5411\u91cf\u5316\u8fd0\u7b97\u4e0e\u5e7f\u64ad\u673a\u5236<\/h5>\n<p>\u5411\u91cf\u5316&#xff08;Vectorization&#xff09;&#xff1a;\u544a\u522b\u4f4e\u6548\u5faa\u73af \u5411\u91cf\u5316\u662f\u6307\u76f4\u63a5\u5bf9\u6574\u4e2a\u6570\u7ec4\u6216\u6570\u7ec4\u95f4\u8fdb\u884c\u8fd0\u7b97&#xff0c;\u800c\u65e0\u9700\u7f16\u5199\u663e\u5f0f\u7684for\u5faa\u73af\u3002<\/p>\n<p># \u975e\u5411\u91cf\u5316 (\u6162)<br \/>\na &#061; list(range(1000000))<br \/>\nb &#061; list(range(1000000))<br \/>\nc &#061; []<br \/>\nfor i in range(len(a)):<br \/>\n    c.append(a[i] &#043; b[i])<\/p>\n<p># \u5411\u91cf\u5316 (\u5feb)<br \/>\na_np &#061; np.arange(1000000)<br \/>\nb_np &#061; np.arange(1000000)<br \/>\nc_np &#061; a_np &#043; b_np <\/p>\n<p>\u5728\u6027\u80fd\u4e0a&#xff0c;\u5411\u91cf\u5316\u7248\u672c\u901a\u5e38\u6bd4\u5faa\u73af\u7248\u672c\u5feb\u51e0\u4e2a\u6570\u91cf\u7ea7\u3002<\/p>\n<p>\u5e7f\u64ad&#xff08;Broadcasting&#xff09;&#xff1a;\u667a\u80fd\u7684\u7ef4\u5ea6\u5bf9\u9f50 \u5e7f\u64ad\u662fNumPy\u6700\u5f3a\u5927\u7684\u7279\u6027\u4e4b\u4e00&#xff0c;\u5b83\u5141\u8bb8\u4e0d\u540c\u5f62\u72b6\u7684\u6570\u7ec4\u5728\u4e00\u5b9a\u89c4\u5219\u4e0b\u8fdb\u884c\u7b97\u672f\u8fd0\u7b97\u3002<\/p>\n<p>\u89c4\u5219\u6838\u5fc3&#xff1a;\u4ece\u4e24\u4e2a\u6570\u7ec4\u7684\u5c3e\u90e8\u7ef4\u5ea6\u5f00\u59cb\u9010\u4e00\u6bd4\u8f83&#xff0c;\u5982\u679c\u7ef4\u5ea6\u76f8\u7b49&#xff0c;\u6216\u5176\u4e2d\u4e00\u4e2a\u4e3a1&#xff0c;\u5219\u53ef\u4ee5\u5e7f\u64ad&#xff1b;\u5426\u5219\u62a5\u9519\u3002\u7f3a\u5931\u7684\u7ef4\u5ea6\u4f1a\u88ab\u89c6\u4e3a1\u3002<\/p>\n<p>matrix &#061; np.array([[1, 2, 3], [4, 5, 6]]) # shape (2, 3)<br \/>\nvector &#061; np.array([10, 20, 30])          # shape (3,)<br \/>\n# \u5e7f\u64ad\u673a\u5236\u4f1a\u81ea\u52a8\u5c06vector\u201c\u6269\u5c55\u201d\u4e3a[[10,20,30], [10,20,30]]<br \/>\nresult &#061; matrix &#043; vector<br \/>\n# result is [[11, 22, 33], [14, 25, 36]]<\/p>\n<h4>2.5 Pandas&#xff1a;\u9a7e\u9a6d\u7ed3\u6784\u5316\u6570\u636e\u7684\u745e\u58eb\u519b\u5200<\/h4>\n<p>Pandas\u662f\u5efa\u7acb\u5728NumPy\u4e4b\u4e0a\u7684\u6570\u636e\u5206\u6790\u5e93&#xff0c;\u63d0\u4f9b\u4e86Series\u548cDataFrame\u4e24\u79cd\u5f3a\u5927\u7684\u6570\u636e\u7ed3\u6784&#xff0c;\u662f\u5904\u7406\u548c\u5206\u6790\u8868\u683c\u7c7b&#xff08;\u7ed3\u6784\u5316&#xff09;\u6570\u636e\u7684\u9996\u9009\u5de5\u5177\u3002<\/p>\n<h5>2.5.1\u00a0Series\u4e0eDataFrame&#xff1a;\u5e26\u6807\u7b7e\u7684\u6570\u636e\u5bb9\u5668<\/h5>\n<ul>\n<li>Series&#xff1a;\u4e00\u4e2a\u5e26\u6807\u7b7e\u7684\u4e00\u7ef4\u6570\u7ec4&#xff0c;\u53ef\u4ee5\u770b\u4f5c\u662fNumPy\u6570\u7ec4\u548cPython\u5b57\u5178\u7684\u7ed3\u5408\u4f53\u3002<\/li>\n<li>DataFrame&#xff1a;\u4e00\u4e2a\u4e8c\u7ef4\u7684\u3001\u5e26\u6807\u7b7e\u7684\u8868\u683c\u578b\u6570\u636e\u7ed3\u6784&#xff0c;\u62e5\u6709\u884c\u7d22\u5f15\u548c\u5217\u7d22\u5f15\u3002\u5b83\u662fPandas\u4e2d\u4f7f\u7528\u6700\u5e7f\u6cdb\u7684\u6838\u5fc3\u3002<\/li>\n<\/ul>\n<h5>2.5.2 \u6570\u636e\u7684\u63a2\u67e5\u3001\u6e05\u6d17\u4e0e\u9884\u5904\u7406<\/h5>\n<p>import pandas as pd<br \/>\n# \u5047\u8bbe\u6211\u4eec\u6709\u4e00\u4e2aCSV\u6587\u4ef6 &#039;titanic.csv&#039;<br \/>\ndf &#061; pd.read_csv(&#039;titanic.csv&#039;)<\/p>\n<p># 2.5.2.1 \u6570\u636e\u63a2\u67e5<br \/>\nprint(df.head())      # \u67e5\u770b\u524d5\u884c<br \/>\nprint(df.info())      # \u67e5\u770b\u5217\u4fe1\u606f\u3001\u6570\u636e\u7c7b\u578b\u3001\u975e\u7a7a\u503c\u6570\u91cf<br \/>\nprint(df.describe())  # \u67e5\u770b\u6570\u503c\u5217\u7684\u7edf\u8ba1\u6458\u8981<\/p>\n<p># 2.5.2.2 \u6570\u636e\u6e05\u6d17<br \/>\n# \u5904\u7406\u7f3a\u5931\u503c&#xff1a;\u5c06Age\u5217\u7684\u7f3a\u5931\u503c\u7528\u5e74\u9f84\u4e2d\u4f4d\u6570\u586b\u5145<br \/>\nmedian_age &#061; df[&#039;Age&#039;].median()<br \/>\ndf[&#039;Age&#039;].fillna(median_age, inplace&#061;True)<\/p>\n<p># 2.5.2.3 \u6570\u636e\u8f6c\u6362\u4e0e\u64cd\u4f5c<br \/>\n# \u9009\u62e9\u4e0e\u8fc7\u6ee4&#xff1a;\u9009\u62e9\u6240\u6709\u5e74\u9f84\u5927\u4e8e60\u5c81\u7684\u4e58\u5ba2<br \/>\nseniors &#061; df.loc[df[&#039;Age&#039;] &gt; 60]<\/p>\n<p># \u5206\u7ec4\u4e0e\u805a\u5408&#xff1a;\u6309\u6027\u522b\u8ba1\u7b97\u5e73\u5747\u7968\u4ef7<br \/>\navg_fare_by_sex &#061; df.groupby(&#039;Sex&#039;)[&#039;Fare&#039;].mean()<br \/>\nprint(avg_fare_by_sex)<\/p>\n<h4>2.6 Matplotlib &amp; Seaborn&#xff1a;\u8ba9\u6570\u636e\u5f00\u53e3\u8bf4\u8bdd\u7684\u827a\u672f<\/h4>\n<p>\u6570\u636e\u53ef\u89c6\u5316\u662f\u7406\u89e3\u6570\u636e\u3001\u5c55\u793a\u6d1e\u89c1\u7684\u5f3a\u5927\u624b\u6bb5\u3002Matplotlib\u662fPython\u53ef\u89c6\u5316\u7684\u57fa\u7840\u5e93&#xff0c;\u800cSeaborn\u5219\u662f\u5728\u5176\u4e4b\u4e0a\u6784\u5efa\u7684\u3001\u66f4\u4fa7\u91cd\u7edf\u8ba1\u7f8e\u5b66\u7684\u9ad8\u7ea7\u5e93\u3002<\/p>\n<h5>2.6.1 Matplotlib&#xff1a;Python\u53ef\u89c6\u5316\u7684\u57fa\u77f3<\/h5>\n<p>\u5feb\u901f\u7ed8\u56fe&#xff1a;Pyplot\u63a5\u53e3<\/p>\n<p>import matplotlib.pyplot as plt<br \/>\nx &#061; np.linspace(0, 10, 100)<br \/>\ny &#061; np.sin(x)<br \/>\nplt.plot(x, y, label&#061;&#039;sin(x)&#039;)<br \/>\nplt.xlabel(&#039;x axis&#039;)<br \/>\nplt.ylabel(&#039;y axis&#039;)<br \/>\nplt.title(&#039;A Simple Plot&#039;)<br \/>\nplt.legend()<br \/>\nplt.show()<\/p>\n<p>\u7cbe\u7ec6\u63a7\u5236&#xff1a;\u9762\u5411\u5bf9\u8c61\u63a5\u53e3<\/p>\n<p>fig, ax &#061; plt.subplots() # \u521b\u5efa\u4e00\u4e2aFigure\u5bf9\u8c61\u548c\u4e00\u4e2aAxes\u5bf9\u8c61<br \/>\nax.plot(x, y, label&#061;&#039;sin(x)&#039;)<br \/>\nax.set_xlabel(&#039;x axis&#039;)<br \/>\nax.set_ylabel(&#039;y axis&#039;)<br \/>\nax.set_title(&#039;A Simple Plot (OO Style)&#039;)<br \/>\nax.legend()<br \/>\nplt.show()<\/p>\n<p>\u9762\u5411\u5bf9\u8c61\u7684\u65b9\u5f0f\u5728\u7ed8\u5236\u590d\u6742\u56fe\u8868\u65f6\u66f4\u5177\u4f18\u52bf\u3002<\/p>\n<h5>2.6.2 Seaborn&#xff1a;\u57fa\u4e8eMatplotlib\u7684\u7edf\u8ba1\u7f8e\u5b66<\/h5>\n<p>Seaborn\u7684\u4f18\u52bf Seaborn\u7684\u4f18\u52bf\u5728\u4e8e\u5176\u7f8e\u89c2\u7684\u9ed8\u8ba4\u6837\u5f0f\u548c\u5185\u7f6e\u7684\u3001\u9762\u5411\u7edf\u8ba1\u5206\u6790\u7684\u9ad8\u7ea7\u7ed8\u56fe\u51fd\u6570\u3002<\/p>\n<p>\u5e38\u7528\u9ad8\u7ea7\u56fe\u8868<\/p>\n<p>import seaborn as sns<br \/>\n# \u4f7f\u7528pandas\u52a0\u8f7d\u7684titanic\u6570\u636e\u96c6 df<br \/>\n# \u7ed8\u5236\u4e00\u4e2a\u7bb1\u5f62\u56fe&#xff0c;\u89c2\u5bdf\u4e0d\u540c\u8239\u8231\u7b49\u7ea7\u4e58\u5ba2\u7684\u5e74\u9f84\u5206\u5e03<br \/>\nsns.boxplot(x&#061;&#039;Pclass&#039;, y&#061;&#039;Age&#039;, data&#061;df)<br \/>\nplt.title(&#039;Age Distribution by Passenger Class&#039;)<br \/>\nplt.show()<\/p>\n<p># \u7ed8\u5236\u4e00\u4e2a\u70ed\u529b\u56fe&#xff0c;\u53ef\u89c6\u5316\u6570\u503c\u7279\u5f81\u95f4\u7684\u76f8\u5173\u6027<br \/>\nnumeric_cols &#061; df.select_dtypes(include&#061;np.number)<br \/>\ncorrelation_matrix &#061; numeric_cols.corr()<br \/>\nsns.heatmap(correlation_matrix, annot&#061;True, cmap&#061;&#039;coolwarm&#039;)<br \/>\nplt.title(&#039;Correlation Matrix of Numeric Features&#039;)<br \/>\nplt.show()<\/p>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u7b2c\u4e8c\u7ae0\u201c\u5185\u529f\u5fc3\u6cd5\u201d\u7684\u5168\u90e8\u4fee\u884c\u3002\u4ece\u7ebf\u6027\u4ee3\u6570\u7684\u7a7a\u95f4\u53d8\u6362&#xff0c;\u5230\u5fae\u79ef\u5206\u7684\u4f18\u5316\u827a\u672f&#xff0c;\u518d\u5230\u6982\u7387\u8bba\u7684\u91cf\u5316\u4e0d\u786e\u5b9a\u6027&#xff0c;\u6211\u4eec\u4e3a\u7406\u89e3\u6df1\u5ea6\u5b66\u4e60\u7b97\u6cd5\u94fa\u8bbe\u4e86\u575a\u5b9e\u7684\u7406\u8bba\u57fa\u77f3\u3002\u7d27\u63a5\u7740&#xff0c;\u6211\u4eec\u4eb2\u624b\u638c\u63e1\u4e86NumPy\u3001Pandas\u3001Matplotlib\/Seaborn\u8fd9\u4e09\u5927\u7f16\u7a0b\u5229\u5668&#xff0c;\u5b83\u4eec\u662f\u5c06\u7406\u8bba\u4ed8\u8bf8\u5b9e\u8df5\u3001\u4e0e\u6570\u636e\u5bf9\u8bdd\u7684\u5f3a\u5927\u5de5\u5177\u3002<\/p>\n<p>\u8bfb\u8005\u73b0\u5728\u5e94\u6df1\u523b\u7406\u89e3&#xff0c;\u6570\u5b66\u5e76\u975e\u7b97\u6cd5\u7684\u969c\u788d&#xff0c;\u800c\u662f\u6d1e\u5bdf\u5176\u672c\u8d28\u7684\u94a5\u5319&#xff1b;\u4ee3\u7801\u4ea6\u975e\u51b0\u51b7\u7684\u6307\u4ee4&#xff0c;\u800c\u662f\u5b9e\u73b0\u521b\u60f3\u7684\u753b\u7b14\u3002\u5f53\u201c\u7406\u201d\u4e0e\u201c\u5668\u201d\u5728\u60a8\u624b\u4e2d\u878d\u4f1a\u8d2f\u901a&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u5927\u95e8\u5df2\u7136\u4e3a\u60a8\u655e\u5f00\u3002\u5e26\u7740\u8fd9\u4efd\u5185\u529f&#xff0c;\u6211\u4eec\u5c06\u5728\u540e\u7eed\u7684\u7ae0\u8282\u4e2d&#xff0c;\u5145\u6ee1\u4fe1\u5fc3\u5730\u6784\u5efa\u3001\u8bad\u7ec3\u5e76\u5256\u6790\u5404\u79cd\u590d\u6742\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u4e09\u7ae0&#xff1a;\u673a\u5668\u5b66\u4e60\u7ecf\u5178\u56de\u987e \u2014\u2014 \u6e29\u6545\u800c\u77e5\u65b0<\/h3>\n<ul>\n<li>3.1 \u673a\u5668\u5b66\u4e60\u7684\u4e09\u5927\u8303\u5f0f&#xff1a;\u95ee\u9053\u4e8e\u5929\u5730\u3002<\/li>\n<li>3.2 \u7ecf\u5178\u6a21\u578b\u5256\u6790&#xff1a;\u4e00\u82b1\u4e00\u4e16\u754c&#xff0c;\u4e00\u53f6\u4e00\u83e9\u63d0\u3002<\/li>\n<li>3.3 \u6a21\u578b\u7684\u8bc4\u4f30\u4e0e\u4f18\u5316&#xff1a;\u77e5\u5176\u7136&#xff0c;\u66f4\u8981\u77e5\u5176\u6240\u4ee5\u7136\u3002<\/li>\n<\/ul>\n<p>\u4e3a\u4f55\u8981\u56de\u987e\u7ecf\u5178&#xff1f;<\/p>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005\u670b\u53cb&#xff0c;\u5728\u60a8\u5373\u5c06\u8e0f\u5165\u6df1\u5ea6\u5b66\u4e60\u90a3\u7247\u5e7f\u88a4\u800c\u6df1\u9083\u7684\u68ee\u6797\u4e4b\u524d&#xff0c;\u6211\u4eec\u8981\u5148\u5e26\u60a8\u8d70\u8fc7\u4e00\u7247\u98ce\u666f\u79c0\u4e3d\u3001\u57fa\u77f3\u575a\u56fa\u7684\u4e18\u9675\u3002\u8fd9\u7247\u4e18\u9675&#xff0c;\u4fbf\u662f\u7531\u90a3\u4e9b\u7ecf\u5178\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u6784\u6210\u7684\u3002\u6216\u8bb8\u60a8\u4f1a\u95ee&#xff0c;\u6211\u4eec\u4e0d\u662f\u8981\u5b66\u6700\u524d\u6cbf\u7684\u6df1\u5ea6\u5b66\u4e60\u5417&#xff1f;\u4e3a\u4f55\u8981\u5728\u6b64\u201c\u9017\u7559\u201d&#xff1f;<\/p>\n<p>\u7b54\u6848\u5f88\u7b80\u5355&#xff1a;\u601d\u60f3&#xff0c;\u603b\u6709\u5176\u4f20\u627f\u3002<\/p>\n<p>\u60a8\u53ef\u77e5\u9053&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u57fa\u77f3\u2014\u2014\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u5355\u4e2a\u795e\u7ecf\u5143&#xff0c;\u5176\u601d\u60f3\u6e90\u5934\u4fbf\u662f\u672c\u7ae0\u5c06\u8981\u4ecb\u7ecd\u7684\u7ebf\u6027\u6a21\u578b\u5417&#xff1f;\u60a8\u53ef\u77e5\u9053&#xff0c;\u90a3\u4e9b\u52a8\u8f84\u62e5\u6709\u4ebf\u4e07\u53c2\u6570\u7684\u5e9e\u5927\u7f51\u7edc\u6a21\u578b&#xff0c;\u5176\u80cc\u540e\u8574\u542b\u7684\u201c\u7fa4\u4f53\u667a\u6167\u201d&#xff0c;\u65e9\u5df2\u5728\u201c\u968f\u673a\u68ee\u6797\u201d\u8fd9\u6837\u7684\u96c6\u6210\u65b9\u6cd5\u4e2d\u521d\u73b0\u7aef\u502a&#xff1f;\u7ecf\u5178\u6a21\u578b&#xff0c;\u6b63\u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u601d\u60f3\u6e90\u5934\u201d\u548c\u201c\u6d3b\u6c34\u201d\u3002\u4e0d\u6eaf\u5176\u6e90&#xff0c;\u5219\u96be\u4ee5\u77e5\u5176\u6d41\u3002<\/p>\n<p>\u5176\u6b21&#xff0c;\u6211\u4eec\u8981\u6709\u6027\u4ef7\u6bd4\u7684\u667a\u6167\u3002\u6df1\u5ea6\u5b66\u4e60\u56fa\u7136\u5f3a\u5927&#xff0c;\u4f46\u5b83\u4e5f\u50cf\u4e00\u67c4\u9700\u8981\u5de8\u5927\u80fd\u91cf\u50ac\u52a8\u7684\u201c\u5c60\u9f99\u5200\u201d\u3002\u5728\u8bb8\u591a\u73b0\u5b9e\u573a\u666f\u4e2d&#xff0c;\u95ee\u9898\u53ef\u80fd\u53ea\u662f\u4e00\u53ea\u201c\u9e21\u201d&#xff0c;\u7528\u4e00\u628a\u7cbe\u81f4\u7684\u201c\u6c34\u679c\u5200\u201d\u2014\u2014\u6bd4\u5982\u903b\u8f91\u56de\u5f52\u6216\u51b3\u7b56\u6811\u2014\u2014\u4fbf\u80fd\u4ee5\u6781\u4f4e\u7684\u8ba1\u7b97\u6210\u672c\u3001\u6781\u5feb\u7684\u65f6\u95f4\u548c\u5b8c\u7f8e\u7684\u89e3\u91ca\u6027\u6765\u89e3\u51b3\u3002\u5b66\u4f1a\u201c\u6740\u9e21\u7528\u725b\u5200\u201d\u4e4b\u524d&#xff0c;\u5fc5\u5148\u8981\u61c2\u5f97\u4f55\u65f6\u8be5\u7528\u3001\u5982\u4f55\u7528\u597d\u90a3\u628a\u201c\u6c34\u679c\u5200\u201d\u3002\u8fd9\u662f\u4e00\u79cd\u5de5\u7a0b\u5e08\u7684\u52a1\u5b9e&#xff0c;\u4e5f\u662f\u4e00\u79cd\u5b9e\u8df5\u5bb6\u7684\u667a\u6167\u3002<\/p>\n<p>\u6700\u540e&#xff0c;\u4e5f\u662f\u6700\u91cd\u8981\u7684&#xff0c;\u5b66\u4e60\u8fd9\u4e9b\u76f8\u5bf9\u7b80\u5355\u7684\u6a21\u578b&#xff0c;\u662f\u4e3a\u4e86\u5e2e\u52a9\u60a8\u5efa\u7acb\u4e00\u79cd\u5b9d\u8d35\u7684\u201c\u7b97\u6cd5\u76f4\u89c9\u201d\u3002\u901a\u8fc7\u4eb2\u624b\u5b9e\u73b0\u4e00\u4e2a\u7ebf\u6027\u56de\u5f52&#xff0c;\u60a8\u4f1a\u76f4\u89c2\u5730\u611f\u53d7\u5230\u4ec0\u4e48\u662f\u201c\u62df\u5408\u201d&#xff1b;\u901a\u8fc7\u8c03\u8bd5\u4e00\u68f5\u51b3\u7b56\u6811&#xff0c;\u60a8\u4f1a\u6df1\u523b\u5730\u7406\u89e3\u4ec0\u4e48\u662f\u201c\u8fc7\u62df\u5408\u201d&#xff1b;\u901a\u8fc7\u6bd4\u8f83\u4e0d\u540c\u7684\u8bc4\u4f30\u6307\u6807&#xff0c;\u60a8\u4f1a\u660e\u767d\u201c\u6cdb\u5316\u80fd\u529b\u201d\u7684\u771f\u6b63\u542b\u4e49\u3002\u8fd9\u4e9b\u6838\u5fc3\u6982\u5ff5&#xff0c;\u5728\u7ecf\u5178\u6a21\u578b\u4e0a\u663e\u5f97\u6e05\u6670\u800c\u5177\u4f53\u3002\u6709\u4e86\u8fd9\u4efd\u76f4\u89c9&#xff0c;\u5f53\u60a8\u672a\u6765\u9762\u5bf9\u6df1\u5ea6\u5b66\u4e60\u8fd9\u4e2a\u590d\u6742\u7684\u201c\u9ed1\u7bb1\u201d\u65f6&#xff0c;\u624d\u80fd\u4e0d\u754f\u60e7\u3001\u4e0d\u8ff7\u832b&#xff0c;\u62e5\u6709\u6d1e\u5bdf\u5176\u672c\u8d28\u7684\u773c\u5149\u3002<\/p>\n<p>\u6240\u4ee5&#xff0c;\u8bf7\u5148\u9759\u4e0b\u5fc3\u6765\u3002\u8ba9\u6211\u4eec\u4e00\u8d77&#xff0c;\u7ad9\u4e0a\u8fd9\u4e9b\u5de8\u4eba\u7684\u80a9\u8180&#xff0c;\u4e0d\u662f\u4e3a\u4e86\u505c\u7559&#xff0c;\u800c\u662f\u4e3a\u4e86\u770b\u5f97\u66f4\u8fdc\u3001\u66f4\u6e05\u3002\u8fd9\u8d9f\u201c\u6e29\u6545\u77e5\u65b0\u201d\u7684\u65c5\u7a0b&#xff0c;\u5c06\u4e3a\u60a8\u672a\u6765\u7684\u6df1\u5ea6\u5b66\u4e60\u4e4b\u8def&#xff0c;\u6253\u4e0b\u6700\u575a\u5b9e\u7684\u601d\u60f3\u5730\u57fa\u3002<\/p>\n<h4>3.1 \u673a\u5668\u5b66\u4e60\u7684\u4e09\u5927\u8303\u5f0f&#xff1a;\u95ee\u9053\u4e8e\u5929\u5730<\/h4>\n<p>\u5728\u6b63\u5f0f\u63a5\u89e6\u5177\u4f53\u7684\u7b97\u6cd5\u4e4b\u524d&#xff0c;\u8bfb\u8005\u9700\u8981\u5148\u5efa\u7acb\u4e00\u4e2a\u5b8f\u89c2\u7684\u8ba4\u77e5\u6846\u67b6\u3002\u673a\u5668\u5b66\u4e60\u6839\u636e\u5176\u5b66\u4e60\u65b9\u5f0f\u548c\u6570\u636e\u7279\u70b9&#xff0c;\u4e3b\u8981\u53ef\u4ee5\u5206\u4e3a\u4e09\u5927\u8303\u5f0f&#xff1a;\u76d1\u7763\u5b66\u4e60\u3001\u65e0\u76d1\u7763\u5b66\u4e60\u548c\u5f3a\u5316\u5b66\u4e60\u3002\u8fd9\u4e09\u79cd\u8303\u5f0f&#xff0c;\u53ef\u4ee5\u770b\u4f5c\u662f\u673a\u5668\u5411\u4e16\u754c\u201c\u95ee\u9053\u201d\u7684\u4e09\u79cd\u4e0d\u540c\u9014\u5f84&#xff0c;\u6bcf\u4e00\u79cd\u90fd\u5bf9\u5e94\u7740\u4e00\u7c7b\u72ec\u7279\u7684\u73b0\u5b9e\u95ee\u9898\u3002<\/p>\n<h5>3.1.1 \u76d1\u7763\u5b66\u4e60&#xff1a;\u6709\u5e08\u4e4b\u5b66&#xff0c;\u4f9d\u6807\u800c\u884c<\/h5>\n<p>\u6838\u5fc3\u601d\u60f3 \u76d1\u7763\u5b66\u4e60\u662f\u76ee\u524d\u5e94\u7528\u6700\u5e7f\u6cdb\u3001\u6700\u6210\u719f\u7684\u673a\u5668\u5b66\u4e60\u8303\u5f0f\u3002\u5b83\u7684\u6838\u5fc3\u601d\u60f3&#xff0c;\u662f\u4ece\u5e26\u6709\u660e\u786e\u201c\u6807\u7b7e\u201d&#xff08;Label&#xff09;\u6216\u201c\u7b54\u6848\u201d&#xff08;Answer&#xff09;\u7684\u6570\u636e\u4e2d\u8fdb\u884c\u5b66\u4e60\u3002\u8fd9\u91cc\u7684\u201c\u6807\u7b7e\u201d\u5c31\u662f\u6211\u4eec\u9884\u5148\u77e5\u9053\u7684\u3001\u6b63\u786e\u7684\u8f93\u51fa\u3002\u6574\u4e2a\u5b66\u4e60\u8fc7\u7a0b&#xff0c;\u5c31\u5982\u540c\u6709\u4e00\u4f4d\u65e0\u6240\u4e0d\u77e5\u7684\u8001\u5e08&#xff08;\u5373\u6807\u7b7e\u6570\u636e&#xff09;\u5728\u65c1\u8fb9\u8fdb\u884c\u6307\u5bfc&#xff0c;\u6bcf\u5f53\u6a21\u578b\u505a\u51fa\u4e00\u6b21\u9884\u6d4b&#xff0c;\u8001\u5e08\u5c31\u4f1a\u544a\u8bc9\u5b83\u6b63\u786e\u7b54\u6848\u662f\u4ec0\u4e48&#xff0c;\u6a21\u578b\u5219\u6839\u636e\u8fd9\u6b21\u53cd\u9988\u6765\u4fee\u6b63\u81ea\u5df1&#xff0c;\u529b\u6c42\u4e0b\u6b21\u505a\u5f97\u66f4\u597d\u3002<\/p>\n<p>\u56e0\u6b64&#xff0c;\u201c\u76d1\u7763\u201d\u4e8c\u5b57\u7684\u7cbe\u9ad3\u5728\u4e8e&#xff0c;\u5b66\u4e60\u7684\u6bcf\u4e00\u6b65\u90fd\u6709\u6765\u81ea\u201c\u6b63\u786e\u7b54\u6848\u201d\u7684\u76f4\u63a5\u53cd\u9988\u3002<\/p>\n<p>\u4e24\u5927\u4efb\u52a1 \u6839\u636e\u6807\u7b7e\u7c7b\u578b\u7684\u4e0d\u540c&#xff0c;\u76d1\u7763\u5b66\u4e60\u4e3b\u8981\u89e3\u51b3\u4e24\u7c7b\u95ee\u9898&#xff1a;<\/p>\n<p>\u56de\u5f52&#xff08;Regression&#xff09;&#xff1a;\u5f53\u6211\u4eec\u7684\u76ee\u6807\u662f\u9884\u6d4b\u4e00\u4e2a\u8fde\u7eed\u7684\u6570\u503c\u65f6&#xff0c;\u8fd9\u7c7b\u95ee\u9898\u5c31\u88ab\u79f0\u4e3a\u56de\u5f52\u3002\u9884\u6d4b\u503c\u7684\u8303\u56f4\u662f\u8fde\u7eed\u7684&#xff0c;\u53ef\u4ee5\u53d6\u4efb\u610f\u5b9e\u6570\u3002<\/p>\n<ul>\n<li>\u8b6c\u5982&#xff1a;\u6839\u636e\u623f\u5c4b\u7684\u9762\u79ef\u3001\u4f4d\u7f6e\u3001\u623f\u9f84\u7b49\u7279\u5f81&#xff0c;\u9884\u6d4b\u5176\u5e02\u573a\u4ef7\u683c&#xff08;\u4e00\u4e2a\u8fde\u7eed\u7684\u91d1\u989d&#xff09;&#xff1b;\u6839\u636e\u5386\u53f2\u6c14\u8c61\u6570\u636e&#xff0c;\u9884\u6d4b\u660e\u5929\u6700\u9ad8\u6c14\u6e29&#xff08;\u4e00\u4e2a\u8fde\u7eed\u7684\u6e29\u5ea6\u503c&#xff09;\u3002<\/li>\n<\/ul>\n<p>\u5206\u7c7b&#xff08;Classification&#xff09;&#xff1a;\u5f53\u6211\u4eec\u7684\u76ee\u6807\u662f\u9884\u6d4b\u4e00\u4e2a\u79bb\u6563\u7684\u7c7b\u522b\u65f6&#xff0c;\u8fd9\u7c7b\u95ee\u9898\u5c31\u662f\u5206\u7c7b\u3002\u9884\u6d4b\u7684\u7ed3\u679c\u662f\u6709\u9650\u4e2a\u7c7b\u522b\u4e2d\u7684\u4e00\u4e2a\u3002<\/p>\n<ul>\n<li>\u8b6c\u5982&#xff1a;\u6839\u636e\u90ae\u4ef6\u7684\u5185\u5bb9\u548c\u53d1\u4ef6\u4eba\u4fe1\u606f&#xff0c;\u5224\u65ad\u8be5\u90ae\u4ef6\u662f**\u201c\u5783\u573e\u90ae\u4ef6\u201d\u8fd8\u662f\u201c\u975e\u5783\u573e\u90ae\u4ef6\u201d&#xff08;\u4e8c\u5206\u7c7b\u95ee\u9898&#xff09;&#xff1b;\u6839\u636e\u4e00\u5f20\u52a8\u7269\u56fe\u7247&#xff0c;\u8bc6\u522b\u51fa\u5b83\u662f\u201c\u732b\u201d\u3001\u201c\u72d7\u201d\u8fd8\u662f\u201c\u9e1f\u201d**&#xff08;\u591a\u5206\u7c7b\u95ee\u9898&#xff09;\u3002<\/li>\n<\/ul>\n<p>\u5b66\u4e60\u7684\u672c\u8d28 \u4ece\u6570\u5b66\u7684\u89d2\u5ea6\u770b&#xff0c;\u76d1\u7763\u5b66\u4e60\u7684\u672c\u8d28&#xff0c;\u5c31\u662f\u5bfb\u627e\u4e00\u4e2a\u6700\u4f18\u7684\u6620\u5c04\u51fd\u6570 f\u3002\u8fd9\u4e2a\u51fd\u6570\u80fd\u591f\u5efa\u7acb\u8d77\u4ece\u8f93\u5165\u7279\u5f81 X \u5230\u8f93\u51fa\u6807\u7b7e Y \u4e4b\u95f4\u7684\u7a33\u5b9a\u5173\u7cfb&#xff0c;\u5373 Y \u2248 f(X)\u3002\u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u901a\u8fc7\u6d77\u91cf\u7684\u5df2\u77e5 (X, Y) \u6570\u636e\u5bf9&#xff0c;\u6765\u4e0d\u65ad\u8c03\u6574\u51fd\u6570 f \u7684\u5185\u90e8\u53c2\u6570&#xff0c;\u4f7f\u5176\u5c3d\u53ef\u80fd\u5730\u903c\u8fd1\u8fd9\u4e2a\u771f\u5b9e\u5b58\u5728\u7684\u3001\u4f46\u6211\u4eec\u672a\u77e5\u7684\u6f5c\u5728\u6620\u5c04\u89c4\u5f8b\u3002<\/p>\n<h5>3.1.2 \u65e0\u76d1\u7763\u5b66\u4e60&#xff1a;\u65e0\u5e08\u4e4b\u5b66&#xff0c;\u89c2\u7269\u81ea\u7701<\/h5>\n<p>\u6838\u5fc3\u601d\u60f3 \u4e0e\u76d1\u7763\u5b66\u4e60\u622a\u7136\u76f8\u53cd&#xff0c;\u65e0\u76d1\u7763\u5b66\u4e60\u6240\u9762\u5bf9\u7684\u6570\u636e\u662f**\u5b8c\u5168\u6ca1\u6709\u201c\u6807\u7b7e\u201d**\u7684\u3002\u8fd9\u610f\u5473\u7740\u6ca1\u6709\u73b0\u6210\u7684\u201c\u6b63\u786e\u7b54\u6848\u201d\u53ef\u4f9b\u53c2\u8003\u3002\u5b66\u4e60\u7684\u76ee\u6807&#xff0c;\u662f\u5728\u6ca1\u6709\u5916\u90e8\u6307\u5bfc\u7684\u60c5\u51b5\u4e0b&#xff0c;\u4ec5\u4ec5\u4f9d\u9760\u6570\u636e\u81ea\u8eab&#xff0c;\u53bb\u53d1\u73b0\u5176\u4e2d\u9690\u85cf\u7684\u7ed3\u6784\u3001\u6a21\u5f0f\u6216\u5185\u5728\u89c4\u5f8b\u3002<\/p>\n<p>\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u597d\u6bd4\u4e00\u4f4d\u53e4\u4ee3\u7684\u5929\u6587\u5b66\u5bb6&#xff0c;\u4ed6\u6240\u62e5\u6709\u7684\u53ea\u662f\u6ee1\u5929\u7e41\u661f\u7684\u4f4d\u7f6e\u6570\u636e&#xff08;\u6ca1\u6709\u6807\u7b7e\u7684\u8f93\u5165 X&#xff09;&#xff0c;\u901a\u8fc7\u65e5\u590d\u4e00\u65e5\u7684\u89c2\u5bdf\u4e0e\u601d\u8003&#xff0c;\u4ed6\u81ea\u5df1\u53d1\u73b0\u4e86\u67d0\u4e9b\u661f\u661f\u7ec4\u5408\u8d77\u6765&#xff0c;\u5f62\u6001\u4e0a\u5f88\u76f8\u4f3c&#xff0c;\u4e8e\u662f\u5c06\u5b83\u4eec\u547d\u540d\u4e3a\u201c\u730e\u6237\u5ea7\u201d\u3001\u201c\u5317\u6597\u4e03\u661f\u201d\u7b49&#xff08;\u5373\u805a\u7c7b&#xff09;\u3002\u4ed6\u662f\u5728\u6570\u636e\u4e2d\u81ea\u6211\u53d1\u73b0\u4e86\u89c4\u5f8b\u3002<\/p>\n<p>\u5178\u578b\u4efb\u52a1 \u65e0\u76d1\u7763\u5b66\u4e60\u7684\u5e94\u7528\u573a\u666f\u540c\u6837\u5e7f\u6cdb&#xff0c;\u4e3b\u8981\u5305\u62ec&#xff1a;<\/p>\n<p>\u805a\u7c7b&#xff08;Clustering&#xff09;&#xff1a;\u5c06\u6570\u636e\u96c6\u4e2d\u7684\u6837\u672c&#xff0c;\u6839\u636e\u5b83\u4eec\u5185\u5728\u7684\u76f8\u4f3c\u6027&#xff0c;\u81ea\u52a8\u5730\u5212\u5206\u4e3a\u82e5\u5e72\u4e2a\u7c07&#xff08;Cluster&#xff09;\u3002\u76ee\u6807\u662f\u8ba9\u540c\u4e00\u7c07\u5185\u7684\u6570\u636e\u70b9\u5c3d\u53ef\u80fd\u76f8\u4f3c&#xff0c;\u4e0d\u540c\u7c07\u4e4b\u95f4\u7684\u6570\u636e\u70b9\u5c3d\u53ef\u80fd\u76f8\u5f02\u3002<\/p>\n<ul>\n<li>\u8b6c\u5982&#xff1a;\u7535\u5546\u5e73\u53f0\u6839\u636e\u7528\u6237\u7684\u8d2d\u4e70\u5386\u53f2\u3001\u6d4f\u89c8\u884c\u4e3a\u7b49\u6570\u636e&#xff0c;\u5c06\u5e9e\u5927\u7684\u5ba2\u6237\u7fa4\u4f53\u81ea\u52a8\u5212\u5206\u4e3a**\u201c\u9ad8\u4ef7\u503c\u5ba2\u6237\u201d\u3001\u201c\u6f5c\u529b\u5ba2\u6237\u201d\u3001\u201c\u5f85\u6fc0\u6d3b\u5ba2\u6237\u201d**\u7b49\u4e0d\u540c\u7684\u7fa4\u4f53&#xff0c;\u4ee5\u4fbf\u5b9e\u65bd\u7cbe\u51c6\u8425\u9500\u3002<\/li>\n<\/ul>\n<p>\u964d\u7ef4&#xff08;Dimensionality Reduction&#xff09;&#xff1a;\u5728\u5c3d\u53ef\u80fd\u4fdd\u7559\u539f\u59cb\u6570\u636e\u4e3b\u8981\u4fe1\u606f\u7684\u524d\u63d0\u4e0b&#xff0c;\u7528\u66f4\u5c11\u7684\u7279\u5f81\u6765\u8868\u793a\u6570\u636e\u3002\u8fd9\u6709\u52a9\u4e8e\u6570\u636e\u53ef\u89c6\u5316\u3001\u53bb\u9664\u566a\u58f0\u3001\u4ee5\u53ca\u63d0\u5347\u540e\u7eed\u5176\u4ed6\u5b66\u4e60\u7b97\u6cd5\u7684\u6548\u7387\u3002<\/p>\n<ul>\n<li>\u8b6c\u5982&#xff1a;\u8861\u91cf\u4e00\u4e2a\u5730\u533a\u5b8f\u89c2\u7ecf\u6d4e\u72b6\u51b5\u7684\u6307\u6807\u53ef\u80fd\u6709\u4e0a\u767e\u4e2a&#xff08;GDP\u3001CPI\u3001PMI\u3001\u8fdb\u51fa\u53e3\u989d\u7b49&#xff09;&#xff0c;\u8fd9\u4e9b\u6307\u6807\u95f4\u53ef\u80fd\u5b58\u5728\u9ad8\u5ea6\u76f8\u5173\u6027\u3002\u964d\u7ef4\u6280\u672f\u53ef\u4ee5\u5c06\u8fd9\u4e0a\u767e\u4e2a\u6307\u6807\u538b\u7f29\u4e3a\u51e0\u4e2a\u6838\u5fc3\u7684\u201c\u7ecf\u6d4e\u666f\u6c14\u6307\u6570\u201d&#xff0c;\u6293\u4f4f\u4e3b\u8981\u77db\u76fe\u3002<\/li>\n<\/ul>\n<p>\u5b66\u4e60\u7684\u672c\u8d28 \u65e0\u76d1\u7763\u5b66\u4e60\u7684\u672c\u8d28&#xff0c;\u4e0d\u662f\u53bb\u62df\u5408\u4e00\u4e2a\u8f93\u5165\u5230\u8f93\u51fa\u7684\u6620\u5c04&#xff0c;\u800c\u662f\u6df1\u5165\u63a2\u7d22\u6570\u636e X \u81ea\u8eab\u5185\u5728\u7684\u5206\u5e03\u3001\u5173\u8054\u4e0e\u7ed3\u6784\u3002\u5b83\u8bd5\u56fe\u56de\u7b54\u7684\u95ee\u9898\u662f&#xff1a;\u201c\u8fd9\u5806\u6570\u636e\u672c\u8eab&#xff0c;\u53ef\u4ee5\u88ab\u5982\u4f55\u7ec4\u7ec7\u548c\u7406\u89e3&#xff1f;\u201d<\/p>\n<h5>3.1.3 \u5f3a\u5316\u5b66\u4e60&#xff1a;\u884c\u4e07\u91cc\u8def&#xff0c;\u52a8\u4e2d\u89c9\u609f<\/h5>\n<p>\u6838\u5fc3\u601d\u60f3 \u5f3a\u5316\u5b66\u4e60\u662f\u4e09\u5927\u8303\u5f0f\u4e2d\u6700\u5177\u201c\u884c\u52a8\u667a\u6167\u201d\u7684\u4e00\u79cd\u3002\u5b83\u5173\u6ce8\u7684\u662f\u4e00\u4e2a\u667a\u80fd\u4f53&#xff08;Agent&#xff09;\u5982\u4f55\u5728\u4e00\u4e2a\u590d\u6742\u7684\u3001\u52a8\u6001\u7684\u73af\u5883&#xff08;Environment&#xff09;\u4e2d&#xff0c;\u901a\u8fc7\u4e0e\u73af\u5883\u7684\u4e92\u52a8\u6765\u5b66\u4e60\u5982\u4f55\u505a\u51fa\u6700\u4f18\u51b3\u7b56&#xff0c;\u4ee5\u8fbe\u6210\u4e00\u4e2a\u957f\u8fdc\u7684\u76ee\u6807\u3002<\/p>\n<p>\u5b83\u7684\u5b66\u4e60\u65b9\u5f0f\u65e2\u975e\u4f9d\u8d56\u6807\u7b7e&#xff0c;\u4e5f\u975e\u7eaf\u7cb9\u89c2\u5bdf&#xff0c;\u800c\u662f**\u201c\u8bd5\u9519\u201d&#xff08;Trial and Error&#xff09;\u3002\u667a\u80fd\u4f53\u505a\u51fa\u4e00\u4e2a\u52a8\u4f5c&#xff08;Action&#xff09;&#xff0c;\u73af\u5883\u4f1a\u56e0\u6b64\u53d1\u751f\u6539\u53d8\u5e76\u53cd\u9988\u7ed9\u667a\u80fd\u4f53\u4e00\u4e2a\u5956\u52b1&#xff08;Reward&#xff09;\u6216\u60e9\u7f5a&#xff08;Punishment&#xff09;\u4fe1\u53f7\u3002\u667a\u80fd\u4f53\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u5b66\u4e60\u4e00\u4e2a\u6700\u4f18\u7684\u7b56\u7565&#xff08;Policy&#xff09;\u2014\u2014\u5373\u5728\u4f55\u79cd\u72b6\u6001&#xff08;State&#xff09;\u4e0b\u5e94\u8be5\u91c7\u53d6\u4f55\u79cd\u52a8\u4f5c\u2014\u2014\u6765\u6700\u5927\u5316\u5b83\u5728\u672a\u6765\u80fd\u591f\u83b7\u5f97\u7684\u7d2f\u79ef\u5956\u52b1**\u3002<\/p>\n<p>\u6838\u5fc3\u8981\u7d20 \u4e00\u4e2a\u5f3a\u5316\u5b66\u4e60\u95ee\u9898&#xff0c;\u901a\u5e38\u7531\u4ee5\u4e0b\u51e0\u4e2a\u6838\u5fc3\u8981\u7d20\u6784\u6210&#xff1a;<\/p>\n<ul>\n<li>\u667a\u80fd\u4f53&#xff08;Agent&#xff09;&#xff1a;\u5b66\u4e60\u8005\u548c\u51b3\u7b56\u8005\u3002<\/li>\n<li>\u73af\u5883&#xff08;Environment&#xff09;&#xff1a;\u667a\u80fd\u4f53\u5916\u90e8\u7684\u4e16\u754c&#xff0c;\u4e0e\u667a\u80fd\u4f53\u8fdb\u884c\u4ea4\u4e92\u3002<\/li>\n<li>\u72b6\u6001&#xff08;State&#xff09;&#xff1a;\u5bf9\u5f53\u524d\u73af\u5883\u7684\u4e00\u4e2a\u63cf\u8ff0\u3002<\/li>\n<li>\u52a8\u4f5c&#xff08;Action&#xff09;&#xff1a;\u667a\u80fd\u4f53\u53ef\u4ee5\u91c7\u53d6\u7684\u884c\u4e3a\u3002<\/li>\n<li>\u5956\u52b1&#xff08;Reward&#xff09;&#xff1a;\u73af\u5883\u5bf9\u667a\u80fd\u4f53\u4e0a\u4e00\u6b65\u52a8\u4f5c\u7684\u5373\u65f6\u53cd\u9988\u4fe1\u53f7\u3002<\/li>\n<\/ul>\n<p>\u5b66\u4e60\u7684\u9690\u55bb \u5f3a\u5316\u5b66\u4e60\u7684\u8fc7\u7a0b&#xff0c;\u4e0e\u751f\u7269\u5b66\u4e60\u65b0\u6280\u80fd\u7684\u8fc7\u7a0b\u6781\u4e3a\u76f8\u4f3c\u3002<\/p>\n<p>\u8b6c\u5982&#xff0c;\u5a74\u513f\u5b66\u8d70\u8def\u3002\u5a74\u513f&#xff08;\u667a\u80fd\u4f53&#xff09;\u5728\u5ba2\u5385&#xff08;\u73af\u5883&#xff09;\u4e2d&#xff0c;\u4ed6\u5f53\u524d\u7684\u59ff\u52bf\u548c\u4f4d\u7f6e\u662f&#xff08;\u72b6\u6001&#xff09;\u3002\u4ed6\u5c1d\u8bd5\u8fc8\u51fa\u4e00\u6b65&#xff08;\u52a8\u4f5c&#xff09;\u3002\u5982\u679c\u6210\u529f\u524d\u8fdb\u4e86&#xff0c;\u7236\u6bcd\u7684\u9f13\u52b1\u5c31\u662f&#xff08;\u6b63\u5956\u52b1&#xff09;&#xff1b;\u5982\u679c\u4e0d\u5e78\u6454\u5012\u4e86&#xff0c;\u75bc\u75db\u611f\u5c31\u662f&#xff08;\u8d1f\u5956\u52b1\/\u60e9\u7f5a&#xff09;\u3002\u901a\u8fc7\u65e0\u6570\u6b21\u7684\u5c1d\u8bd5&#xff0c;\u5a74\u513f\u7684\u5927\u8111\u9010\u6e10\u5b66\u4f1a\u4e86\u5728\u5404\u79cd\u72b6\u6001\u4e0b&#xff0c;\u5982\u4f55\u534f\u8c03\u808c\u8089\u505a\u51fa\u6b63\u786e\u7684\u52a8\u4f5c&#xff0c;\u4ee5\u5b9e\u73b0\u201c\u6301\u7eed\u884c\u8d70\u4e0d\u6454\u5012\u201d\u8fd9\u4e2a\u957f\u671f\u76ee\u6807\u3002<\/p>\n<p>\u5728\u6280\u672f\u9886\u57df&#xff0c;\u5f3a\u5316\u5b66\u4e60\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u8bad\u7ec3AI\u4e0b\u68cb&#xff08;AlphaGo&#xff09;\u3001\u73a9\u7535\u5b50\u6e38\u620f\u3001\u63a7\u5236\u673a\u5668\u4eba\u884c\u8d70\u3001\u4f18\u5316\u4ea4\u901a\u4fe1\u53f7\u706f\u8c03\u5ea6\u7b49\u52a8\u6001\u51b3\u7b56\u95ee\u9898\u3002<\/p>\n<hr \/>\n<p>\u597d\u4e86&#xff0c;\u8bfb\u8005\u670b\u53cb\u4eec&#xff0c;\u901a\u8fc7\u4ee5\u4e0a\u4ecb\u7ecd&#xff0c;\u76f8\u4fe1\u5927\u5bb6\u5bf9\u673a\u5668\u5b66\u4e60\u7684\u4e09\u5927\u8303\u5f0f\u5df2\u7ecf\u6709\u4e86\u4e00\u4e2a\u6e05\u6670\u7684\u8ba4\u8bc6\u3002\u5b83\u4eec\u5206\u522b\u4ece\u201c\u6709\u5e08\u4e4b\u5b66\u201d\u3001\u201c\u65e0\u5e08\u4e4b\u5b66\u201d\u548c\u201c\u52a8\u4e2d\u89c9\u609f\u201d\u4e09\u4e2a\u7ef4\u5ea6&#xff0c;\u4e3a\u673a\u5668\u8d4b\u4e88\u4e86\u5b66\u4e60\u7684\u80fd\u529b\u3002\u63a5\u4e0b\u6765&#xff0c;\u6211\u4eec\u5c06\u6df1\u5165\u5230\u76d1\u7763\u5b66\u4e60\u7684\u9635\u8425\u4e2d&#xff0c;\u4ece\u6700\u7ecf\u5178\u3001\u6700\u57fa\u7840\u7684\u6a21\u578b\u5f00\u59cb&#xff0c;\u5256\u6790\u5b83\u4eec\u7684\u5185\u5728\u673a\u7406\u3002<\/p>\n<h4>3.2 \u7ecf\u5178\u6a21\u578b\u5256\u6790&#xff1a;\u4e00\u82b1\u4e00\u4e16\u754c&#xff0c;\u4e00\u53f6\u4e00\u83e9\u63d0<\/h4>\n<p>\u672c\u8282\u5c06\u8981\u4ecb\u7ecd\u7684\u51e0\u4e2a\u7ecf\u5178\u6a21\u578b&#xff0c;\u5404\u81ea\u90fd\u4ee3\u8868\u4e86\u4e00\u79cd\u6838\u5fc3\u7684\u5efa\u6a21\u601d\u60f3\u3002\u5b83\u4eec\u5c31\u50cf\u51e0\u6247\u4e0d\u540c\u7684\u7a97\u6237&#xff0c;\u8ba9\u6211\u4eec\u4ece\u4e0d\u540c\u7684\u89d2\u5ea6\u7aa5\u89c1\u673a\u5668\u5b66\u4e60\u89e3\u51b3\u95ee\u9898\u7684\u667a\u6167\u3002\u6bcf\u4e00\u4e2a\u6a21\u578b&#xff0c;\u90fd\u662f\u4e00\u4e2a\u81ea\u6210\u4f53\u7cfb\u7684\u5c0f\u4e16\u754c\u3002<\/p>\n<h5>3.2.1 \u7ebf\u6027\u56de\u5f52\u4e0e\u903b\u8f91\u56de\u5f52&#xff1a;\u4ece\u7ebf\u6027\u5230\u975e\u7ebf\u6027\u7684\u6865\u6881<\/h5>\n<p>\u7ebf\u6027\u6a21\u578b\u662f\u673a\u5668\u5b66\u4e60\u4e16\u754c\u4e2d\u6700\u57fa\u7840\u3001\u6700\u91cd\u8981\u7684\u6a21\u578b\u5bb6\u65cf\u3002\u5b83\u4eec\u4ee5\u5176\u7b80\u6d01\u3001\u9ad8\u6548\u548c\u9ad8\u5ea6\u7684\u53ef\u89e3\u91ca\u6027&#xff0c;\u6210\u4e3a\u4e86\u8bb8\u591a\u590d\u6742\u7b97\u6cd5\u7684\u57fa\u77f3\u3002<\/p>\n<p>\u7ebf\u6027\u56de\u5f52&#xff08;Linear Regression&#xff09;&#xff1a;\u5927\u9053\u81f3\u7b80&#xff0c;\u76f4\u6765\u76f4\u5f80<\/p>\n<p>\u6a21\u578b\u54f2\u5b66 \u7ebf\u6027\u56de\u5f52\u7684\u80cc\u540e&#xff0c;\u662f\u4e00\u79cd\u6734\u7d20\u800c\u5f3a\u5927\u7684\u4e16\u754c\u89c2&#xff1a;\u5b83\u76f8\u4fe1\u4e16\u95f4\u4e07\u7269\u7684\u8bb8\u591a\u8054\u7cfb\u662f\u7b80\u5355\u7684\u3001\u7ebf\u6027\u7684\u3002\u5b83\u8bd5\u56fe\u7528\u4e00\u6761\u7b14\u76f4\u7684\u7ebf&#xff08;\u5728\u4e00\u7ef4\u7a7a\u95f4\u4e2d&#xff09;\u6216\u4e00\u4e2a\u5e73\u6574\u7684\u9762&#xff08;\u5728\u591a\u7ef4\u7a7a\u95f4\u4e2d&#xff09;&#xff0c;\u6765\u63cf\u8ff0\u8f93\u5165\u7279\u5f81\u4e0e\u8f93\u51fa\u503c\u4e4b\u95f4\u7684\u5173\u7cfb\u3002\u8fd9\u79cd\u201c\u76f4\u6765\u76f4\u5f80\u201d\u7684\u5047\u8bbe&#xff0c;\u867d\u7136\u7b80\u5355&#xff0c;\u5374\u60ca\u4eba\u5730\u6709\u6548&#xff0c;\u80fd\u591f\u6293\u4f4f\u8bb8\u591a\u73b0\u5b9e\u95ee\u9898\u7684\u6838\u5fc3\u8d8b\u52bf\u3002<\/p>\n<p>\u6570\u5b66\u5f62\u5f0f \u5bf9\u4e8e\u4e00\u4e2a\u62e5\u6709 n \u4e2a\u7279\u5f81\u7684\u6837\u672c x &#061; (x\u2081, x\u2082, &#8230;, x\u2099)&#xff0c;\u7ebf\u6027\u56de\u5f52\u6a21\u578b\u9884\u6d4b\u5176\u8f93\u51fa \u0177 (\u8bfb\u4f5c y-hat) \u7684\u516c\u5f0f\u4e3a&#xff1a; \u0177 &#061; w\u2081x\u2081 &#043; w\u2082x\u2082 &#043; &#8230; &#043; w\u2099x\u2099 &#043; b \u8fd9\u4e2a\u516c\u5f0f\u53ef\u4ee5\u7528\u66f4\u7b80\u6d01\u7684\u5411\u91cf\u5f62\u5f0f\u8868\u793a&#xff1a;\u0177 &#061; w\u1d40x &#043; b<\/p>\n<ul>\n<li>\u6743\u91cd&#xff08;weights&#xff09;w&#xff1a;\u5411\u91cf\u00a0w &#061; (w\u2081, w\u2082, &#8230;, w\u2099)\u00a0\u4e2d\u7684\u6bcf\u4e00\u4e2a\u5143\u7d20\u00a0w\u1d62&#xff0c;\u4ee3\u8868\u4e86\u7b2c\u00a0i\u00a0\u4e2a\u7279\u5f81\u00a0x\u1d62\u00a0\u5bf9\u6700\u7ec8\u9884\u6d4b\u7ed3\u679c\u7684\u5f71\u54cd\u529b\u3002w\u1d62\u00a0\u7684\u7edd\u5bf9\u503c\u8d8a\u5927&#xff0c;\u8bf4\u660e\u8be5\u7279\u5f81\u8d8a\u91cd\u8981\u3002<\/li>\n<li>\u504f\u7f6e&#xff08;bias&#xff09;b&#xff1a;\u5b83\u662f\u4e00\u4e2a\u622a\u8ddd\u9879&#xff0c;\u4ee3\u8868\u4e86\u5728\u4e0d\u8003\u8651\u4efb\u4f55\u8f93\u5165\u7279\u5f81\u7684\u60c5\u51b5\u4e0b&#xff0c;\u9884\u6d4b\u503c\u7684\u57fa\u51c6\u6c34\u5e73\u3002<\/li>\n<\/ul>\n<p>\u635f\u5931\u51fd\u6570&#xff1a;\u6700\u5c0f\u4e8c\u4e58\u6cd5 \u6a21\u578b\u5982\u4f55\u77e5\u9053\u54ea\u6761\u76f4\u7ebf\u624d\u662f\u201c\u6700\u4f73\u201d\u7684\u5462&#xff1f;\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u6807\u51c6\u6765\u8861\u91cf\u201c\u597d\u201d\u4e0e\u201c\u574f\u201d\u3002\u5728\u7ebf\u6027\u56de\u5f52\u4e2d&#xff0c;\u8fd9\u4e2a\u6807\u51c6\u5c31\u662f\u5747\u65b9\u8bef\u5dee&#xff08;Mean Squared Error, MSE&#xff09;&#xff0c;\u4e5f\u79f0\u4e3a\u6700\u5c0f\u4e8c\u4e58\u6cd5\u3002 \u5b83\u7684\u601d\u60f3\u76f4\u89c2\u800c\u4f18\u7f8e&#xff1a;\u5bf9\u4e8e\u6bcf\u4e00\u4e2a\u8bad\u7ec3\u6837\u672c&#xff0c;\u6a21\u578b\u90fd\u6709\u4e00\u4e2a\u9884\u6d4b\u503c \u0177 \u548c\u4e00\u4e2a\u771f\u5b9e\u503c y\u3002\u5b83\u4eec\u4e4b\u95f4\u7684\u5dee\u8ddd (y &#8211; \u0177) \u5c31\u662f\u9884\u6d4b\u8bef\u5dee\u3002\u6211\u4eec\u5c06\u6240\u6709\u6837\u672c\u7684\u8fd9\u4e2a\u8bef\u5dee\u8fdb\u884c\u5e73\u65b9&#xff08;\u4ee5\u6d88\u9664\u6b63\u8d1f\u53f7\u5e76\u653e\u5927\u8f83\u5927\u8bef\u5dee\u7684\u5f71\u54cd&#xff09;&#xff0c;\u7136\u540e\u6c42\u5176\u5e73\u5747\u503c\u3002 \u51e0\u4f55\u610f\u4e49&#xff1a;\u6700\u5c0f\u5316\u5747\u65b9\u8bef\u5dee&#xff0c;\u7b49\u4ef7\u4e8e\u5bfb\u627e\u4e00\u6761\u76f4\u7ebf&#xff0c;\u4f7f\u5f97\u6240\u6709\u6570\u636e\u70b9\u5230\u8fd9\u6761\u76f4\u7ebf\u7684\u7ad6\u76f4\u8ddd\u79bb\u7684\u5e73\u65b9\u548c\u6700\u5c0f\u3002\u8fd9\u6761\u76f4\u7ebf&#xff0c;\u5c31\u662f\u5bf9\u6570\u636e\u62df\u5408\u5f97\u6700\u597d\u7684\u76f4\u7ebf\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218 \u5728Python\u4e2d&#xff0c;Scikit-Learn\u5e93\u4e3a\u6211\u4eec\u5b9e\u73b0\u7ebf\u6027\u56de\u5f52\u63d0\u4f9b\u4e86\u6781\u5927\u7684\u4fbf\u5229\u3002\u4e0b\u9762\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u793a\u4f8b&#xff0c;\u6211\u4eec\u5c06\u4f7f\u7528\u5b83\u6765\u9884\u6d4b\u4e00\u4e2a\u57fa\u4e8e\u201c\u5b66\u4e60\u5c0f\u65f6\u6570\u201d\u7684\u201c\u8003\u8bd5\u5206\u6570\u201d\u3002<\/p>\n<p># \u5bfc\u5165\u6240\u9700\u5e93<br \/>\nimport numpy as np<br \/>\nfrom sklearn.linear_model import LinearRegression<br \/>\nimport matplotlib.pyplot as plt<\/p>\n<p># 1. \u51c6\u5907\u6570\u636e (X: \u5b66\u4e60\u5c0f\u65f6\u6570, y: \u8003\u8bd5\u5206\u6570)<br \/>\nX &#061; np.array([[2], [4], [5], [6], [8], [10]])<br \/>\ny &#061; np.array([55, 60, 68, 75, 85, 95])<\/p>\n<p># 2. \u521b\u5efa\u5e76\u8bad\u7ec3\u6a21\u578b<br \/>\nmodel &#061; LinearRegression()<br \/>\nmodel.fit(X, y)<\/p>\n<p># 3. \u67e5\u770b\u6a21\u578b\u53c2\u6570<br \/>\nw &#061; model.coef_[0]  # \u6743\u91cd w<br \/>\nb &#061; model.intercept_ # \u504f\u7f6e b<br \/>\nprint(f&#034;\u5b66\u4e60\u5230\u7684\u6a21\u578b: \u5206\u6570 &#061; {w:.2f} * \u5c0f\u65f6\u6570 &#043; {b:.2f}&#034;)<\/p>\n<p># 4. \u8fdb\u884c\u9884\u6d4b\u5e76\u53ef\u89c6\u5316<br \/>\nplt.scatter(X, y, color&#061;&#039;blue&#039;, label&#061;&#039;\u771f\u5b9e\u6570\u636e&#039;) # \u7ed8\u5236\u539f\u59cb\u6570\u636e\u70b9<br \/>\nplt.plot(X, model.predict(X), color&#061;&#039;red&#039;, label&#061;&#039;\u56de\u5f52\u76f4\u7ebf&#039;) # \u7ed8\u5236\u62df\u5408\u7684\u76f4\u7ebf<br \/>\nplt.xlabel(&#039;\u5b66\u4e60\u5c0f\u65f6\u6570&#039;)<br \/>\nplt.ylabel(&#039;\u8003\u8bd5\u5206\u6570&#039;)<br \/>\nplt.legend()<br \/>\nplt.show()<\/p>\n<p>\u901a\u8fc7\u8fd9\u6bb5\u4ee3\u7801&#xff0c;\u8bfb\u8005\u53ef\u4ee5\u4eb2\u773c\u770b\u5230\u4e00\u6761\u7ea2\u8272\u7684\u76f4\u7ebf&#xff0c;\u5b83\u4ee5\u201c\u6700\u5c0f\u4e8c\u4e58\u201d\u7684\u51c6\u5219&#xff0c;\u4f18\u96c5\u5730\u7a7f\u8fc7\u4e86\u84dd\u8272\u7684\u6570\u636e\u70b9&#xff0c;\u63ed\u793a\u4e86\u5b66\u4e60\u65f6\u95f4\u4e0e\u5206\u6570\u4e4b\u95f4\u7684\u7ebf\u6027\u5173\u7cfb\u3002<\/p>\n<p>\u903b\u8f91\u56de\u5f52&#xff08;Logistic Regression&#xff09;&#xff1a;\u5728\u8fb9\u754c\u5904\u4f18\u96c5\u8f6c\u8eab<\/p>\n<p>\u4e3a\u4f55\u9700\u8981 \u5982\u679c\u6211\u4eec\u60f3\u9884\u6d4b\u7684\u4e0d\u662f\u8fde\u7eed\u7684\u5206\u6570&#xff0c;\u800c\u662f\u201c\u53ca\u683c\u201d\u4e0e\u201c\u4e0d\u53ca\u683c\u201d\u8fd9\u4e24\u4e2a\u7c7b\u522b\u5462&#xff1f;\u7ebf\u6027\u56de\u5f52\u7684\u8f93\u51fa\u662f (-\u221e, &#043;\u221e) \u7684\u8fde\u7eed\u503c&#xff0c;\u65e0\u6cd5\u76f4\u63a5\u7528\u4e8e\u5206\u7c7b\u3002\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u6a21\u578b&#xff0c;\u5b83\u80fd\u8f93\u51fa\u4e00\u4e2a\u8868\u793a\u201c\u6982\u7387\u201d\u7684\u503c\u3002\u903b\u8f91\u56de\u5f52\u5e94\u8fd0\u800c\u751f&#xff0c;\u5b83\u867d\u7136\u540d\u5b57\u91cc\u6709\u201c\u56de\u5f52\u201d&#xff0c;\u4f46\u5176\u672c\u8d28\u662f\u4e00\u4e2a\u7ecf\u5178\u7684\u4e8c\u5206\u7c7b\u7b97\u6cd5\u3002<\/p>\n<p>\u6838\u5fc3\u673a\u5173&#xff1a;Sigmoid\u51fd\u6570 \u903b\u8f91\u56de\u5f52\u7684\u5de7\u5999\u4e4b\u5904&#xff0c;\u5728\u4e8e\u5b83\u5728\u7ebf\u6027\u56de\u5f52\u7684\u57fa\u7840\u4e0a&#xff0c;\u5957\u4e0a\u4e86\u4e00\u4e2a\u201c\u9a6c\u7532\u201d\u2014\u2014Sigmoid\u51fd\u6570&#xff08;\u4e5f\u79f0Logistic\u51fd\u6570&#xff09;\u3002 Sigmoid\u51fd\u6570\u7684\u6570\u5b66\u5f62\u5f0f\u4e3a&#xff1a;\u03c3(z) &#061; 1 \/ (1 &#043; e\u207b\u1dbb) \u5b83\u7684\u795e\u5947\u4e4b\u5904\u5728\u4e8e&#xff0c;\u65e0\u8bba\u8f93\u5165 z \u662f\u591a\u5927\u6216\u591a\u5c0f\u7684\u5b9e\u6570&#xff0c;\u5b83\u7684\u8f93\u51fa\u503c \u03c3(z) \u6c38\u8fdc\u88ab\u538b\u7f29\u5728 (0, 1) \u8fd9\u4e2a\u533a\u95f4\u5185\u3002\u8fd9\u6b63\u597d\u7b26\u5408\u6982\u7387\u7684\u5b9a\u4e49&#xff01; \u903b\u8f91\u56de\u5f52\u7684\u505a\u6cd5\u662f&#xff1a;<\/p>\n<li>\u5148\u50cf\u7ebf\u6027\u56de\u5f52\u4e00\u6837&#xff0c;\u8ba1\u7b97\u4e00\u4e2a\u7ebf\u6027\u8f93\u51fa\u00a0z &#061; w\u1d40x &#043; b\u3002<\/li>\n<li>\u7136\u540e&#xff0c;\u5c06\u8fd9\u4e2a\u00a0z\u00a0\u503c\u9001\u5165Sigmoid\u51fd\u6570&#xff0c;\u5f97\u5230\u00a0p &#061; \u03c3(z)\u3002 \u8fd9\u4e2a\u8f93\u51fa\u00a0p\u00a0\u5c31\u53ef\u4ee5\u88ab\u89e3\u91ca\u4e3a\u6837\u672c\u5c5e\u4e8e\u201c\u6b63\u7c7b\u201d&#xff08;\u901a\u5e38\u75281\u8868\u793a&#xff09;\u7684\u6982\u7387\u3002\u5982\u679c\u00a0p &gt; 0.5&#xff0c;\u6211\u4eec\u5c31\u9884\u6d4b\u4e3a\u6b63\u7c7b&#xff1b;\u53cd\u4e4b&#xff0c;\u5219\u9884\u6d4b\u4e3a\u8d1f\u7c7b\u3002<\/li>\n<p>\u51b3\u7b56\u8fb9\u754c&#xff08;Decision Boundary&#xff09; p &#061; 0.5 \u662f\u5206\u7c7b\u7684\u4e34\u754c\u70b9&#xff0c;\u5b83\u53d1\u751f\u5728 z &#061; 0 \u7684\u65f6\u5019\u3002\u4e5f\u5c31\u662f\u8bf4&#xff0c;w\u1d40x &#043; b &#061; 0 \u8fd9\u6761\u7ebf&#xff08;\u6216\u9ad8\u7ef4\u5e73\u9762&#xff09;\u6210\u4e3a\u4e86\u4e24\u4e2a\u7c7b\u522b\u7684\u5206\u754c\u7ebf&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u51b3\u7b56\u8fb9\u754c\u3002\u5728\u51b3\u7b56\u8fb9\u754c\u4e00\u4fa7\u7684\u70b9\u88ab\u5212\u5206\u4e3a\u4e00\u7c7b&#xff0c;\u53e6\u4e00\u4fa7\u7684\u70b9\u88ab\u5212\u5206\u4e3a\u53e6\u4e00\u7c7b\u3002\u903b\u8f91\u56de\u5f52\u7684\u201c\u5b66\u4e60\u201d&#xff0c;\u5c31\u662f\u5728\u5bfb\u627e\u8fd9\u6761\u6700\u4f73\u7684\u51b3\u7b56\u8fb9\u754c\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218 \u6211\u4eec\u4f7f\u7528Scikit-Learn\u6765\u5b8c\u6210\u4e00\u4e2a\u7b80\u5355\u7684\u4e8c\u5206\u7c7b\u4efb\u52a1&#xff0c;\u6bd4\u5982\u6839\u636e\u80bf\u7624\u5927\u5c0f\u9884\u6d4b\u5176\u4e3a\u826f\u6027\u8fd8\u662f\u6076\u6027\u3002<\/p>\n<p># \u5bfc\u5165\u6240\u9700\u5e93<br \/>\nfrom sklearn.linear_model import LogisticRegression<br \/>\nfrom sklearn.datasets import make_classification<\/p>\n<p># 1. \u751f\u6210\u6a21\u62df\u5206\u7c7b\u6570\u636e<br \/>\nX, y &#061; make_classification(n_samples&#061;100, n_features&#061;1, n_informative&#061;1,<br \/>\n                           n_redundant&#061;0, n_clusters_per_class&#061;1, random_state&#061;42)<\/p>\n<p># 2. \u521b\u5efa\u5e76\u8bad\u7ec3\u6a21\u578b<br \/>\nmodel &#061; LogisticRegression()<br \/>\nmodel.fit(X, y)<\/p>\n<p># 3. \u53ef\u89c6\u5316\u51b3\u7b56\u8fb9\u754c (\u6b64\u5904\u4ec5\u4e3a\u793a\u610f&#xff0c;\u771f\u5b9e\u53ef\u89c6\u5316\u8f83\u590d\u6742)<br \/>\n# \u51b3\u7b56\u8fb9\u754c\u5728 z&#061;0 \u5904&#xff0c;\u5373 w*x &#043; b &#061; 0, x &#061; -b\/w<br \/>\nboundary_x &#061; -model.intercept_ \/ model.coef_[0]<br \/>\nprint(f&#034;\u51b3\u7b56\u8fb9\u754c\u4f4d\u4e8e x &#061; {boundary_x[0]:.2f}&#034;)<\/p>\n<p># \u7ed8\u5236\u6570\u636e\u70b9<br \/>\nplt.scatter(X, y, c&#061;y, cmap&#061;&#039;bwr&#039;, edgecolor&#061;&#039;k&#039;)<br \/>\n# \u7ed8\u5236\u51b3\u7b56\u8fb9\u754c<br \/>\nplt.axvline(x&#061;boundary_x, color&#061;&#039;green&#039;, linestyle&#061;&#039;&#8211;&#039;, label&#061;&#039;\u51b3\u7b56\u8fb9\u754c&#039;)<br \/>\nplt.xlabel(&#039;\u80bf\u7624\u5927\u5c0f&#039;)<br \/>\nplt.ylabel(&#039;\u7c7b\u522b (0: \u826f\u6027, 1: \u6076\u6027)&#039;)<br \/>\nplt.legend()<br \/>\nplt.show()<\/p>\n<p>\u8fd9\u6bb5\u4ee3\u7801\u5c55\u793a\u4e86\u903b\u8f91\u56de\u5f52\u5982\u4f55\u627e\u5230\u4e00\u4e2a\u5206\u754c\u70b9&#xff08;\u7eff\u8272\u7684\u865a\u7ebf&#xff09;&#xff0c;\u5c06\u4ee3\u8868\u826f\u6027\u4e0e\u6076\u6027\u7684\u4e24\u7c7b\u6570\u636e\u70b9\u6e05\u6670\u5730\u5206\u5f00\u3002<\/p>\n<h5>3.2.2 \u51b3\u7b56\u6811\u4e0e\u968f\u673a\u68ee\u6797&#xff1a;\u96c6\u6210\u601d\u60f3\u7684\u521d\u6b65\u4f53\u73b0<\/h5>\n<p>\u5982\u679c\u8bf4\u7ebf\u6027\u6a21\u578b\u662f\u7528\u4e00\u4e2a\u5168\u5c40\u7684\u3001\u7edf\u4e00\u7684\u51fd\u6570\u6765\u5212\u5206\u4e16\u754c&#xff0c;\u90a3\u4e48\u51b3\u7b56\u6811\u5219\u662f\u4e00\u79cd\u5b8c\u5168\u4e0d\u540c\u7684\u3001\u5206\u800c\u6cbb\u4e4b\u7684\u54f2\u5b66\u3002<\/p>\n<p>\u51b3\u7b56\u6811&#xff1a;\u683c\u7269\u81f4\u77e5\u7684\u5178\u8303<\/p>\n<p>\u6a21\u578b\u54f2\u5b66 \u51b3\u7b56\u6811\u7684\u601d\u8003\u65b9\u5f0f&#xff0c;\u4e0e\u4eba\u7c7b\u7684\u51b3\u7b56\u8fc7\u7a0b\u5982\u51fa\u4e00\u8f99\u3002\u5b83\u901a\u8fc7\u63d0\u51fa\u4e00\u7cfb\u5217\u201c\u662f\/\u5426\u201d\u7684\u95ee\u9898&#xff0c;\u5bf9\u6570\u636e\u8fdb\u884c\u5c42\u5c42\u5212\u5206&#xff0c;\u6700\u7ec8\u5bfc\u5411\u4e00\u4e2a\u7ed3\u8bba\u3002<\/p>\n<p>\u4f8b\u5982&#xff0c;\u5224\u65ad\u4e00\u4e2a\u74dc\u662f\u4e0d\u662f\u597d\u74dc&#xff0c;\u6211\u4eec\u53ef\u80fd\u4f1a\u95ee&#xff1a;\u201c\u5b83\u7684\u989c\u8272\u662f\u9752\u7eff\u5417&#xff1f;\u201d -&gt; \u201c\u662f\u7684\u3002\u201d -&gt; \u201c\u5b83\u7684\u6839\u8482\u662f\u8737\u7f29\u7684\u5417&#xff1f;\u201d -&gt; \u201c\u662f\u7684\u3002\u201d -&gt; \u201c\u5b83\u7684\u6572\u58f0\u662f\u6c89\u95f7\u7684\u5417&#xff1f;\u201d -&gt; \u201c\u4e0d\u662f\u3002\u201d -&gt; \u201c\u7ed3\u8bba&#xff1a;\u8fd9\u662f\u4e2a\u574f\u74dc\u3002\u201d \u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u4e00\u68f5\u51b3\u7b56\u6811\u3002<\/p>\n<p>\u5b83\u7684\u6700\u5927\u4f18\u70b9\u662f\u6a21\u578b\u5177\u6709\u6781\u5f3a\u7684\u53ef\u89e3\u91ca\u6027&#xff0c;\u6211\u4eec\u53ef\u4ee5\u6e05\u6670\u5730\u770b\u5230\u6bcf\u4e00\u4e2a\u51b3\u7b56\u7684\u4f9d\u636e\u3002<\/p>\n<p>\u6784\u5efa\u8fc7\u7a0b&#xff1a;\u5982\u4f55\u9009\u62e9\u6700\u4f73\u95ee\u9898 \u51b3\u7b56\u6811\u6784\u5efa\u7684\u5173\u952e&#xff0c;\u662f\u5728\u6bcf\u4e00\u6b65&#xff0c;\u5982\u4f55\u9009\u62e9\u4e00\u4e2a\u201c\u6700\u4f73\u201d\u7684\u95ee\u9898&#xff08;\u5373\u7279\u5f81&#xff09;\u6765\u5212\u5206\u5f53\u524d\u7684\u6570\u636e\u96c6&#xff0c;\u4f7f\u5f97\u5212\u5206\u540e\u7684\u6570\u636e\u201c\u7eaf\u5ea6\u201d\u6700\u9ad8\u3002\u8fd9\u91cc\u7684\u201c\u7eaf\u5ea6\u201d\u6307\u7684\u662f\u6570\u636e\u7684\u7c7b\u522b\u4e00\u81f4\u6027\u3002 \u6211\u4eec\u4f7f\u7528**\u4fe1\u606f\u589e\u76ca&#xff08;Information Gain&#xff09;\u6216\u57fa\u5c3c\u4e0d\u7eaf\u5ea6&#xff08;Gini Impurity&#xff09;**\u6765\u8861\u91cf\u4e00\u4e2a\u95ee\u9898\u7684\u597d\u574f\u3002\u4e00\u4e2a\u597d\u7684\u95ee\u9898&#xff0c;\u5e94\u8be5\u80fd\u8ba9\u5212\u5206\u540e\u7684\u5b50\u96c6&#xff0c;\u5176\u4e0d\u786e\u5b9a\u6027&#xff08;\u4fe1\u606f\u71b5&#xff09;\u6216\u4e0d\u7eaf\u5ea6&#xff08;\u57fa\u5c3c\u6307\u6570&#xff09;\u5927\u5e45\u4e0b\u964d\u3002\u51b3\u7b56\u6811\u4f1a\u8d2a\u5fc3\u5730\u9009\u62e9\u90a3\u4e2a\u80fd\u5e26\u6765\u6700\u5927\u4fe1\u606f\u589e\u76ca\u7684\u7279\u5f81\u8fdb\u884c\u5206\u88c2\u3002<\/p>\n<p>\u526a\u679d&#xff08;Pruning&#xff09; \u5982\u679c\u4efb\u7531\u51b3\u7b56\u6811\u751f\u957f&#xff0c;\u5b83\u4f1a\u4e3a\u6bcf\u4e00\u4e2a\u8bad\u7ec3\u6837\u672c\u90fd\u627e\u5230\u4e00\u6761\u5b8c\u7f8e\u7684\u8def\u5f84&#xff0c;\u6700\u7ec8\u5bfc\u81f4\u6811\u53d8\u5f97\u5f02\u5e38\u201c\u8302\u76db\u201d\u548c\u590d\u6742\u3002\u8fd9\u6837\u7684\u6811\u5bf9\u8bad\u7ec3\u6570\u636e\u62df\u5408\u5f97\u5f88\u597d&#xff0c;\u4f46\u5bf9\u65b0\u6570\u636e\u7684\u6cdb\u5316\u80fd\u529b\u4f1a\u5f88\u5dee&#xff0c;\u8fd9\u5c31\u662f\u8fc7\u62df\u5408\u3002 \u4e3a\u4e86\u9632\u6b62\u8fd9\u79cd\u60c5\u51b5&#xff0c;\u6211\u4eec\u9700\u8981\u5bf9\u6811\u8fdb\u884c\u526a\u679d\u3002\u53ef\u4ee5\u5728\u6811\u751f\u957f\u65f6\u5c31\u9650\u5236\u5176\u6df1\u5ea6\u6216\u53f6\u5b50\u8282\u70b9\u6570\u91cf&#xff08;\u9884\u526a\u679d&#xff09;&#xff0c;\u4e5f\u53ef\u4ee5\u7b49\u6811\u5b8c\u5168\u957f\u6210\u540e\u518d\u780d\u6389\u4e00\u4e9b\u4e0d\u5fc5\u8981\u7684\u679d\u53f6&#xff08;\u540e\u526a\u679d&#xff09;\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218 \u4f7f\u7528Scikit-Learn\u6784\u5efa\u51b3\u7b56\u6811&#xff0c;\u5e76\u5229\u7528graphviz\u5de5\u5177\u5c06\u5176\u51b3\u7b56\u8fc7\u7a0b\u53ef\u89c6\u5316\u3002<\/p>\n<p>from sklearn.tree import DecisionTreeClassifier, export_graphviz<br \/>\nfrom sklearn.datasets import load_iris<br \/>\nimport graphviz<\/p>\n<p># 1. \u52a0\u8f7d\u9e22\u5c3e\u82b1\u6570\u636e\u96c6<br \/>\niris &#061; load_iris()<br \/>\nX, y &#061; iris.data, iris.target<\/p>\n<p># 2. \u521b\u5efa\u5e76\u8bad\u7ec3\u51b3\u7b56\u6811\u6a21\u578b<br \/>\nmodel &#061; DecisionTreeClassifier(max_depth&#061;3) # \u9650\u5236\u6700\u5927\u6df1\u5ea6\u4ee5\u9632\u8fc7\u62df\u5408<br \/>\nmodel.fit(X, y)<\/p>\n<p># 3. \u5c06\u51b3\u7b56\u6811\u53ef\u89c6\u5316<br \/>\ndot_data &#061; export_graphviz(model, out_file&#061;None,<br \/>\n                          feature_names&#061;iris.feature_names,<br \/>\n                          class_names&#061;iris.target_names,<br \/>\n                          filled&#061;True, rounded&#061;True,<br \/>\n                          special_characters&#061;True)<br \/>\ngraph &#061; graphviz.Source(dot_data)<br \/>\n# graph.render(&#034;iris_decision_tree&#034;) # \u53ef\u4ee5\u4fdd\u5b58\u4e3a\u6587\u4ef6<br \/>\n# \u5728Jupyter Notebook\u4e2d&#xff0c;\u53ef\u4ee5\u76f4\u63a5\u663e\u793agraph\u5bf9\u8c61<br \/>\n# graph <\/p>\n<p>\u8fd0\u884c\u4e0a\u8ff0\u4ee3\u7801&#xff08;\u9700\u5b89\u88c5graphviz\u5e93&#xff09;&#xff0c;\u8bfb\u8005\u5c06\u80fd\u5f97\u5230\u4e00\u5f20\u6e05\u6670\u7684\u51b3\u7b56\u6811\u56fe&#xff0c;\u76f4\u89c2\u5730\u770b\u5230\u6a21\u578b\u662f\u5982\u4f55\u6839\u636e\u82b1\u74e3\u3001\u82b1\u843c\u7684\u957f\u5bbd\u6765\u5224\u65ad\u9e22\u5c3e\u82b1\u54c1\u79cd\u7684\u3002<\/p>\n<p>\u968f\u673a\u68ee\u6797&#xff08;Random Forest&#xff09;&#xff1a;\u4e09\u4e2a\u81ed\u76ae\u5320&#xff0c;\u8d5b\u8fc7\u8bf8\u845b\u4eae<\/p>\n<p>\u96c6\u6210\u5b66\u4e60&#xff08;Ensemble Learning&#xff09;\u601d\u60f3 \u5355\u68f5\u51b3\u7b56\u6811&#xff0c;\u5c24\u5176\u662f\u6df1\u5ea6\u8f83\u5927\u7684\u51b3\u7b56\u6811&#xff0c;\u5bb9\u6613\u8fc7\u62df\u5408&#xff0c;\u5bf9\u6570\u636e\u7684\u5fae\u5c0f\u6270\u52a8\u5f88\u654f\u611f\u3002\u4e00\u4e2a\u81ea\u7136\u7684\u60f3\u6cd5\u662f&#xff1a;\u6211\u4eec\u80fd\u4e0d\u80fd\u7efc\u5408\u591a\u68f5\u4e0d\u540c\u51b3\u7b56\u6811\u7684\u610f\u89c1\u6765\u505a\u51b3\u5b9a&#xff1f;\u8fd9\u5c31\u662f\u96c6\u6210\u5b66\u4e60\u7684\u6838\u5fc3\u601d\u60f3&#xff0c;\u4fd7\u79f0\u201c\u7fa4\u4f53\u667a\u6167\u201d\u3002\u968f\u673a\u68ee\u6797\u6b63\u662f\u8fd9\u79cd\u601d\u60f3\u6700\u6770\u51fa\u7684\u4ee3\u8868\u3002<\/p>\n<p>\u201c\u968f\u673a\u201d\u4f53\u73b0\u5728\u4f55\u5904 \u4e3a\u4e86\u8ba9\u68ee\u6797\u4e2d\u7684\u6bcf\u4e00\u68f5\u6811\u90fd\u6709\u6240\u4e0d\u540c&#xff0c;\u4ece\u800c\u5f62\u6210\u4e92\u8865&#xff0c;\u968f\u673a\u68ee\u6797\u5f15\u5165\u4e86\u4e24\u4e2a\u201c\u968f\u673a\u201d\u673a\u5236&#xff1a;<\/p>\n<li>\u6570\u636e\u968f\u673a&#xff08;\u884c\u62bd\u6837&#xff09;&#xff1a;\u68ee\u6797\u4e2d\u7684\u6bcf\u4e00\u68f5\u6811&#xff0c;\u90fd\u4e0d\u662f\u7528\u5168\u90e8\u7684\u8bad\u7ec3\u6570\u636e\u6765\u8bad\u7ec3\u7684\u3002\u800c\u662f\u901a\u8fc7\u81ea\u52a9\u91c7\u6837\u6cd5&#xff08;Bootstrap Aggregating&#xff0c;\u7b80\u79f0Bagging&#xff09;&#xff0c;\u4ece\u539f\u59cb\u6570\u636e\u96c6\u4e2d\u6709\u653e\u56de\u5730\u62bd\u53d6\u4e00\u4e2a\u4e0e\u539f\u59cb\u6570\u636e\u96c6\u540c\u6837\u5927\u5c0f\u7684\u5b50\u96c6\u3002\u8fd9\u6837&#xff0c;\u6bcf\u68f5\u6811\u770b\u5230\u7684\u201c\u4e16\u754c\u201d&#xff08;\u8bad\u7ec3\u6570\u636e&#xff09;\u90fd\u662f\u4e0d\u5b8c\u5168\u4e00\u6837\u7684\u3002<\/li>\n<li>\u7279\u5f81\u968f\u673a&#xff08;\u5217\u62bd\u6837&#xff09;&#xff1a;\u5728\u6784\u5efa\u6bcf\u4e00\u68f5\u6811\u7684\u6bcf\u4e00\u4e2a\u8282\u70b9\u65f6&#xff0c;\u4e0d\u518d\u662f\u4ece\u5168\u90e8\u7279\u5f81\u4e2d\u9009\u62e9\u6700\u4f18\u5206\u88c2\u7279\u5f81&#xff0c;\u800c\u662f\u968f\u673a\u62bd\u53d6\u4e00\u90e8\u5206\u7279\u5f81&#xff0c;\u518d\u4ece\u8fd9\u90e8\u5206\u7279\u5f81\u4e2d\u9009\u62e9\u6700\u4f18\u7684\u3002\u8fd9\u8fdb\u4e00\u6b65\u589e\u52a0\u4e86\u6811\u4e4b\u95f4\u7684\u5dee\u5f02\u6027\u3002<\/li>\n<p>\u4e3a\u4f55\u6709\u6548 \u8fd9\u4e24\u4e2a\u201c\u968f\u673a\u201d\u673a\u5236&#xff0c;\u6781\u5927\u5730\u4fdd\u8bc1\u4e86\u68ee\u6797\u4e2d\u6811\u7684\u591a\u6837\u6027&#xff08;Diversity&#xff09;\u3002\u6709\u4e9b\u6811\u53ef\u80fd\u5728\u8fd9\u4e2a\u95ee\u9898\u4e0a\u72af\u9519&#xff0c;\u4f46\u53e6\u4e00\u4e9b\u6811\u53ef\u80fd\u6070\u597d\u662f\u6b63\u786e\u7684\u3002\u5f53\u68ee\u6797\u8fdb\u884c\u9884\u6d4b\u65f6&#xff0c;\u5b83\u4f1a\u91c7\u7528\u201c\u6295\u7968\u201d\u7684\u65b9\u5f0f&#xff08;\u5206\u7c7b\u95ee\u9898&#xff09;\u6216\u201c\u53d6\u5e73\u5747\u201d\u7684\u65b9\u5f0f&#xff08;\u56de\u5f52\u95ee\u9898&#xff09;&#xff0c;\u7efc\u5408\u6240\u6709\u6811\u7684\u610f\u89c1\u3002\u8fd9\u79cd\u673a\u5236\u53ef\u4ee5\u6709\u6548\u5730\u964d\u4f4e\u6574\u4f53\u6a21\u578b\u7684\u65b9\u5dee&#xff0c;\u4f7f\u5f97\u968f\u673a\u68ee\u6797\u5177\u6709\u975e\u5e38\u597d\u7684\u6297\u8fc7\u62df\u5408\u80fd\u529b\u548c\u7a33\u5b9a\u6027\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218 \u6211\u4eec\u4f7f\u7528Scikit-Learn\u7684RandomForestClassifier&#xff0c;\u5728\u540c\u6837\u7684\u6570\u636e\u4e0a\u6bd4\u8f83\u5b83\u4e0e\u5355\u68f5\u51b3\u7b56\u6811\u7684\u6027\u80fd\u3002<\/p>\n<p>from sklearn.ensemble import RandomForestClassifier<br \/>\nfrom sklearn.model_selection import cross_val_score<\/p>\n<p># \u4ecd\u7136\u4f7f\u7528\u9e22\u5c3e\u82b1\u6570\u636e\u96c6<br \/>\nX, y &#061; iris.data, iris.target<\/p>\n<p># \u521b\u5efa\u5355\u68f5\u51b3\u7b56\u6811\u6a21\u578b<br \/>\ntree_clf &#061; DecisionTreeClassifier(random_state&#061;42)<br \/>\n# \u521b\u5efa\u968f\u673a\u68ee\u6797\u6a21\u578b<br \/>\nforest_clf &#061; RandomForestClassifier(n_estimators&#061;100, random_state&#061;42) # n_estimators\u662f\u6811\u7684\u6570\u91cf<\/p>\n<p># \u4f7f\u7528\u4ea4\u53c9\u9a8c\u8bc1\u6bd4\u8f83\u6027\u80fd<br \/>\ntree_scores &#061; cross_val_score(tree_clf, X, y, cv&#061;5)<br \/>\nforest_scores &#061; cross_val_score(forest_clf, X, y, cv&#061;5)<\/p>\n<p>print(f&#034;\u5355\u68f5\u51b3\u7b56\u6811\u7684\u5e73\u5747\u51c6\u786e\u7387: {tree_scores.mean():.4f}&#034;)<br \/>\nprint(f&#034;\u968f\u673a\u68ee\u6797\u7684\u5e73\u5747\u51c6\u786e\u7387: {forest_scores.mean():.4f}&#034;)<\/p>\n<p>\u901a\u5e38\u60c5\u51b5\u4e0b&#xff0c;\u8bfb\u8005\u4f1a\u89c2\u5bdf\u5230\u968f\u673a\u68ee\u6797\u7684\u5e73\u5747\u51c6\u786e\u7387\u4f1a\u66f4\u7a33\u5b9a&#xff0c;\u4e5f\u53ef\u80fd\u66f4\u9ad8\u3002<\/p>\n<h5>3.2.3 \u652f\u6301\u5411\u91cf\u673a&#xff08;SVM&#xff09;&#xff1a;\u6700\u5927\u5316\u95f4\u9694\u7684\u827a\u672f<\/h5>\n<p>\u652f\u6301\u5411\u91cf\u673a\u662f\u4e00\u4e2a\u529f\u80fd\u5f3a\u5927\u4e14\u7406\u8bba\u4f18\u7f8e\u7684\u5206\u7c7b\u7b97\u6cd5\u3002\u5b83\u7684\u6838\u5fc3\u601d\u60f3&#xff0c;\u4e0d\u662f\u7b80\u5355\u5730\u5c06\u6570\u636e\u5206\u5f00&#xff0c;\u800c\u662f\u8981\u4ee5\u6700\u201c\u7a33\u5065\u201d\u7684\u65b9\u5f0f\u5206\u5f00\u3002<\/p>\n<p>\u6838\u5fc3\u601d\u60f3&#xff1a;\u5bfb\u627e\u6700\u201c\u5bbd\u201d\u7684\u8857\u9053<\/p>\n<ul>\n<li>\n<p>\u6a21\u578b\u54f2\u5b66 \u60f3\u8c61\u4e00\u4e0b&#xff0c;\u5728\u4e24\u79cd\u4e0d\u540c\u989c\u8272\u7684\u6570\u636e\u70b9\u4e4b\u95f4&#xff0c;\u53ef\u4ee5\u753b\u51fa\u65e0\u6570\u6761\u7ebf\u5c06\u5b83\u4eec\u5206\u5f00\u3002SVM\u8ba4\u4e3a&#xff0c;\u6700\u597d\u7684\u90a3\u6761\u5206\u754c\u7ebf&#xff0c;\u5e94\u8be5\u662f\u8fd9\u6837\u4e00\u6761\u7ebf&#xff1a;\u5b83\u4f4d\u4e8e\u4e24\u7c7b\u6570\u636e\u70b9\u7684\u6b63\u4e2d\u95f4&#xff0c;\u5e76\u4e14\u79bb\u4e24\u8fb9\u6700\u8fd1\u7684\u6570\u636e\u70b9&#xff08;\u6211\u4eec\u79f0\u4e4b\u4e3a\u652f\u6301\u5411\u91cf&#xff09;\u7684\u8ddd\u79bb\u662f\u6700\u8fdc\u7684\u3002 \u8fd9\u6761\u5206\u754c\u7ebf\u5c31\u50cf\u4e00\u6761\u8857\u9053&#xff0c;\u800c\u8857\u9053\u4e24\u4fa7\u5230\u6700\u8fd1\u7684\u5efa\u7b51&#xff08;\u652f\u6301\u5411\u91cf&#xff09;\u4e4b\u95f4\u7684\u8ddd\u79bb&#xff0c;\u5c31\u662f\u8857\u9053\u7684\u5bbd\u5ea6&#xff0c;\u8fd9\u4e2a\u5bbd\u5ea6\u5728SVM\u4e2d\u88ab\u79f0\u4e3a\u95f4\u9694&#xff08;Margin&#xff09;\u3002SVM\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u6700\u5927\u5316\u8fd9\u4e2a\u95f4\u9694\u3002\u4e00\u6761\u66f4\u5bbd\u7684\u201c\u8857\u9053\u201d\u610f\u5473\u7740\u6a21\u578b\u5bf9\u672a\u77e5\u6570\u636e\u7684\u5bb9\u5fcd\u5ea6\u66f4\u9ad8&#xff0c;\u6cdb\u5316\u80fd\u529b\u66f4\u5f3a\u3002<\/p>\n<\/li>\n<li>\n<p>\u7ebf\u6027SVM \u5728\u7ebf\u6027\u53ef\u5206\u7684\u60c5\u51b5\u4e0b&#xff0c;SVM\u5c31\u662f\u8981\u627e\u5230\u7531\u51b3\u7b56\u8fb9\u754c&#xff08;\u8857\u9053\u4e2d\u5fc3\u7ebf&#xff09;\u3001\u95f4\u9694\u548c\u652f\u6301\u5411\u91cf&#xff08;\u8857\u9053\u4e24\u65c1\u7684\u5efa\u7b51&#xff09;\u6784\u6210\u7684\u8fd9\u4e2a\u6700\u4f18\u201c\u8857\u9053\u7cfb\u7edf\u201d\u3002\u4e00\u4e2a\u6709\u8da3\u7684\u7279\u70b9\u662f&#xff0c;\u6700\u7ec8\u7684\u51b3\u7b56\u8fb9\u754c\u4ec5\u4ec5\u7531\u8fd9\u4e9b\u5c11\u6570\u7684\u652f\u6301\u5411\u91cf\u51b3\u5b9a&#xff0c;\u4e0e\u5176\u4ed6\u6570\u636e\u70b9\u65e0\u5173\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u6838\u6280\u5de7&#xff08;The Kernel Trick&#xff09;&#xff1a;\u4e7e\u5764\u5927\u632a\u79fb<\/p>\n<ul>\n<li>\n<p>\u5904\u7406\u975e\u7ebf\u6027\u95ee\u9898 \u73b0\u5b9e\u4e16\u754c\u7684\u6570\u636e\u5f80\u5f80\u4e0d\u662f\u7ebf\u6027\u53ef\u5206\u7684\u3002\u6bd4\u5982&#xff0c;\u4e00\u7c7b\u6570\u636e\u70b9\u88ab\u53e6\u4e00\u7c7b\u6570\u636e\u70b9\u5305\u56f4&#xff0c;\u65e0\u6cd5\u7528\u4e00\u6761\u76f4\u7ebf\u5c06\u5b83\u4eec\u5206\u5f00\u3002\u6b64\u65f6&#xff0c;SVM\u5982\u4f55\u5e94\u5bf9&#xff1f;<\/p>\n<\/li>\n<li>\n<p>\u5347\u7ef4\u6253\u51fb SVM\u7684\u7cbe\u9ad3\u2014\u2014\u6838\u6280\u5de7\u767b\u573a\u4e86\u3002\u5b83\u7684\u601d\u60f3&#xff0c;\u582a\u79f0\u201c\u4e7e\u5764\u5927\u632a\u79fb\u201d\u3002\u5b83\u901a\u8fc7\u4e00\u4e2a\u79f0\u4e3a\u6838\u51fd\u6570&#xff08;Kernel Function&#xff09;\u7684\u6570\u5b66\u53d8\u6362&#xff0c;\u5c06\u6570\u636e\u4ece\u5f53\u524d\u7684\u4f4e\u7ef4\u7279\u5f81\u7a7a\u95f4&#xff0c;\u5de7\u5999\u5730\u6620\u5c04\u5230\u4e00\u4e2a\u66f4\u9ad8\u7ef4\u7684\u7a7a\u95f4\u3002\u795e\u5947\u7684\u662f&#xff0c;\u5728\u4f4e\u7ef4\u7a7a\u95f4\u4e2d\u7ebf\u6027\u4e0d\u53ef\u5206\u7684\u6570\u636e&#xff0c;\u5728\u6620\u5c04\u5230\u9ad8\u7ef4\u7a7a\u95f4\u540e&#xff0c;\u5f80\u5f80\u5c31\u53d8\u5f97\u7ebf\u6027\u53ef\u5206\u4e86\u3002<\/p>\n<p>\u4e00\u4e2a\u6bd4\u55bb&#xff1a;\u60f3\u8c61\u684c\u9762\u4e0a\u6709\u4e00\u5806\u7ea2\u8272\u548c\u84dd\u8272\u7684\u73e0\u5b50\u6df7\u5728\u4e00\u8d77&#xff0c;\u65e0\u6cd5\u5728\u684c\u9762\u4e0a\u753b\u4e00\u6761\u76f4\u7ebf\u5206\u5f00\u5b83\u4eec\u3002\u73b0\u5728\u60a8\u731b\u5730\u4e00\u62cd\u684c\u5b50&#xff0c;\u6240\u6709\u73e0\u5b50\u90fd\u98de\u5230\u4e86\u7a7a\u4e2d&#xff08;\u5347\u7ef4&#xff09;\u3002\u5728\u5b83\u4eec\u98de\u5230\u6700\u9ad8\u70b9\u7684\u90a3\u4e00\u77ac\u95f4&#xff0c;\u60a8\u7528\u4e00\u4e2a\u6c34\u5e73\u9762&#xff08;\u9ad8\u7ef4\u7a7a\u95f4\u7684\u201c\u76f4\u7ebf\u201d&#xff09;\u5c31\u80fd\u8f7b\u6613\u5730\u5c06\u7ea2\u8272\u548c\u84dd\u8272\u7684\u73e0\u5b50\u5206\u5f00\u3002<\/p>\n<\/li>\n<li>\n<p>\u201c\u6838\u201d\u7684\u5de7\u5999 \u66f4\u4ee4\u4eba\u60ca\u53f9\u7684\u662f&#xff0c;SVM\u7684\u6838\u6280\u5de7\u65e0\u9700\u771f\u6b63\u5730\u8ba1\u7b97\u6570\u636e\u70b9\u5728\u9ad8\u7ef4\u7a7a\u95f4\u4e2d\u7684\u5750\u6807&#xff0c;\u8fd9\u5728\u7ef4\u5ea6\u6781\u9ad8\u65f6&#xff08;\u751a\u81f3\u662f\u65e0\u9650\u7ef4&#xff09;\u662f\u65e0\u6cd5\u505a\u5230\u7684\u3002\u5b83\u53ea\u9700\u8981\u901a\u8fc7\u6838\u51fd\u6570\u8ba1\u7b97\u51fa\u6570\u636e\u70b9\u5728\u4f4e\u7ef4\u7a7a\u95f4\u4e2d\u7684\u5185\u79ef&#xff0c;\u5176\u7ed3\u679c\u5c31\u7b49\u4ef7\u4e8e\u5b83\u4eec\u5728\u9ad8\u7ef4\u7a7a\u95f4\u4e2d\u7684\u5185\u79ef\u3002\u8fd9\u6781\u5927\u5730\u964d\u4f4e\u4e86\u8ba1\u7b97\u7684\u590d\u6742\u5ea6&#xff0c;\u4f7f\u5f97\u201c\u5347\u7ef4\u6253\u51fb\u201d\u6210\u4e3a\u53ef\u80fd\u3002\u5e38\u7528\u7684\u6838\u51fd\u6570\u6709\u9ad8\u65af\u6838&#xff08;RBF&#xff09;\u3001\u591a\u9879\u5f0f\u6838\u7b49\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u4ee3\u7801\u5b9e\u6218 \u6211\u4eec\u4f7f\u7528Scikit-Learn\u7684SVC&#xff08;Support Vector Classifier&#xff09;\u6765\u5904\u7406\u4e00\u4e2a\u975e\u7ebf\u6027\u53ef\u5206\u7684\u6570\u636e\u96c6\u3002<\/p>\n<p>from sklearn.svm import SVC<br \/>\nfrom sklearn.datasets import make_circles<\/p>\n<p># 1. \u751f\u6210\u4e00\u4e2a\u73af\u5f62&#xff08;\u975e\u7ebf\u6027&#xff09;\u6570\u636e\u96c6<br \/>\nX, y &#061; make_circles(n_samples&#061;100, noise&#061;0.1, factor&#061;0.5, random_state&#061;42)<\/p>\n<p># 2. \u521b\u5efa\u5e76\u8bad\u7ec3\u4e00\u4e2a\u5e26RBF\u6838\u7684SVM\u6a21\u578b<br \/>\nmodel &#061; SVC(kernel&#061;&#039;rbf&#039;, C&#061;1.0, gamma&#061;&#039;auto&#039;) # rbf\u662f\u9ad8\u65af\u6838<br \/>\nmodel.fit(X, y)<\/p>\n<p># 3. \u53ef\u89c6\u5316\u51b3\u7b56\u8fb9\u754c (\u6b64\u5904\u4ee3\u7801\u8f83\u590d\u6742&#xff0c;\u4ec5\u4e3a\u793a\u610f&#xff0c;\u9700\u8981\u8f85\u52a9\u51fd\u6570)<br \/>\n# \u6838\u5fc3\u662f\u7ed8\u5236\u51fa model.predict() \u7ed3\u679c\u7684\u7b49\u9ad8\u7ebf\u56fe<br \/>\n# \u5728\u56fe\u4e2d&#xff0c;\u8bfb\u8005\u4f1a\u770b\u5230\u4e00\u6761\u5e73\u6ed1\u7684\u5706\u5f62\u8fb9\u754c&#xff0c;\u5b8c\u7f8e\u5730\u5206\u5f00\u4e86\u5185\u5916\u4e24\u5708\u6570\u636e\u70b9<br \/>\n# \u5e76\u4e14\u53ef\u4ee5\u770b\u5230&#xff0c;\u53ea\u6709\u8fb9\u754c\u9644\u8fd1\u7684\u70b9&#xff08;\u652f\u6301\u5411\u91cf&#xff09;\u5bf9\u51b3\u7b56\u8fb9\u754c\u6709\u5f71\u54cd<\/p>\n<p>\u901a\u8fc7\u8fd9\u4e2a\u4f8b\u5b50&#xff0c;\u8bfb\u8005\u53ef\u4ee5\u76f4\u89c2\u5730\u611f\u53d7\u5230\u6838\u6280\u5de7\u7684\u5a01\u529b&#xff0c;\u5b83\u5982\u4f55\u6beb\u4e0d\u8d39\u529b\u5730\u89e3\u51b3\u4e86\u7ebf\u6027\u6a21\u578b\u65e0\u6cd5\u89e3\u51b3\u7684\u95ee\u9898\u3002<\/p>\n<h4>3.3 \u6a21\u578b\u7684\u8bc4\u4f30\u4e0e\u4f18\u5316&#xff1a;\u77e5\u5176\u7136&#xff0c;\u66f4\u8981\u77e5\u5176\u6240\u4ee5\u7136<\/h4>\n<p>\u5efa\u7acb\u6a21\u578b\u53ea\u662f\u7b2c\u4e00\u6b65\u3002\u6211\u4eec\u5982\u4f55\u77e5\u9053\u4e00\u4e2a\u6a21\u578b\u662f\u597d\u662f\u574f&#xff1f;\u53c8\u5982\u4f55\u8bca\u65ad\u5b83\u5b58\u5728\u7684\u95ee\u9898\u5e76\u52a0\u4ee5\u6539\u8fdb&#xff1f;\u672c\u8282\u5c06\u4ecb\u7ecd\u4e00\u5957\u79d1\u5b66\u7684\u8bc4\u4f30\u4e0e\u4f18\u5316\u65b9\u6cd5\u8bba\u3002<\/p>\n<h5>3.3.1 \u6a21\u578b\u7684\u8bc4\u4f30&#xff1a;\u660e\u8fa8\u662f\u975e\u7684\u6807\u5c3a<\/h5>\n<p>\u9009\u62e9\u5408\u9002\u7684\u8bc4\u4f30\u6307\u6807\u81f3\u5173\u91cd\u8981&#xff0c;\u5b83\u51b3\u5b9a\u4e86\u6211\u4eec\u4f18\u5316\u6a21\u578b\u7684\u65b9\u5411\u3002<\/p>\n<p>\u5206\u7c7b\u6a21\u578b\u8bc4\u4f30\u6307\u6807<\/p>\n<ul>\n<li>\n<p>\u6df7\u6dc6\u77e9\u9635&#xff08;Confusion Matrix&#xff09; \u5b83\u662f\u6240\u6709\u5206\u7c7b\u6307\u6807\u7684\u57fa\u77f3\u3002\u5bf9\u4e8e\u4e00\u4e2a\u4e8c\u5206\u7c7b\u95ee\u9898&#xff0c;\u6df7\u6dc6\u77e9\u9635\u662f\u4e00\u4e2a2&#215;2\u7684\u8868\u683c&#xff0c;\u6e05\u6670\u5730\u5c55\u793a\u4e86\u6a21\u578b\u9884\u6d4b\u7684\u56db\u79cd\u60c5\u51b5&#xff1a;<\/p>\n<ul>\n<li>\u771f\u6b63\u4f8b (TP): \u771f\u5b9e\u4e3a\u6b63&#xff0c;\u9884\u6d4b\u4e5f\u4e3a\u6b63\u3002<\/li>\n<li>\u5047\u6b63\u4f8b (FP): \u771f\u5b9e\u4e3a\u8d1f&#xff0c;\u9884\u6d4b\u5374\u4e3a\u6b63\u3002&#xff08;\u8bef\u62a5&#xff09;<\/li>\n<li>\u771f\u8d1f\u4f8b (TN): \u771f\u5b9e\u4e3a\u8d1f&#xff0c;\u9884\u6d4b\u4e5f\u4e3a\u8d1f\u3002<\/li>\n<li>\u5047\u8d1f\u4f8b (FN): \u771f\u5b9e\u4e3a\u6b63&#xff0c;\u9884\u6d4b\u5374\u4e3a\u8d1f\u3002&#xff08;\u6f0f\u62a5&#xff09;<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u51c6\u786e\u7387&#xff08;Accuracy&#xff09; Accuracy &#061; (TP &#043; TN) \/ (TP &#043; FP &#043; TN &#043; FN) \u8fd9\u662f\u6700\u76f4\u89c2\u7684\u6307\u6807&#xff0c;\u5373\u201c\u9884\u6d4b\u6b63\u786e\u7684\u6837\u672c\u6570 \/ \u603b\u6837\u672c\u6570\u201d\u3002\u4f46\u5728\u6837\u672c\u4e0d\u5747\u8861&#xff08;\u4f8b\u598299%\u7684\u90ae\u4ef6\u90fd\u662f\u6b63\u5e38\u90ae\u4ef6&#xff09;\u7684\u60c5\u51b5\u4e0b&#xff0c;\u5b83\u5177\u6709\u6781\u5927\u7684\u8bef\u5bfc\u6027\u3002\u4e00\u4e2a\u65e0\u8111\u9884\u6d4b\u6240\u6709\u90ae\u4ef6\u90fd\u6b63\u5e38\u7684\u6a21\u578b&#xff0c;\u51c6\u786e\u7387\u4e5f\u80fd\u8fbe\u523099%\u3002<\/p>\n<\/li>\n<li>\n<p>\u7cbe\u786e\u7387&#xff08;Precision&#xff09;\u4e0e\u53ec\u56de\u7387&#xff08;Recall&#xff09;<\/p>\n<ul>\n<li>\u7cbe\u786e\u7387\u00a0Precision &#061; TP \/ (TP &#043; FP)&#xff1a;\u5728\u6240\u6709\u88ab\u6a21\u578b\u9884\u6d4b\u4e3a\u201c\u6b63\u201d\u7684\u6837\u672c\u4e2d&#xff0c;\u6709\u591a\u5c11\u662f\u771f\u6b63\u7684\u6b63\u6837\u672c\u3002\u5b83\u8861\u91cf\u7684\u662f\u6a21\u578b\u7684**\u201c\u67e5\u51c6\u7387\u201d**&#xff0c;\u5173\u5fc3\u7684\u662f\u201c\u522b\u628a\u574f\u4eba\u5f53\u597d\u4eba\u201d\u3002<\/li>\n<li>\u53ec\u56de\u7387\u00a0Recall &#061; TP \/ (TP &#043; FN)&#xff1a;\u5728\u6240\u6709\u771f\u5b9e\u4e3a\u201c\u6b63\u201d\u7684\u6837\u672c\u4e2d&#xff0c;\u6709\u591a\u5c11\u88ab\u6a21\u578b\u6210\u529f\u5730\u627e\u4e86\u51fa\u6765\u3002\u5b83\u8861\u91cf\u7684\u662f\u6a21\u578b\u7684**\u201c\u67e5\u5168\u7387\u201d**&#xff0c;\u5173\u5fc3\u7684\u662f\u201c\u522b\u653e\u8fc7\u4efb\u4f55\u4e00\u4e2a\u574f\u4eba\u201d\u3002 \u7cbe\u786e\u7387\u548c\u53ec\u56de\u7387\u5f80\u5f80\u662f\u4e00\u5bf9\u77db\u76fe\u7684\u6307\u6807\u3002\u5728\u75be\u75c5\u8bca\u65ad\u4e2d&#xff0c;\u6211\u4eec\u66f4\u5173\u5fc3\u9ad8\u53ec\u56de\u7387&#xff08;\u5b81\u53ef\u8bef\u8bca&#xff0c;\u4e0d\u80fd\u6f0f\u8bca&#xff09;&#xff1b;\u5728\u5783\u573e\u90ae\u4ef6\u8fc7\u6ee4\u4e2d&#xff0c;\u6211\u4eec\u66f4\u5173\u5fc3\u9ad8\u7cbe\u786e\u7387&#xff08;\u5b81\u53ef\u653e\u8fc7\u4e00\u4e9b\u5783\u573e\u90ae\u4ef6&#xff0c;\u4e0d\u80fd\u628a\u91cd\u8981\u90ae\u4ef6\u9519\u5224\u4e3a\u5783\u573e&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>F1\u5206\u6570&#xff08;F1-Score&#xff09; F1 &#061; 2 * (Precision * Recall) \/ (Precision &#043; Recall) \u5b83\u662f\u7cbe\u786e\u7387\u548c\u53ec\u56de\u7387\u7684\u8c03\u548c\u5e73\u5747\u6570&#xff0c;\u662f\u4e00\u4e2a\u517c\u987e\u4e24\u8005\u7684\u7efc\u5408\u6027\u6307\u6807\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u56de\u5f52\u6a21\u578b\u8bc4\u4f30\u6307\u6807<\/p>\n<ul>\n<li>\n<p>\u5747\u65b9\u8bef\u5dee&#xff08;MSE&#xff09;\u4e0e\u5747\u65b9\u6839\u8bef\u5dee&#xff08;RMSE&#xff09; MSE &#061; (1\/n) * \u03a3(y &#8211; \u0177)\u00b2 RMSE &#061; sqrt(MSE) \u5b83\u4eec\u8861\u91cf\u7684\u662f\u9884\u6d4b\u503c\u4e0e\u771f\u5b9e\u503c\u4e4b\u5dee\u7684\u5e73\u65b9\u7684\u5747\u503c\u3002RMSE\u56e0\u4e3a\u5f00\u4e86\u6839\u53f7&#xff0c;\u5176\u91cf\u7eb2\u4e0e\u539f\u59cb\u6570\u636e\u76f8\u540c&#xff0c;\u66f4\u6613\u4e8e\u89e3\u91ca\u3002\u5b83\u4eec\u5bf9\u8f83\u5927\u7684\u8bef\u5dee&#xff08;\u5f02\u5e38\u503c&#xff09;\u60e9\u7f5a\u66f4\u91cd\u3002<\/p>\n<\/li>\n<li>\n<p>\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee&#xff08;MAE&#xff09; MAE &#061; (1\/n) * \u03a3|y &#8211; \u0177| \u5b83\u8861\u91cf\u7684\u662f\u9884\u6d4b\u503c\u4e0e\u771f\u5b9e\u503c\u4e4b\u5dee\u7684\u7edd\u5bf9\u503c\u7684\u5747\u503c\u3002\u5b83\u5bf9\u6240\u6709\u8bef\u5dee\u4e00\u89c6\u540c\u4ec1&#xff0c;\u5bf9\u5f02\u5e38\u503c\u7684\u9c81\u68d2\u6027\u6bd4MSE\/RMSE\u66f4\u597d\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>3.3.2 \u504f\u5dee\u4e0e\u65b9\u5dee&#xff1a;\u6267\u7740\u4e8e\u4e00\u4e0e\u6e38\u79fb\u4e0d\u5b9a<\/h5>\n<p>\u8fd9\u662f\u7406\u89e3\u6a21\u578b\u884c\u4e3a\u3001\u8bca\u65ad\u95ee\u9898\u7684\u6838\u5fc3\u7406\u8bba\u3002<\/p>\n<p>\u6982\u5ff5\u8fa8\u6790 \u4e00\u4e2a\u6a21\u578b\u7684\u6cdb\u5316\u8bef\u5dee&#xff0c;\u53ef\u4ee5\u5206\u89e3\u4e3a\u504f\u5dee\u3001\u65b9\u5dee\u548c\u566a\u58f0\u4e09\u90e8\u5206\u3002\u6211\u4eec\u4e3b\u8981\u5173\u6ce8\u524d\u4e24\u8005\u3002<\/p>\n<ul>\n<li>\u504f\u5dee&#xff08;Bias&#xff09;&#xff1a;\u63cf\u8ff0\u7684\u662f\u6a21\u578b\u7684\u9884\u6d4b\u503c\u7684\u671f\u671b\u4e0e\u771f\u5b9e\u503c\u4e4b\u95f4\u7684\u5dee\u8ddd\u3002\u9ad8\u504f\u5dee\u6e90\u4e8e\u6a21\u578b\u672c\u8eab\u7684\u201c\u504f\u89c1\u201d\u6216\u201c\u56fa\u6267\u201d&#xff0c;\u5b83\u65e0\u6cd5\u6355\u6349\u5230\u6570\u636e\u7684\u771f\u5b9e\u89c4\u5f8b\u3002\u9ad8\u504f\u5dee\u901a\u5e38\u8868\u73b0\u4e3a\u6b20\u62df\u5408&#xff08;Underfitting&#xff09;\u3002<\/li>\n<li>\u65b9\u5dee&#xff08;Variance&#xff09;&#xff1a;\u63cf\u8ff0\u7684\u662f\u6a21\u578b\u5728\u4e0d\u540c\u8bad\u7ec3\u96c6\u4e0a\u8fdb\u884c\u8bad\u7ec3\u65f6&#xff0c;\u5176\u9884\u6d4b\u7ed3\u679c\u7684\u6ce2\u52a8\u6027\u6216\u4e0d\u7a33\u5b9a\u6027\u3002\u9ad8\u65b9\u5dee\u6e90\u4e8e\u6a21\u578b\u5bf9\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u566a\u58f0\u548c\u7ec6\u8282\u8fc7\u5ea6\u201c\u654f\u611f\u201d&#xff0c;\u628a\u5076\u7136\u5f53\u5fc5\u7136\u3002\u9ad8\u65b9\u5dee\u901a\u5e38\u8868\u73b0\u4e3a\u8fc7\u62df\u5408&#xff08;Overfitting&#xff09;\u3002<\/li>\n<\/ul>\n<p>\u504f\u5dee-\u65b9\u5dee\u7684\u6743\u8861&#xff08;Bias-Variance Tradeoff&#xff09; \u8fd9\u662f\u673a\u5668\u5b66\u4e60\u4e2d\u4e00\u4e2a\u6c38\u6052\u7684\u6838\u5fc3\u77db\u76fe\u3002<\/p>\n<ul>\n<li>\u4e00\u4e2a\u7b80\u5355\u7684\u6a21\u578b&#xff08;\u5982\u7ebf\u6027\u56de\u5f52&#xff09;&#xff0c;\u5bf9\u6570\u636e\u89c4\u5f8b\u7684\u5047\u8bbe\u5f88\u5f3a&#xff0c;\u504f\u5dee\u8f83\u9ad8&#xff0c;\u4f46\u5bf9\u4e0d\u540c\u8bad\u7ec3\u6570\u636e\u4e0d\u654f\u611f&#xff0c;\u65b9\u5dee\u8f83\u4f4e\u3002<\/li>\n<li>\u4e00\u4e2a\u590d\u6742\u7684\u6a21\u578b&#xff08;\u5982\u9ad8\u9636\u591a\u9879\u5f0f\u56de\u5f52\u6216\u6df1\u5ea6\u51b3\u7b56\u6811&#xff09;&#xff0c;\u80fd\u62df\u5408\u5404\u79cd\u590d\u6742\u89c4\u5f8b&#xff0c;\u504f\u5dee\u8f83\u4f4e&#xff0c;\u4f46\u6781\u6613\u5b66\u4e60\u5230\u8bad\u7ec3\u6570\u636e\u7684\u566a\u58f0&#xff0c;\u5bfc\u81f4\u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0a\u7ed3\u679c\u5dee\u5f02\u5de8\u5927&#xff0c;\u65b9\u5dee\u8f83\u9ad8\u3002\u00a0\u6211\u4eec\u7684\u76ee\u6807&#xff0c;\u4e0d\u662f\u8ffd\u6c42\u96f6\u504f\u5dee\u6216\u96f6\u65b9\u5dee&#xff0c;\u800c\u662f\u5728\u4e24\u8005\u4e4b\u95f4\u627e\u5230\u4e00\u4e2a\u6700\u4f73\u7684\u5e73\u8861\u70b9&#xff0c;\u4f7f\u5f97\u603b\u7684\u6cdb\u5316\u8bef\u5dee\u6700\u5c0f\u3002<\/li>\n<\/ul>\n<h5>3.3.3 \u8fc7\u62df\u5408\u4e0e\u6b20\u62df\u5408&#xff1a;\u5b66\u201c\u5e9f\u201d\u4e86\u4e0e\u6ca1\u5b66\u4f1a<\/h5>\n<p>\u8fd9\u662f\u504f\u5dee\u4e0e\u65b9\u5dee\u5728\u6a21\u578b\u8bad\u7ec3\u4e2d\u7684\u5177\u4f53\u8868\u73b0\u3002<\/p>\n<p>\u73b0\u8c61\u4e0e\u8bca\u65ad&#xff1a;<\/p>\n<p>\u6b20\u62df\u5408&#xff08;Underfitting&#xff09;&#xff1a;\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6\u4e0a\u8868\u73b0\u90fd\u5f88\u5dee\u3002<\/p>\n<ul>\n<li>\u75c5\u56e0&#xff1a;\u6a21\u578b\u592a\u7b80\u5355\u4e86&#xff08;\u9ad8\u504f\u5dee&#xff09;&#xff0c;\u8fde\u8bad\u7ec3\u6570\u636e\u7684\u57fa\u672c\u89c4\u5f8b\u90fd\u6ca1\u5b66\u4f1a\u3002<\/li>\n<li>\u8bca\u65ad&#xff1a;\u89c2\u5bdf\u5230\u8bad\u7ec3\u8bef\u5dee\u548c\u6d4b\u8bd5\u8bef\u5dee\u90fd\u5f88\u9ad8\u3002<\/li>\n<\/ul>\n<p>\u8fc7\u62df\u5408&#xff08;Overfitting&#xff09;&#xff1a;\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u4e0a\u8868\u73b0\u6781\u597d&#xff0c;\u4f46\u5728\u672a\u89c1\u8fc7\u7684\u6d4b\u8bd5\u96c6\u4e0a\u8868\u73b0\u5f88\u5dee\u3002<\/p>\n<ul>\n<li>\u75c5\u56e0&#xff1a;\u6a21\u578b\u592a\u590d\u6742\u4e86&#xff08;\u9ad8\u65b9\u5dee&#xff09;&#xff0c;\u628a\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u566a\u58f0\u548c\u5076\u7136\u6027\u4e5f\u5f53\u6210\u4e86\u666e\u9002\u89c4\u5f8b\u6765\u5b66\u4e60&#xff0c;\u5931\u53bb\u4e86\u201c\u6cdb\u5316\u201d\u5230\u65b0\u6570\u636e\u7684\u80fd\u529b\u3002<\/li>\n<li>\u8bca\u65ad&#xff1a;\u89c2\u5bdf\u5230\u8bad\u7ec3\u8bef\u5dee\u5f88\u4f4e&#xff0c;\u4f46\u6d4b\u8bd5\u8bef\u5dee\u8fdc\u9ad8\u4e8e\u8bad\u7ec3\u8bef\u5dee\u3002<\/li>\n<\/ul>\n<p>\u201c\u8bca\u65ad\u201d\u65b9\u6cd5&#xff1a;\u5b66\u4e60\u66f2\u7ebf \u7ed8\u5236\u5b66\u4e60\u66f2\u7ebf&#xff08;Learning Curves&#xff09;\u662f\u4e00\u79cd\u6709\u6548\u7684\u8bca\u65ad\u65b9\u6cd5\u3002\u8be5\u66f2\u7ebf\u7684\u6a2a\u8f74\u662f\u8bad\u7ec3\u6837\u672c\u7684\u6570\u91cf&#xff0c;\u7eb5\u8f74\u662f\u8bef\u5dee&#xff08;\u6216\u51c6\u786e\u7387&#xff09;\u3002\u901a\u8fc7\u89c2\u5bdf\u8bad\u7ec3\u8bef\u5dee\u66f2\u7ebf\u548c\u9a8c\u8bc1\u8bef\u5dee\u66f2\u7ebf\u7684\u8d70\u52bf\u548c\u5dee\u8ddd&#xff0c;\u53ef\u4ee5\u6e05\u6670\u5730\u5224\u65ad\u51fa\u6a21\u578b\u6b63\u5904\u4e8e\u6b20\u62df\u5408\u3001\u8fc7\u62df\u5408\u8fd8\u662f\u7406\u60f3\u72b6\u6001\u3002<\/p>\n<h5>3.3.4 \u4ea4\u53c9\u9a8c\u8bc1&#xff1a;\u66f4\u516c\u5e73\u7684\u201c\u6a21\u62df\u8003\u8bd5\u201d<\/h5>\n<p>\u4e3a\u4f55\u9700\u8981 \u5728\u6a21\u578b\u5f00\u53d1\u4e2d&#xff0c;\u6211\u4eec\u901a\u5e38\u4f1a\u5c06\u6570\u636e\u5212\u5206\u4e3a\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6\u3002\u4f46\u8fd9\u79cd\u4e00\u6b21\u6027\u7684\u5212\u5206\u5177\u6709\u5f88\u5927\u7684\u5076\u7136\u6027\u3002\u53ef\u80fd\u6211\u4eec\u78b0\u5de7\u5206\u5230\u4e86\u4e00\u4efd\u201c\u7b80\u5355\u201d\u7684\u6d4b\u8bd5\u96c6&#xff0c;\u5bfc\u81f4\u6a21\u578b\u5f97\u5206\u865a\u9ad8&#xff1b;\u4e5f\u53ef\u80fd\u78b0\u5de7\u5206\u5230\u4e86\u4e00\u4efd\u201c\u56f0\u96be\u201d\u7684\u6d4b\u8bd5\u96c6&#xff0c;\u5bfc\u81f4\u597d\u6a21\u578b\u88ab\u51a4\u6789\u3002<\/p>\n<p>K\u6298\u4ea4\u53c9\u9a8c\u8bc1&#xff08;K-Fold Cross-Validation&#xff09; \u4e3a\u4e86\u5f97\u5230\u4e00\u4e2a\u66f4\u7a33\u5065\u3001\u66f4\u516c\u5e73\u7684\u6a21\u578b\u8bc4\u4f30\u7ed3\u679c&#xff0c;\u6211\u4eec\u91c7\u7528\u4ea4\u53c9\u9a8c\u8bc1\u3002K\u6298\u4ea4\u53c9\u9a8c\u8bc1\u662f\u5176\u4e2d\u6700\u5e38\u7528\u7684\u65b9\u6cd5\u3002<\/p>\n<p>\u8fd9\u79cd\u65b9\u6cd5\u8ba9\u6240\u6709\u6570\u636e\u90fd\u6709\u673a\u4f1a\u6210\u4e3a\u9a8c\u8bc1\u96c6&#xff0c;\u5927\u5927\u964d\u4f4e\u4e86\u5355\u6b21\u5212\u5206\u5e26\u6765\u7684\u5076\u7136\u6027&#xff0c;\u5f97\u51fa\u7684\u8bc4\u4f30\u7ed3\u679c\u66f4\u5177\u8bf4\u670d\u529b\u3002<\/p>\n<li>\u5c06\u5168\u90e8\u8bad\u7ec3\u6570\u636e\u968f\u673a\u5206\u6210\u00a0K\u00a0\u4e2a\u5927\u5c0f\u76f8\u7b49\u7684\u3001\u4e92\u4e0d\u76f8\u4ea4\u7684\u5b50\u96c6&#xff08;\u79f0\u4e3a\u201c\u6298\u201d&#xff09;\u3002<\/li>\n<li>\u8fdb\u884c\u00a0K\u00a0\u8f6e\u201c\u6a21\u62df\u8003\u8bd5\u201d&#xff1a;\n<ul>\n<li>\u5728\u6bcf\u4e00\u8f6e\u4e2d&#xff0c;\u9009\u62e9\u5176\u4e2d 1 \u4e2a\u201c\u6298\u201d\u4f5c\u4e3a\u9a8c\u8bc1\u96c6&#xff08;\u6d4b\u8bd5\u6a21\u578b&#xff09;&#xff0c;\u5176\u4f59 K-1 \u4e2a\u201c\u6298\u201d\u5408\u5e76\u8d77\u6765\u4f5c\u4e3a\u8bad\u7ec3\u96c6&#xff08;\u8bad\u7ec3\u6a21\u578b&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u6700\u7ec8&#xff0c;\u6211\u4eec\u5f97\u5230 K \u4e2a\u9a8c\u8bc1\u96c6\u4e0a\u7684\u5f97\u5206&#xff0c;\u5c06\u5b83\u4eec\u53d6\u5e73\u5747\u503c&#xff0c;\u4f5c\u4e3a\u6a21\u578b\u7684\u6700\u7ec8\u8bc4\u4f30\u5206\u6570\u3002<\/li>\n<p>\u4ee3\u7801\u5b9e\u6218 Scikit-Learn\u4e2d\u7684cross_val_score\u51fd\u6570\u8ba9\u6267\u884c\u4ea4\u53c9\u9a8c\u8bc1\u53d8\u5f97\u5f02\u5e38\u7b80\u5355&#xff0c;\u6211\u4eec\u5728\u4e4b\u524d\u7684\u968f\u673a\u68ee\u6797\u4ee3\u7801\u5b9e\u6218\u4e2d\u5df2\u7ecf\u6f14\u793a\u4e86\u5b83\u7684\u7528\u6cd5\u3002\u5b83\u5c01\u88c5\u4e86\u6574\u4e2aK\u6298\u5212\u5206\u548c\u5faa\u73af\u8bad\u7ec3\u8bc4\u4f30\u7684\u8fc7\u7a0b&#xff0c;\u4e00\u884c\u4ee3\u7801\u5c31\u80fd\u8fd4\u56deK\u6b21\u201c\u6a21\u8003\u201d\u7684\u6210\u7ee9\u5217\u8868\u3002<\/p>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u5bf9\u673a\u5668\u5b66\u4e60\u7ecf\u5178\u7b97\u6cd5\u7684\u56de\u987e\u3002\u4ece\u4e09\u5927\u8303\u5f0f\u7684\u9ad8\u5c4b\u5efa\u74f4&#xff0c;\u5230\u7ebf\u6027\u6a21\u578b\u3001\u51b3\u7b56\u6811\u3001\u968f\u673a\u68ee\u6797\u3001\u652f\u6301\u5411\u91cf\u673a\u7b49\u6838\u5fc3\u7b97\u6cd5\u7684\u6df1\u5165\u5256\u6790&#xff0c;\u518d\u5230\u4e00\u5957\u79d1\u5b66\u7684\u6a21\u578b\u8bc4\u4f30\u4e0e\u4f18\u5316\u65b9\u6cd5\u8bba&#xff0c;\u8bfb\u8005\u5e94\u5f53\u5df2\u7ecf\u5efa\u7acb\u8d77\u4e86\u575a\u5b9e\u7684\u673a\u5668\u5b66\u4e60\u57fa\u7840\u3002<\/p>\n<p>\u8fd9\u4e9b\u7ecf\u5178\u6a21\u578b&#xff0c;\u4e0d\u4ec5\u662f\u89e3\u51b3\u73b0\u5b9e\u95ee\u9898\u7684\u6709\u529b\u5de5\u5177&#xff0c;\u66f4\u662f\u901a\u5f80\u6df1\u5ea6\u5b66\u4e60\u6bbf\u5802\u7684\u5fc5\u7ecf\u4e4b\u8def\u3002\u5b83\u4eec\u6240\u8574\u542b\u7684\u5173\u4e8e\u62df\u5408\u3001\u6cdb\u5316\u3001\u504f\u5dee\u3001\u65b9\u5dee\u3001\u96c6\u6210\u7b49\u6838\u5fc3\u601d\u60f3&#xff0c;\u5c06\u5728\u540e\u7eed\u7684\u7ae0\u8282\u4e2d\u53cd\u590d\u51fa\u73b0\u5e76\u4e0d\u65ad\u5347\u534e\u3002\u5e26\u7740\u8fd9\u4e9b\u5b9d\u8d35\u7684\u77e5\u8bc6\u548c\u76f4\u89c9&#xff0c;\u73b0\u5728&#xff0c;\u6211\u4eec\u7ec8\u4e8e\u51c6\u5907\u597d&#xff0c;\u53bb\u8fce\u63a5\u4e0b\u4e00\u7ae0\u2014\u2014\u771f\u6b63\u8e0f\u5165\u795e\u7ecf\u7f51\u7edc\u7684\u4e16\u754c\u3002<\/p>\n<hr \/>\n<h2>\u7b2c\u4e8c\u90e8\u5206&#xff1a;\u6838\u5fc3\u7bc7 \u2014\u2014 \u6df1\u5165\u795e\u7ecf\u7f51\u7edc\u7684\u6bbf\u5802<\/h2>\n<hr \/>\n<h3>\u7b2c\u56db\u7ae0&#xff1a;\u795e\u7ecf\u7f51\u7edc\u57fa\u7840<\/h3>\n<p>\u4ece\u4eff\u751f\u5b66\u5230\u6570\u5b66\u6a21\u578b \u2014\u2014 \u667a\u80fd\u7684\u8ba1\u7b97\u4e4b\u8def<\/p>\n<p>\u6b22\u8fce\u6765\u5230\u6df1\u5ea6\u5b66\u4e60\u7684\u6838\u5fc3\u8179\u5730\u3002\u5728\u524d\u9762\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u56de\u987e\u4e86\u7ecf\u5178\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5&#xff0c;\u5b83\u4eec\u5982\u540c\u80fd\u5de5\u5de7\u5320&#xff0c;\u4f7f\u7528\u7740\u8bbe\u8ba1\u7cbe\u5de7\u7684\u5de5\u5177\u3002\u4ece\u672c\u7ae0\u5f00\u59cb&#xff0c;\u6211\u4eec\u5c06\u5b66\u4e60\u5982\u4f55\u6784\u5efa\u4e00\u4e2a\u80fd\u591f\u81ea\u6211\u5b66\u4e60\u3001\u81ea\u6211\u6f14\u5316\u7684\u201c\u6709\u673a\u4f53\u201d\u2014\u2014\u795e\u7ecf\u7f51\u7edc\u3002<\/p>\n<p>\u8fd9\u4e2a\u5b8f\u5927\u6784\u60f3\u7684\u6700\u521d\u706b\u82b1&#xff0c;\u6e90\u4e8e\u4eba\u7c7b\u5bf9\u81ea\u5df1\u667a\u6167\u6e90\u6cc9\u2014\u2014\u5927\u8111\u2014\u2014\u7684\u63a2\u5bfb\u3002\u79d1\u5b66\u5bb6\u4eec\u6e34\u671b\u5728\u51b0\u51b7\u7684\u7845\u57fa\u82af\u7247\u4e0a&#xff0c;\u6a21\u62df\u51fa\u7531\u4ebf\u4e07\u795e\u7ecf\u5143\u6784\u6210\u7684\u751f\u7269\u795e\u7ecf\u7f51\u7edc\u7684\u60ca\u4eba\u80fd\u529b\u3002\u8fd9&#xff0c;\u4fbf\u662f\u795e\u7ecf\u7f51\u7edc\u4eff\u751f\u5b66\u7684\u6d6a\u6f2b\u8d77\u70b9\u3002\u7136\u800c&#xff0c;\u6d6a\u6f2b\u7684\u7075\u611f\u5fc5\u987b\u7ecf\u8fc7\u4e25\u8c28\u7684\u6570\u5b66\u62bd\u8c61\u548c\u5de5\u7a0b\u5b9e\u8df5\u7684\u6dec\u70bc&#xff0c;\u624d\u80fd\u5316\u4e3a\u771f\u6b63\u5f3a\u5927\u7684\u8ba1\u7b97\u6a21\u578b\u3002\u73b0\u4ee3\u795e\u7ecf\u7f51\u7edc&#xff0c;\u65e9\u5df2\u8d85\u8d8a\u4e86\u7b80\u5355\u7684\u6a21\u4eff&#xff0c;\u53d1\u5c55\u6210\u4e3a\u4e00\u95e8\u535a\u5927\u7cbe\u6df1\u7684\u3001\u4ee5\u6570\u5b66\u548c\u8ba1\u7b97\u673a\u79d1\u5b66\u4e3a\u57fa\u7840\u7684\u72ec\u7acb\u5b66\u79d1\u3002<\/p>\n<p>\u5728\u672c\u7ae0\u7684\u65c5\u7a0b\u4e2d&#xff0c;\u6211\u4eec\u5c06\u8ffd\u968f\u5148\u9a71\u7684\u811a\u6b65&#xff0c;\u5b8c\u6210\u8fd9\u6b21\u4ece\u4eff\u751f\u5230\u5efa\u6a21\u7684\u4f1f\u5927\u8de8\u8d8a\u3002\u6211\u4eec\u5c06\u4ece\u6784\u6210\u667a\u80fd\u7684\u6700\u57fa\u672c\u5355\u5143\u201c\u795e\u7ecf\u5143\u201d\u51fa\u53d1&#xff0c;\u89c2\u5bdf\u5b83\u662f\u5982\u4f55\u4ece\u751f\u7269\u6982\u5ff5\u88ab\u62bd\u8c61\u4e3a\u6570\u5b66\u51fd\u6570\u3002\u63a5\u7740&#xff0c;\u6211\u4eec\u5c06\u7528\u5b83\u642d\u5efa\u51fa\u6700\u7b80\u5355\u7684\u201c\u611f\u77e5\u673a\u201d\u7f51\u7edc&#xff0c;\u5e76\u76f4\u9762\u5176\u5386\u53f2\u6027\u7684\u5c40\u9650\u3002\u7136\u540e&#xff0c;\u6211\u4eec\u5c06\u89c1\u8bc1\u201c\u6df1\u5ea6\u201d\u7684\u8bde\u751f\u2014\u2014\u901a\u8fc7\u5f15\u5165\u9690\u85cf\u5c42\u6784\u5efa\u591a\u5c42\u611f\u77e5\u673a&#xff0c;\u770b\u5b83\u5982\u4f55\u7a81\u7834\u74f6\u9888\u3002\u6211\u4eec\u8fd8\u5c06\u5de1\u793c\u8d4b\u4e88\u7f51\u7edc\u975e\u7ebf\u6027\u529b\u91cf\u7684\u201c\u6fc0\u6d3b\u51fd\u6570\u201d&#xff0c;\u5e76\u4e3a\u7f51\u7edc\u88c5\u4e0a\u8bc4\u5224\u662f\u975e\u7684\u201c\u635f\u5931\u51fd\u6570\u201d\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u63ed\u5f00\u7f51\u7edc\u5b66\u4e60\u7684\u7ec8\u6781\u5965\u79d8\u2014\u2014\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5&#xff0c;\u770b\u68af\u5ea6\u4e0b\u964d\u4e0e\u94fe\u5f0f\u6cd5\u5219\u5982\u4f55\u5b8c\u7f8e\u7ed3\u5408&#xff0c;\u9a71\u52a8\u6574\u4e2a\u7f51\u7edc\u8fdb\u884c\u81ea\u6211\u4f18\u5316\u3002<\/p>\n<p>\u51c6\u5907\u597d\u4e86\u5417&#xff1f;\u8ba9\u6211\u4eec\u6df1\u5165\u6e90\u5934&#xff0c;\u63a2\u5bfb\u795e\u7ecf\u7f51\u7edc\u601d\u60f3\u7684\u6700\u521d\u7075\u611f&#xff0c;\u63ed\u5f00\u795e\u7ecf\u7f51\u7edc\u7684\u795e\u79d8\u9762\u7eb1&#xff0c;\u5e76\u6700\u7ec8\u4eb2\u624b\u6784\u5efa\u8d77\u901a\u5f80\u4eba\u5de5\u667a\u80fd\u7684\u9636\u68af\u3002<\/p>\n<hr \/>\n<h4>4.1 \u4ece\u751f\u7269\u795e\u7ecf\u5143\u5230\u4eba\u5de5\u795e\u7ecf\u5143&#xff1a;\u6a21\u578b\u7684\u7075\u611f\u6765\u6e90<\/h4>\n<p>\u5728\u8ba1\u7b97\u673a\u79d1\u5b66\u7684\u6bbf\u5802\u91cc&#xff0c;\u5f88\u5c11\u6709\u54ea\u4e2a\u9886\u57df\u50cf\u795e\u7ecf\u7f51\u7edc\u4e00\u6837&#xff0c;\u5176\u6700\u521d\u7684\u6784\u60f3\u4e0e\u751f\u547d\u79d1\u5b66\u7684\u8054\u7cfb\u5982\u6b64\u7d27\u5bc6\u3002\u4e3a\u4e86\u521b\u9020\u667a\u80fd&#xff0c;\u5148\u9a71\u4eec\u9996\u5148\u5c06\u76ee\u5149\u6295\u5411\u4e86\u5df2\u77e5\u7684\u3001\u6700\u9ad8\u6548\u7684\u667a\u80fd\u7cfb\u7edf\u2014\u2014\u4eba\u7c7b\u5927\u8111\u3002<\/p>\n<h5>4.1.1 \u751f\u7269\u795e\u7ecf\u5143&#xff08;Biological Neuron&#xff09;&#xff1a;\u751f\u547d\u667a\u80fd\u7684\u57fa\u77f3<\/h5>\n<p>\u6211\u4eec\u7684\u5927\u8111\u662f\u7531\u5927\u7ea6860\u4ebf\u4e2a\u79f0\u4e3a\u795e\u7ecf\u5143\u7684\u7279\u6b8a\u7ec6\u80de\u7ec4\u6210\u7684\u590d\u6742\u7f51\u7edc\u3002\u8fd9\u4e9b\u795e\u7ecf\u5143\u662f\u4fe1\u606f\u5904\u7406\u7684\u57fa\u672c\u5355\u4f4d&#xff0c;\u5b83\u4eec\u534f\u540c\u5de5\u4f5c&#xff0c;\u6784\u6210\u4e86\u6211\u4eec\u7684\u601d\u60f3\u3001\u8bb0\u5fc6\u548c\u611f\u77e5\u3002\u4e00\u4e2a\u5178\u578b\u7684\u751f\u7269\u795e\u7ecf\u5143&#xff0c;\u5176\u7ed3\u6784\u548c\u5de5\u4f5c\u65b9\u5f0f\u53ef\u4ee5\u88ab\u7b80\u5316\u4e3a\u4ee5\u4e0b\u51e0\u4e2a\u5173\u952e\u90e8\u5206&#xff1a;<\/p>\n<p>\u6838\u5fc3\u7ed3\u6784<\/p>\n<ul>\n<li>\u6811\u7a81&#xff08;Dendrites&#xff09;&#xff1a;\u5b83\u4eec\u50cf\u5929\u7ebf\u4e00\u6837&#xff0c;\u662f\u795e\u7ecf\u5143\u7684\u201c\u8f93\u5165\u7aef\u201d\u3002\u6811\u7a81\u4ece\u6210\u5343\u4e0a\u4e07\u4e2a\u5176\u4ed6\u7684\u4e0a\u6e38\u795e\u7ecf\u5143\u90a3\u91cc\u63a5\u6536\u7535\u5316\u5b66\u4fe1\u53f7\u3002<\/li>\n<li>\u7ec6\u80de\u4f53&#xff08;Soma&#xff09;&#xff1a;\u8fd9\u662f\u795e\u7ecf\u5143\u7684\u201c\u5904\u7406\u4e2d\u5fc3\u201d\u3002\u5b83\u5c06\u6240\u6709\u4ece\u6811\u7a81\u63a5\u6536\u5230\u7684\u4fe1\u53f7\u8fdb\u884c\u6574\u5408\u3001\u7d2f\u52a0\u3002<\/li>\n<li>\u8f74\u7a81&#xff08;Axon&#xff09;&#xff1a;\u8fd9\u662f\u4e00\u6761\u957f\u957f\u7684\u201c\u8f93\u51fa\u7ebf\u201d\u3002\u5f53\u7ec6\u80de\u4f53\u6574\u5408\u540e\u7684\u4fe1\u53f7\u5f3a\u5ea6\u8fbe\u5230\u67d0\u4e2a\u4e34\u754c\u70b9\u65f6&#xff0c;\u8f74\u7a81\u4f1a\u5c06\u4e00\u4e2a\u6807\u51c6\u7684\u7535\u4fe1\u53f7\u4f20\u9012\u7ed9\u4e0b\u6e38\u7684\u5176\u4ed6\u795e\u7ecf\u5143\u3002<\/li>\n<\/ul>\n<p>\u5de5\u4f5c\u539f\u7406&#xff1a; \u201c\u5168\u6216\u65e0\u201d\u7684\u6fc0\u6d3b\u6a21\u5f0f \u751f\u7269\u795e\u7ecf\u5143\u7684\u5de5\u4f5c\u65b9\u5f0f&#xff0c;\u9075\u5faa\u4e00\u79cd**\u201c\u5168\u6216\u65e0\u201d&#xff08;All-or-None&#xff09;\u7684\u539f\u5219\u3002\u7ec6\u80de\u4f53\u6301\u7eed\u5730\u7d2f\u52a0\u6765\u81ea\u6811\u7a81\u7684\u8f93\u5165\u4fe1\u53f7\u3002\u53ea\u6709\u5f53\u8fd9\u4e9b\u4fe1\u53f7\u7684\u603b\u548c&#xff0c;\u5728\u77ed\u65f6\u95f4\u5185\u8d85\u8fc7\u4e00\u4e2a\u7279\u5b9a\u7684\u6fc0\u6d3b\u9608\u503c&#xff08;Activation Threshold&#xff09;**\u65f6&#xff0c;\u8fd9\u4e2a\u795e\u7ecf\u5143\u624d\u4f1a\u88ab\u201c\u6fc0\u6d3b\u201d&#xff08;\u6216\u79f0\u201c\u5174\u594b\u201d\u3001\u201c\u653e\u7535\u201d&#xff09;\u3002<\/p>\n<p>\u4e00\u65e6\u88ab\u6fc0\u6d3b&#xff0c;\u5b83\u5c31\u4f1a\u901a\u8fc7\u8f74\u7a81&#xff0c;\u5411\u4e0b\u6e38\u7684\u795e\u7ecf\u5143\u53d1\u9001\u4e00\u4e2a\u5f3a\u5ea6\u548c\u6ce2\u5f62\u90fd\u5b8c\u5168\u76f8\u540c\u7684\u6807\u51c6\u7535\u5316\u5b66\u8109\u51b2\u3002\u5982\u679c\u4fe1\u53f7\u603b\u548c\u6ca1\u6709\u8fbe\u5230\u9608\u503c&#xff0c;\u795e\u7ecf\u5143\u5219\u4fdd\u6301\u9759\u9ed8&#xff0c;\u4e0d\u4ea7\u751f\u4efb\u4f55\u8f93\u51fa\u3002\u8fd9\u79cd\u975e0\u53731\u7684\u5f00\u5173\u7279\u6027&#xff0c;\u662f\u751f\u7269\u667a\u80fd\u8fdb\u884c\u4fe1\u606f\u7f16\u7801\u548c\u4f20\u9012\u7684\u57fa\u7840\u3002<\/p>\n<h5>4.1.2 \u4eba\u5de5\u795e\u7ecf\u5143&#xff08;Artificial Neuron&#xff09;&#xff1a;\u4e00\u6b21\u4f18\u96c5\u7684\u6570\u5b66\u62bd\u8c61<\/h5>\n<p>\u65e9\u671f\u7684\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u8005\u4eec&#xff0c;\u88ab\u751f\u7269\u795e\u7ecf\u5143\u8fd9\u79cd\u7b80\u6d01\u800c\u5f3a\u5927\u7684\u5de5\u4f5c\u6a21\u5f0f\u6240\u542f\u53d1&#xff0c;\u8bd5\u56fe\u7528\u6570\u5b66\u7684\u8bed\u8a00\u6765\u5bf9\u5176\u8fdb\u884c\u5efa\u6a21\u3002\u8fd9\u4e2a\u6a21\u578b&#xff0c;\u5c31\u662f\u4eba\u5de5\u795e\u7ecf\u5143&#xff0c;\u5b83\u6784\u6210\u4e86\u6240\u6709\u795e\u7ecf\u7f51\u7edc\u7684\u6700\u57fa\u672c\u5355\u5143\u3002<\/p>\n<p>\u6a21\u578b\u7684\u5efa\u7acb<\/p>\n<p>\u8ba9\u6211\u4eec\u770b\u770b&#xff0c;\u751f\u7269\u8fc7\u7a0b\u662f\u5982\u4f55\u88ab\u4e00\u6b65\u6b65\u7ffb\u8bd1\u4e3a\u6570\u5b66\u8bed\u8a00\u7684&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u8f93\u5165\u4e0e\u6743\u91cd \u751f\u7269\u795e\u7ecf\u5143\u901a\u8fc7\u6811\u7a81\u63a5\u6536\u6765\u81ea\u591a\u4e2a\u4e0a\u6e38\u795e\u7ecf\u5143\u7684\u4fe1\u53f7\u3002\u5728\u4eba\u5de5\u795e\u7ecf\u5143\u6a21\u578b\u4e2d&#xff0c;\u8fd9\u88ab\u62bd\u8c61\u4e3a\u4e00\u7ec4\u8f93\u5165 x\u2081, x\u2082, &#8230;, x\u2099\u3002\u6bcf\u4e2a\u8f93\u5165&#xff0c;\u90fd\u4e0e\u4e00\u4e2a\u6743\u91cd w\u2081, w\u2082, &#8230;, w\u2099 \u76f8\u5173\u8054\u3002\u8fd9\u4e2a\u6743\u91cd w\u1d62&#xff0c;\u6a21\u62df\u4e86\u751f\u7269\u795e\u7ecf\u5143\u4e2d\u7a81\u89e6\u8fde\u63a5\u7684\u201c\u5f3a\u5ea6\u201d\u3002\u4e00\u4e2a\u5927\u7684\u6b63\u6743\u91cd&#xff0c;\u610f\u5473\u7740\u8fd9\u4e2a\u8f93\u5165\u4fe1\u53f7\u5bf9\u795e\u7ecf\u5143\u7684\u6fc0\u6d3b\u6709\u5f88\u5f3a\u7684\u201c\u5174\u594b\u201d\u4f5c\u7528&#xff1b;\u4e00\u4e2a\u5927\u7684\u8d1f\u6743\u91cd&#xff0c;\u5219\u4ee3\u8868\u6709\u5f88\u5f3a\u7684\u201c\u6291\u5236\u201d\u4f5c\u7528\u3002<\/p>\n<\/li>\n<li>\n<p>\u6574\u5408\u4e0e\u504f\u7f6e \u751f\u7269\u795e\u7ecf\u5143\u7684\u7ec6\u80de\u4f53\u8d1f\u8d23\u6574\u5408\u6240\u6709\u8f93\u5165\u4fe1\u53f7\u3002\u5728\u6570\u5b66\u6a21\u578b\u4e2d&#xff0c;\u8fd9\u4e2a\u8fc7\u7a0b\u88ab\u7b80\u5316\u4e3a\u5bf9\u6240\u6709\u8f93\u5165\u7684\u52a0\u6743\u6c42\u548c&#xff1a;z &#061; (w\u2081x\u2081 &#043; w\u2082x\u2082 &#043; &#8230; &#043; w\u2099x\u2099)\u3002 \u6b64\u5916&#xff0c;\u6a21\u578b\u8fd8\u5f15\u5165\u4e86\u4e00\u4e2a\u989d\u5916\u7684\u53c2\u6570&#xff0c;\u79f0\u4e3a\u504f\u7f6e&#xff08;Bias&#xff09;b\u3002\u52a0\u6743\u548c\u4f1a\u4e0e\u8fd9\u4e2a\u504f\u7f6e\u9879\u76f8\u52a0&#xff1a;z &#061; (\u03a3w\u1d62x\u1d62) &#043; b\u3002\u504f\u7f6e\u9879\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u795e\u7ecf\u5143\u56fa\u6709\u7684\u6fc0\u6d3b\u503e\u5411\u3002\u4e00\u4e2a\u5927\u7684\u6b63\u504f\u7f6e&#xff0c;\u610f\u5473\u7740\u8fd9\u4e2a\u795e\u7ecf\u5143\u672c\u8eab\u5c31\u66f4\u5bb9\u6613\u88ab\u6fc0\u6d3b&#xff0c;\u5373\u4f7f\u8f93\u5165\u4fe1\u53f7\u4e0d\u5f3a\u3002\u5b83\u5728\u6570\u5b66\u4e0a\u786e\u4fdd\u4e86\u5373\u4f7f\u6240\u6709\u8f93\u5165\u90fd\u4e3a0&#xff0c;\u795e\u7ecf\u5143\u4e5f\u80fd\u4ea7\u751f\u8f93\u51fa&#xff0c;\u589e\u52a0\u4e86\u6a21\u578b\u7684\u7075\u6d3b\u6027\u3002<\/p>\n<\/li>\n<li>\n<p>\u6fc0\u6d3b\u51fd\u6570 \u751f\u7269\u795e\u7ecf\u5143\u201c\u5168\u6216\u65e0\u201d\u7684\u6fc0\u6d3b\u673a\u5236&#xff0c;\u88ab\u4e00\u4e2a\u79f0\u4e3a\u6fc0\u6d3b\u51fd\u6570&#xff08;Activation Function&#xff09;f() \u7684\u6570\u5b66\u51fd\u6570\u6240\u53d6\u4ee3\u3002\u8fd9\u4e2a\u51fd\u6570\u63a5\u6536\u6574\u5408\u540e\u7684\u4fe1\u53f7 z \u4f5c\u4e3a\u8f93\u5165&#xff0c;\u5e76\u4ea7\u751f\u795e\u7ecf\u5143\u7684\u6700\u7ec8\u8f93\u51fa y\u3002\u6fc0\u6d3b\u51fd\u6570\u51b3\u5b9a\u4e86\u795e\u7ecf\u5143\u5728\u63a5\u6536\u5230\u7279\u5b9a\u5f3a\u5ea6\u7684\u4fe1\u53f7\u540e&#xff0c;\u5e94\u8be5\u505a\u51fa\u4f55\u79cd\u7a0b\u5ea6\u7684\u54cd\u5e94\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u6570\u5b66\u5f62\u5f0f<\/p>\n<p>\u7efc\u4e0a\u6240\u8ff0&#xff0c;\u4e00\u4e2a\u4eba\u5de5\u795e\u7ecf\u5143\u7684\u5b8c\u6574\u6570\u5b66\u8868\u8fbe\u5f62\u5f0f\u4e3a&#xff1a;<\/p>\n<p>y &#061; f( (w\u2081x\u2081 &#043; w\u2082x\u2082 &#043; &#8230; &#043; w\u2099x\u2099) &#043; b )<\/p>\n<p>\u4f7f\u7528\u6211\u4eec\u5728\u7b2c\u4e8c\u7ae0\u5b66\u8fc7\u7684\u7ebf\u6027\u4ee3\u6570\u77e5\u8bc6&#xff0c;\u8fd9\u4e2a\u516c\u5f0f\u53ef\u4ee5\u88ab\u66f4\u7b80\u6d01\u5730\u5199\u6210\u5411\u91cf\u5f62\u5f0f&#xff1a;<\/p>\n<p>y &#061; f(w\u1d40x &#043; b)<\/p>\n<p>\u5176\u4e2d&#xff1a;<\/p>\n<ul>\n<li>x\u00a0\u662f\u8f93\u5165\u7684\u7279\u5f81\u5411\u91cf\u3002<\/li>\n<li>w\u00a0\u662f\u6743\u91cd\u7684\u53c2\u6570\u5411\u91cf\u3002<\/li>\n<li>b\u00a0\u662f\u504f\u7f6e\u6807\u91cf\u3002<\/li>\n<li>w\u1d40x\u00a0\u662f\u6743\u91cd\u4e0e\u8f93\u5165\u7684\u70b9\u79ef\u3002<\/li>\n<li>f\u00a0\u662f\u6fc0\u6d3b\u51fd\u6570\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e2a\u7b80\u6d01\u800c\u4f18\u96c5\u7684\u6570\u5b66\u6a21\u578b&#xff0c;\u5c31\u662f\u4eba\u5de5\u795e\u7ecf\u5143\u7684\u6838\u5fc3\u3002\u5b83\u867d\u7136\u662f\u5bf9\u590d\u6742\u751f\u7269\u8fc7\u7a0b\u7684\u6781\u5927\u7b80\u5316&#xff0c;\u5374\u6293\u4f4f\u4e86\u201c\u52a0\u6743\u8f93\u5165\u3001\u6574\u5408\u3001\u975e\u7ebf\u6027\u6fc0\u6d3b\u201d\u8fd9\u4e00\u6838\u5fc3\u601d\u60f3\u3002\u5b83\u5c06\u4f5c\u4e3a\u6211\u4eec\u6784\u5efa\u6240\u6709\u590d\u6742\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u4e50\u9ad8\u79ef\u6728\u201d\u3002<\/p>\n<hr \/>\n<h4>4.2 \u611f\u77e5\u673a&#xff08;Perceptron&#xff09;&#xff1a;\u6700\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc<\/h4>\n<p>\u6709\u4e86\u4eba\u5de5\u795e\u7ecf\u5143\u8fd9\u4e2a\u57fa\u672c\u6784\u5efa\u5757&#xff0c;\u6211\u4eec\u5c31\u53ef\u4ee5\u642d\u5efa\u51fa\u53f2\u4e0a\u7b2c\u4e00\u4e2a\u3001\u4e5f\u662f\u6700\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u2014\u2014\u611f\u77e5\u673a\u3002\u5b83\u7531\u79d1\u5b66\u5bb6\u5f17\u5170\u514b\u00b7\u7f57\u68ee\u5e03\u62c9\u7279\u4e8e1957\u5e74\u63d0\u51fa&#xff0c;\u662f\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u7684\u5f00\u5c71\u4e4b\u4f5c\u3002<\/p>\n<h5>4.2.1 \u5355\u5c42\u611f\u77e5\u673a&#xff1a;\u7ebf\u6027\u5206\u7c7b\u7684\u5148\u884c\u8005<\/h5>\n<p>\u6a21\u578b\u7ed3\u6784 \u4e00\u4e2a\u5355\u5c42\u611f\u77e5\u673a&#xff0c;\u5176\u7ed3\u6784\u6781\u5176\u7b80\u5355&#xff1a;\u5b83\u4ec5\u5305\u542b\u4e00\u4e2a\u8f93\u5165\u5c42\u548c\u4e00\u4e2a\u8f93\u51fa\u5c42\u3002\u8f93\u51fa\u5c42\u901a\u5e38\u53ea\u6709\u4e00\u4e2a\u6211\u4eec\u521a\u521a\u5b9a\u4e49\u7684\u4eba\u5de5\u795e\u7ecf\u5143\u3002\u6570\u636e\u4ece\u8f93\u5165\u5c42\u8fdb\u5165&#xff0c;\u7ecf\u8fc7\u8fd9\u4e2a\u795e\u7ecf\u5143\u7684\u5904\u7406\u540e&#xff0c;\u76f4\u63a5\u5f97\u5230\u6700\u7ec8\u7684\u8f93\u51fa\u7ed3\u679c\u3002<\/p>\n<p>\u6fc0\u6d3b\u51fd\u6570&#xff1a;\u9636\u8dc3\u51fd\u6570 \u5728\u6700\u521d\u7684\u611f\u77e5\u673a\u6a21\u578b\u4e2d&#xff0c;\u4e3a\u4e86\u6700\u76f4\u63a5\u5730\u6a21\u62df\u751f\u7269\u795e\u7ecf\u5143\u201c\u5168\u6216\u65e0\u201d\u7684\u7279\u6027&#xff0c;\u5b83\u4f7f\u7528\u7684\u6fc0\u6d3b\u51fd\u6570\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u9636\u8dc3\u51fd\u6570&#xff08;Step Function&#xff09;&#xff0c;\u4e5f\u79f0\u4e3aHeaviside\u51fd\u6570&#xff1a; f(z) &#061; 1 \u5982\u679c z \u2265 0 f(z) &#061; 0 \u5982\u679c z &lt; 0 \u8fd9\u610f\u5473\u7740&#xff0c;\u5982\u679c\u52a0\u6743\u548c w\u1d40x &#043; b \u5927\u4e8e\u6216\u7b49\u4e8e0&#xff0c;\u795e\u7ecf\u5143\u5c31\u8f93\u51fa1&#xff08;\u88ab\u6fc0\u6d3b&#xff09;&#xff1b;\u5426\u5219&#xff0c;\u5c31\u8f93\u51fa0&#xff08;\u4e0d\u6fc0\u6d3b&#xff09;\u3002\u8fd9\u4f7f\u5f97\u611f\u77e5\u673a\u975e\u5e38\u9002\u5408\u6267\u884c\u4e8c\u5206\u7c7b\u4efb\u52a1\u3002<\/p>\n<p>\u5b66\u4e60\u89c4\u5219 \u611f\u77e5\u673a\u7684\u5b66\u4e60\u7b97\u6cd5\u540c\u6837\u975e\u5e38\u76f4\u89c2\u3002\u5bf9\u4e8e\u4e00\u4e2a\u8bad\u7ec3\u6837\u672c&#xff0c;\u5728\u505a\u51fa\u9884\u6d4b\u540e&#xff1a;<\/p>\n<p>\u901a\u8fc7\u5728\u8bad\u7ec3\u6570\u636e\u4e0a\u53cd\u590d\u8fed\u4ee3\u8fd9\u4e2a\u7b80\u5355\u7684\u66f4\u65b0\u89c4\u5219&#xff0c;\u611f\u77e5\u673a\u80fd\u591f\u81ea\u52a8\u5730\u5b66\u4e60\u5230\u4e00\u7ec4\u53ef\u4ee5\u6b63\u786e\u5212\u5206\u6570\u636e\u7684\u6743\u91cd\u548c\u504f\u7f6e\u3002<\/p>\n<ul>\n<li>\u5982\u679c\u9884\u6d4b\u6b63\u786e&#xff0c;\u6743\u91cd\u00a0w\u00a0\u548c\u504f\u7f6e\u00a0b\u00a0\u4fdd\u6301\u4e0d\u53d8\u3002<\/li>\n<li>\u5982\u679c\u9884\u6d4b\u9519\u8bef&#xff1a;\n<ul>\n<li>\u771f\u5b9e\u4e3a1&#xff0c;\u9884\u6d4b\u4e3a0&#xff1a;\u8bf4\u660e\u52a0\u6743\u548c\u592a\u5c0f\u4e86&#xff0c;\u9700\u8981\u589e\u5927\u3002\u56e0\u6b64&#xff0c;\u5c06\u8f93\u5165\u5411\u91cf\u00a0x\u00a0\u6309\u4e00\u5b9a\u6bd4\u4f8b&#xff08;\u5b66\u4e60\u7387\u00a0\u03b7&#xff09;\u52a0\u5230\u6743\u91cd\u5411\u91cf\u00a0w\u00a0\u4e0a&#xff08;w &#061; w &#043; \u03b7x&#xff09;&#xff0c;\u540c\u65f6\u589e\u5927\u504f\u7f6e\u00a0b\u3002<\/li>\n<li>\u771f\u5b9e\u4e3a0&#xff0c;\u9884\u6d4b\u4e3a1&#xff1a;\u8bf4\u660e\u52a0\u6743\u548c\u592a\u5927\u4e86&#xff0c;\u9700\u8981\u51cf\u5c0f\u3002\u56e0\u6b64&#xff0c;\u4ece\u6743\u91cd\u5411\u91cf\u00a0w\u00a0\u4e2d\u51cf\u53bb\u8f93\u5165\u5411\u91cf\u00a0x&#xff08;w &#061; w &#8211; \u03b7x&#xff09;&#xff0c;\u540c\u65f6\u51cf\u5c0f\u504f\u7f6e\u00a0b\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>4.2.2 \u611f\u77e5\u673a\u7684\u5c40\u9650\u6027&#xff1a;XOR\u95ee\u9898\u7684\u56f0\u5883<\/h5>\n<p>\u611f\u77e5\u673a\u5728\u5f53\u65f6\u5f15\u8d77\u4e86\u5de8\u5927\u7684\u8f70\u52a8&#xff0c;\u4eba\u4eec\u4e00\u5ea6\u8ba4\u4e3a\u901a\u5f80\u673a\u5668\u667a\u80fd\u7684\u9053\u8def\u5df2\u7ecf\u655e\u5f00\u3002\u7136\u800c&#xff0c;\u5b83\u7684\u4e00\u4e2a\u81f4\u547d\u7f3a\u9677\u5f88\u5feb\u88ab\u63ed\u793a\u51fa\u6765&#xff0c;\u8fd9\u4e2a\u53d1\u73b0\u76f4\u63a5\u5bfc\u81f4\u4e86\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u7684\u7b2c\u4e00\u6b21\u201c\u5bd2\u51ac\u201d\u3002<\/p>\n<p>\u7ebf\u6027\u53ef\u5206\u6027 \u8ba9\u6211\u4eec\u56de\u987e\u4e00\u4e0b\u611f\u77e5\u673a\u7684\u51b3\u7b56\u8fc7\u7a0b\u3002\u5b83\u5c06\u6570\u636e\u5206\u4e3a\u4e24\u7c7b\u7684\u8fb9\u754c\u662f w\u1d40x &#043; b &#061; 0\u3002\u5728\u4e8c\u7ef4\u7a7a\u95f4\u4e2d&#xff0c;\u8fd9\u662f\u4e00\u4e2a\u65b9\u7a0b w\u2081x\u2081 &#043; w\u2082x\u2082 &#043; b &#061; 0&#xff0c;\u8fd9\u6b63\u662f\u4e00\u6761\u76f4\u7ebf\u7684\u65b9\u7a0b\u3002\u5728\u4e09\u7ef4\u7a7a\u95f4\u4e2d&#xff0c;\u5b83\u662f\u4e00\u4e2a\u5e73\u9762&#xff1b;\u5728\u66f4\u9ad8\u7ef4\u7684\u7a7a\u95f4\u4e2d&#xff0c;\u5b83\u662f\u4e00\u4e2a\u8d85\u5e73\u9762\u3002<\/p>\n<p>\u8fd9\u610f\u5473\u7740&#xff0c;\u5355\u5c42\u611f\u77e5\u673a\u672c\u8d28\u4e0a\u662f\u4e00\u4e2a\u7ebf\u6027\u5206\u7c7b\u5668\u3002\u5b83\u53ea\u80fd\u89e3\u51b3\u90a3\u4e9b\u53ef\u4ee5\u7528\u4e00\u6761\u76f4\u7ebf&#xff08;\u6216\u4e00\u4e2a\u8d85\u5e73\u9762&#xff09;\u5c31\u80fd\u5c06\u4e24\u7c7b\u6837\u672c\u5b8c\u7f8e\u5206\u5f00\u7684\u95ee\u9898\u3002\u8fd9\u7c7b\u95ee\u9898&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u7ebf\u6027\u53ef\u5206&#xff08;Linearly Separable&#xff09;\u95ee\u9898\u3002<\/p>\n<p>XOR\u56f0\u5883 \u95ee\u9898\u5728\u4e8e&#xff0c;\u73b0\u5b9e\u4e16\u754c\u4e2d\u8bb8\u591a\u95ee\u9898\u90fd\u4e0d\u662f\u7ebf\u6027\u53ef\u5206\u7684\u3002\u5176\u4e2d\u6700\u8457\u540d\u3001\u6700\u7b80\u5355\u7684\u4f8b\u5b50&#xff0c;\u5c31\u662f\u5f02\u6216&#xff08;XOR&#xff09;\u95ee\u9898\u3002 XOR\u903b\u8f91\u5173\u7cfb\u5982\u4e0b&#xff1a;<\/p>\n<p>\u5f53\u6211\u4eec\u5c06\u8fd9\u56db\u4e2a\u70b9\u7ed8\u5236\u5728\u4e8c\u7ef4\u5e73\u9762\u4e0a\u65f6&#xff0c;\u4f1a\u53d1\u73b0&#xff1a;\u6211\u4eec\u6c38\u8fdc\u65e0\u6cd5\u7528\u4e00\u6761\u76f4\u7ebf&#xff0c;\u5c06\u4ee3\u8868\u8f93\u51fa0\u7684\u4e24\u4e2a\u70b9 (0,0), (1,1) \u548c\u4ee3\u8868\u8f93\u51fa1\u7684\u4e24\u4e2a\u70b9 (0,1), (1,0) \u5206\u5f00\u3002<\/p>\n<p>\u8fd9\u4e2a\u770b\u4f3c\u7b80\u5355\u7684XOR\u95ee\u9898&#xff0c;\u6210\u4e3a\u4e86\u5355\u5c42\u611f\u77e5\u673a\u65e0\u6cd5\u903e\u8d8a\u7684\u9e3f\u6c9f\u30021969\u5e74&#xff0c;\u9a6c\u6587\u00b7\u660e\u65af\u57fa\u548c\u897f\u6469\u5c14\u00b7\u6d3e\u666e\u7279\u5728\u4ed6\u4eec\u7684\u8457\u4f5c\u300a\u611f\u77e5\u673a\u300b\u4e2d&#xff0c;\u7cfb\u7edf\u5730\u8bba\u8bc1\u4e86\u8fd9\u4e00\u5c40\u9650\u6027&#xff0c;\u5e76\u60b2\u89c2\u5730\u9884\u6d4b\u4e86\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u7684\u524d\u666f\u3002\u8fd9\u7ed9\u5f53\u65f6\u70ed\u60c5\u9ad8\u6da8\u7684AI\u793e\u533a\u6cfc\u4e86\u4e00\u76c6\u51b7\u6c34&#xff0c;\u5bfc\u81f4\u8be5\u9886\u57df\u7684\u7814\u7a76\u7ecf\u8d39\u88ab\u5927\u91cf\u524a\u51cf&#xff0c;\u8fdb\u5165\u4e86\u957f\u8fbe\u8fd1\u4e8c\u5341\u5e74\u7684\u505c\u6ede\u671f\u3002<\/p>\n<ul>\n<li>\u8f93\u5165\u00a0(0, 0)\u00a0-&gt; \u8f93\u51fa\u00a00<\/li>\n<li>\u8f93\u5165\u00a0(0, 1)\u00a0-&gt; \u8f93\u51fa\u00a01<\/li>\n<li>\u8f93\u5165\u00a0(1, 0)\u00a0-&gt; \u8f93\u51fa\u00a01<\/li>\n<li>\u8f93\u5165\u00a0(1, 1)\u00a0-&gt; \u8f93\u51fa\u00a00<\/li>\n<\/ul>\n<p>\u7136\u800c&#xff0c;\u6b63\u5982\u5386\u53f2\u4e0a\u7684\u8bb8\u591a\u5371\u673a\u4e00\u6837&#xff0c;XOR\u56f0\u5883\u4e5f\u5b55\u80b2\u7740\u7a81\u7834\u7684\u79cd\u5b50\u3002\u5982\u4f55\u6253\u7834\u7ebf\u6027\u5206\u7c7b\u7684\u67b7\u9501&#xff1f;\u7b54\u6848&#xff0c;\u5c31\u5728\u4e8e\u589e\u52a0\u7f51\u7edc\u7684\u201c\u6df1\u5ea6\u201d\u3002<\/p>\n<h4>4.3 \u591a\u5c42\u611f\u77e5\u673a&#xff08;MLP&#xff09;&#xff1a;\u6784\u5efa\u6df1\u5ea6\u7f51\u7edc\u7684\u7b2c\u4e00\u6b65<\/h4>\n<p>\u8ba9\u6211\u4eec\u4ece\u90a3\u7247\u66fe\u7ecf\u8ba9\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u9677\u5165\u6c89\u5bc2\u7684\u201cXOR\u56f0\u5883\u201d\u51fa\u53d1&#xff0c;\u89c1\u8bc1\u5b83\u662f\u5982\u4f55\u88ab\u514b\u670d&#xff0c;\u5e76\u7531\u6b64\u5f00\u542f\u901a\u5f80\u201c\u6df1\u5ea6\u201d\u5b66\u4e60\u7684\u5eb7\u5e84\u5927\u9053\u7684\u3002<\/p>\n<p>\u5355\u5c42\u611f\u77e5\u673a\u7684\u5931\u8d25&#xff0c;\u5e76\u975e\u5ba3\u544a\u4e86\u795e\u7ecf\u7f51\u7edc\u601d\u60f3\u7684\u7ec8\u7ed3&#xff0c;\u800c\u662f\u6fc0\u53d1\u4e86\u7814\u7a76\u8005\u4eec\u66f4\u6df1\u5c42\u6b21\u7684\u601d\u8003\u3002\u65e2\u7136\u5355\u5c42\u7f51\u7edc\u7684\u80fd\u529b\u6709\u9650&#xff0c;\u6211\u4eec\u80fd\u5426\u901a\u8fc7\u5806\u53e0\u66f4\u591a\u7684\u5c42\u6b21\u6765\u6784\u5efa\u4e00\u4e2a\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u5462&#xff1f;\u7b54\u6848\u662f\u80af\u5b9a\u7684&#xff0c;\u8fd9\u4fbf\u5f15\u51fa\u4e86\u591a\u5c42\u611f\u77e5\u673a&#xff08;Multi-Layer Perceptron, MLP&#xff09;\u3002<\/p>\n<h5>4.3.1 \u7a81\u7834\u5c40\u9650&#xff1a;\u5f15\u5165\u9690\u85cf\u5c42<\/h5>\n<p>MLP\u7684\u7ed3\u6784 \u591a\u5c42\u611f\u77e5\u673a\u7684\u7ed3\u6784&#xff0c;\u76f8\u8f83\u4e8e\u5355\u5c42\u611f\u77e5\u673a&#xff0c;\u5176\u5173\u952e\u7684\u3001\u9769\u547d\u6027\u7684\u53d8\u5316\u5728\u4e8e&#xff1a;\u5728\u8f93\u5165\u5c42&#xff08;Input Layer&#xff09;\u548c\u8f93\u51fa\u5c42&#xff08;Output Layer&#xff09;\u4e4b\u95f4&#xff0c;\u52a0\u5165\u4e86\u4e00\u4e2a\u6216\u591a\u4e2a\u9690\u85cf\u5c42&#xff08;Hidden Layers&#xff09;\u3002<\/p>\n<ul>\n<li>\u8f93\u5165\u5c42&#xff1a;\u8d1f\u8d23\u63a5\u6536\u539f\u59cb\u6570\u636e&#xff0c;\u5176\u8282\u70b9\u6570\u7b49\u4e8e\u7279\u5f81\u7684\u6570\u91cf\u3002\u5b83\u4e0d\u8fdb\u884c\u8ba1\u7b97&#xff0c;\u4ec5\u4f5c\u4e3a\u6570\u636e\u5165\u53e3\u3002<\/li>\n<li>\u9690\u85cf\u5c42&#xff1a;\u4f4d\u4e8e\u8f93\u5165\u5c42\u548c\u8f93\u51fa\u5c42\u4e4b\u95f4&#xff0c;\u662fMLP\u7684\u6838\u5fc3\u3002\u6bcf\u4e2a\u9690\u85cf\u5c42\u90fd\u7531\u82e5\u5e72\u4e2a\u4eba\u5de5\u795e\u7ecf\u5143\u7ec4\u6210\u3002\u5b83\u4eec\u4e0d\u76f4\u63a5\u4e0e\u5916\u754c\u4ea4\u4e92&#xff08;\u65e2\u4e0d\u63a5\u6536\u539f\u59cb\u8f93\u5165&#xff0c;\u4e5f\u4e0d\u4ea7\u751f\u6700\u7ec8\u8f93\u51fa&#xff09;&#xff0c;\u5176\u5b58\u5728\u5bf9\u7528\u6237\u6765\u8bf4\u662f\u201c\u9690\u85cf\u201d\u7684&#xff0c;\u6545\u5f97\u6b64\u540d\u3002<\/li>\n<li>\u8f93\u51fa\u5c42&#xff1a;\u8d1f\u8d23\u4ea7\u751f\u6a21\u578b\u7684\u6700\u7ec8\u9884\u6d4b\u7ed3\u679c\u3002\u5176\u8282\u70b9\u6570\u53d6\u51b3\u4e8e\u4efb\u52a1\u7c7b\u578b&#xff08;\u56de\u5f52\u4efb\u52a1\u901a\u5e38\u4e3a1\u4e2a&#xff0c;\u591a\u5206\u7c7b\u4efb\u52a1\u901a\u5e38\u4e3a\u7c7b\u522b\u6570&#xff09;\u3002<\/li>\n<\/ul>\n<p>\u201c\u6df1\u5ea6\u201d\u7684\u8bde\u751f \u4e00\u4e2a\u7f51\u7edc\u4e2d\u9690\u85cf\u5c42\u7684\u6570\u91cf&#xff0c;\u51b3\u5b9a\u4e86\u5b83\u7684\u201c\u6df1\u5ea6\u201d\u3002\u62e5\u6709\u4e00\u4e2a\u6216\u591a\u4e2a\u9690\u85cf\u5c42\u7684\u795e\u7ecf\u7f51\u7edc&#xff0c;\u624d\u5f00\u59cb\u5177\u5907\u4e86\u201c\u6df1\u5ea6\u201d\u7684\u542b\u4e49&#xff0c;\u662f**\u6df1\u5ea6\u5b66\u4e60&#xff08;Deep Learning&#xff09;**\u8fd9\u4e2a\u672f\u8bed\u7684\u7531\u6765\u3002\u867d\u7136\u4e00\u4e2a\u4ec5\u542b\u5355\u4e2a\u9690\u85cf\u5c42\u7684MLP\u901a\u5e38\u88ab\u79f0\u4e3a\u201c\u6d45\u5c42\u201d\u7f51\u7edc&#xff0c;\u4f46\u5b83\u5df2\u7ecf\u64ad\u4e0b\u4e86\u901a\u5f80\u201c\u6df1\u5ea6\u201d\u7684\u79cd\u5b50\u3002<\/p>\n<h5>4.3.2 \u9690\u85cf\u5c42\u7684\u4f5c\u7528&#xff1a;\u7279\u5f81\u7684\u7ec4\u5408\u4e0e\u53d8\u6362<\/h5>\n<p>\u9690\u85cf\u5c42\u7a76\u7adf\u65bd\u5c55\u4e86\u600e\u6837\u7684\u201c\u9b54\u6cd5\u201d&#xff0c;\u4ece\u800c\u514b\u670d\u4e86\u5355\u5c42\u611f\u77e5\u673a\u7684\u5c40\u9650\u5462&#xff1f;\u5176\u6838\u5fc3\u5728\u4e8e\u7279\u5f81\u7684\u975e\u7ebf\u6027\u7ec4\u5408\u4e0e\u5c42\u6b21\u5316\u5b66\u4e60\u3002<\/p>\n<p>\u89e3\u51b3XOR\u95ee\u9898 \u8ba9\u6211\u4eec\u56de\u5230XOR\u56f0\u5883\u3002\u4e00\u4e2a\u5355\u5c42\u611f\u77e5\u673a\u53ea\u80fd\u753b\u51fa\u4e00\u6761\u76f4\u7ebf&#xff0c;\u65e0\u6cd5\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u3002\u4f46\u4e00\u4e2a\u5e26\u6709\u4e00\u4e2a\u9690\u85cf\u5c42&#xff08;\u4f8b\u5982&#xff0c;\u5305\u542b\u4e24\u4e2a\u795e\u7ecf\u5143&#xff09;\u7684MLP\u5374\u53ef\u4ee5\u8f7b\u677e\u5e94\u5bf9\u3002<\/p>\n<p>\u4e00\u4e2a\u76f4\u89c2\u7684\u56fe\u89e3\u601d\u8def&#xff1a;<\/p>\n<li>\u9690\u85cf\u5c42\u7684\u7b2c\u4e00\u4e2a\u795e\u7ecf\u5143&#xff1a;\u5b83\u53ef\u4ee5\u5b66\u4e60\u5230\u4e00\u6761\u51b3\u7b56\u8fb9\u754c&#xff08;\u6bd4\u5982\u4e00\u6761\u659c\u7ebf&#xff09;&#xff0c;\u5c06\u70b9\u00a0(1,1)\u00a0\u4e0e\u5176\u4ed6\u4e09\u4e2a\u70b9\u5206\u5f00\u3002<\/li>\n<li>\u9690\u85cf\u5c42\u7684\u7b2c\u4e8c\u4e2a\u795e\u7ecf\u5143&#xff1a;\u5b83\u53ef\u4ee5\u5b66\u4e60\u5230\u53e6\u4e00\u6761\u51b3\u7b56\u8fb9\u754c&#xff08;\u6bd4\u5982\u53e6\u4e00\u6761\u659c\u7ebf&#xff09;&#xff0c;\u5c06\u70b9\u00a0(0,0)\u00a0\u4e0e\u5176\u4ed6\u4e09\u4e2a\u70b9\u5206\u5f00\u3002<\/li>\n<li>\u8f93\u51fa\u5c42\u7684\u795e\u7ecf\u5143&#xff1a;\u5b83\u4e0d\u518d\u76f4\u63a5\u5904\u7406\u539f\u59cb\u7684\u8f93\u5165\u00a0(x\u2081, x\u2082)&#xff0c;\u800c\u662f\u5904\u7406\u6765\u81ea\u4e24\u4e2a\u9690\u85cf\u5c42\u795e\u7ecf\u5143\u7684\u8f93\u51fa\u3002\u5b83\u53ef\u4ee5\u5b66\u4e60\u5230\u4e00\u4e2a\u7b80\u5355\u7684\u903b\u8f91&#xff1a;\u5f53\u4e14\u4ec5\u5f53\u7b2c\u4e00\u4e2a\u9690\u85cf\u795e\u7ecf\u5143\u548c\u7b2c\u4e8c\u4e2a\u9690\u85cf\u795e\u7ecf\u5143\u90fd\u4e0d\u88ab\u6fc0\u6d3b\u65f6&#xff0c;\u624d\u6700\u7ec8\u8f93\u51fa1\u3002<\/li>\n<p>\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;MLP\u4e0d\u518d\u662f\u753b\u4e00\u6761\u76f4\u7ebf&#xff0c;\u800c\u662f\u901a\u8fc7\u9690\u85cf\u5c42\u5b66\u4e60\u5230\u4e86\u4e24\u4e2a\u4e0d\u540c\u7684\u7ebf\u6027\u8fb9\u754c&#xff0c;\u518d\u7531\u8f93\u51fa\u5c42\u5bf9\u8fd9\u4e24\u4e2a\u8fb9\u754c\u7684\u7ed3\u679c\u8fdb\u884c\u7ec4\u5408&#xff0c;\u4ece\u800c\u6784\u9020\u51fa\u4e86\u4e00\u4e2a\u80fd\u591f\u5b8c\u7f8e\u5305\u56f4\u4f4f (0,1) \u548c (1,0) \u7684\u975e\u7ebf\u6027\u51b3\u7b56\u533a\u57df\u3002\u5355\u5c42\u611f\u77e5\u673a\u7684\u7ebf\u6027\u67b7\u9501&#xff0c;\u5c31\u6b64\u88ab\u6253\u7834\u3002<\/p>\n<p>\u7279\u5f81\u7684\u5c42\u6b21\u5316\u5b66\u4e60 \u89e3\u51b3XOR\u95ee\u9898\u53ea\u662f\u725b\u5200\u5c0f\u8bd5\u3002\u9690\u85cf\u5c42\u66f4\u6df1\u523b\u7684\u610f\u4e49\u5728\u4e8e&#xff0c;\u5b83\u5f00\u542f\u4e86\u7279\u5f81\u7684\u5c42\u6b21\u5316\u5b66\u4e60&#xff08;Hierarchical Feature Learning&#xff09;\u3002<\/p>\n<ul>\n<li>\u7b2c\u4e00\u5c42\u9690\u85cf\u5c42&#xff1a;\u76f4\u63a5\u4e0e\u539f\u59cb\u8f93\u5165\u6570\u636e\u76f8\u8fde\u3002\u5b83\u53ef\u4ee5\u5b66\u4e60\u5230\u4e00\u4e9b\u521d\u7ea7\u7684\u3001\u7b80\u5355\u7684\u7279\u5f81\u3002\u4f8b\u5982&#xff0c;\u5728\u56fe\u50cf\u8bc6\u522b\u4efb\u52a1\u4e2d&#xff0c;\u7b2c\u4e00\u5c42\u53ef\u80fd\u5b66\u4f1a\u8bc6\u522b\u8fb9\u7f18\u3001\u89d2\u70b9\u3001\u989c\u8272\u5757\u7b49\u57fa\u672c\u6a21\u5f0f\u3002<\/li>\n<li>\u66f4\u6df1\u7684\u9690\u85cf\u5c42&#xff1a;\u5b83\u4eec\u4e0d\u518d\u770b\u539f\u59cb\u6570\u636e&#xff0c;\u800c\u662f\u4ee5\u4e0a\u4e00\u5c42\u9690\u85cf\u5c42\u7684\u8f93\u51fa&#xff08;\u5373\u521d\u7ea7\u7279\u5f81&#xff09;\u4f5c\u4e3a\u8f93\u5165\u3002\u56e0\u6b64&#xff0c;\u5b83\u4eec\u53ef\u4ee5\u5728\u8fd9\u4e9b\u521d\u7ea7\u7279\u5f81\u7684\u57fa\u7840\u4e0a&#xff0c;\u5b66\u4e60\u5230\u66f4\u590d\u6742\u3001\u66f4\u62bd\u8c61\u7684\u7ec4\u5408\u7279\u5f81\u3002\u4f8b\u5982&#xff0c;\u7b2c\u4e8c\u5c42\u53ef\u80fd\u5b66\u4f1a\u5c06\u8fb9\u7f18\u548c\u89d2\u70b9\u7ec4\u5408\u6210\u773c\u775b\u3001\u9f3b\u5b50\u3001\u8033\u6735\u7b49\u90e8\u4ef6\u3002<\/li>\n<li>\u66f4\u6df1\u2026\u2026&#xff1a;\u7b2c\u4e09\u5c42\u53ef\u80fd\u5c06\u773c\u775b\u3001\u9f3b\u5b50\u7b49\u90e8\u4ef6\u7ec4\u5408\u6210\u4e00\u5f20\u4eba\u8138\u7684\u8f6e\u5ed3\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u6b63\u547c\u5e94\u4e86\u6211\u4eec\u5728\u7b2c\u4e8c\u7ae0\u4e2d\u63d0\u5230\u7684\u201c\u7a7a\u95f4\u53d8\u6362\u201d\u7684\u6982\u5ff5\u3002MLP\u7684\u6bcf\u4e00\u5c42&#xff0c;\u90fd\u5728\u5bf9\u524d\u4e00\u5c42\u6240\u5904\u7684\u7279\u5f81\u7a7a\u95f4\u8fdb\u884c\u4e00\u6b21\u590d\u6742\u7684\u975e\u7ebf\u6027\u53d8\u6362&#xff0c;\u76ee\u7684\u662f\u5c06\u6570\u636e\u9010\u6b65\u6620\u5c04\u5230\u4e00\u4e2a\u65b0\u7684\u7a7a\u95f4&#xff0c;\u5728\u8fd9\u4e2a\u7a7a\u95f4\u91cc&#xff0c;\u4e0d\u540c\u7c7b\u522b\u7684\u6570\u636e\u80fd\u591f\u88ab\u66f4\u5bb9\u6613\u5730\u533a\u5206\u5f00\u6765\u3002\u8fd9\u79cd\u81ea\u52a8\u5b66\u4e60\u6709\u6548\u7279\u5f81\u8868\u793a\u7684\u80fd\u529b&#xff0c;\u662f\u6df1\u5ea6\u5b66\u4e60\u76f8\u6bd4\u4f20\u7edf\u673a\u5668\u5b66\u4e60\u6700\u6838\u5fc3\u7684\u4f18\u52bf\u4e4b\u4e00\u3002\u00a0<\/p>\n<h4>4.4 \u6fc0\u6d3b\u51fd\u6570&#xff1a;\u8d4b\u4e88\u7f51\u7edc\u975e\u7ebf\u6027<\/h4>\n<p>\u5728MLP\u7684\u8ba8\u8bba\u4e2d&#xff0c;\u6211\u4eec\u63d0\u5230\u4e86\u201c\u975e\u7ebf\u6027\u201d\u8fd9\u4e2a\u5173\u952e\u8bcd\u3002\u8d4b\u4e88\u7f51\u7edc\u8fd9\u79cd\u81f3\u5173\u91cd\u8981\u7684\u975e\u7ebf\u6027\u80fd\u529b\u7684\u7ec4\u4ef6&#xff0c;\u6b63\u662f\u6fc0\u6d3b\u51fd\u6570\u3002<\/p>\n<h5>4.4.1 \u4e3a\u4f55\u9700\u8981\u975e\u7ebf\u6027&#xff1f;<\/h5>\n<p>\u7ebf\u6027\u53e0\u52a0\u7684\u74f6\u9888 \u5047\u8bbe\u6211\u4eec\u6784\u5efa\u4e00\u4e2a\u591a\u5c42\u7f51\u7edc&#xff0c;\u4f46\u4e0d\u4f7f\u7528\u4efb\u4f55\u6fc0\u6d3b\u51fd\u6570&#xff08;\u6216\u8005\u8bf4&#xff0c;\u4f7f\u7528\u4e00\u4e2a\u7ebf\u6027\u7684\u6fc0\u6d3b\u51fd\u6570 f(x)&#061;x&#xff09;\u3002\u90a3\u4e48&#xff0c;\u7b2c\u4e00\u5c42\u7684\u8f93\u51fa\u662f W\u2081x &#043; b\u2081\u3002\u7b2c\u4e8c\u5c42\u7684\u8f93\u51fa\u5c06\u662f W\u2082(W\u2081x &#043; b\u2081) &#043; b\u2082 &#061; (W\u2082W\u2081)x &#043; (W\u2082b\u2081 &#043; b\u2082)\u3002 \u65e0\u8bba\u6211\u4eec\u5806\u53e0\u591a\u5c11\u5c42&#xff0c;\u6700\u7ec8\u7684\u7ed3\u679c (W_n&#8230;W\u2082W\u2081)x &#043; &#8230; \u59cb\u7ec8\u53ef\u4ee5\u88ab\u5316\u7b80\u4e3a\u4e00\u4e2a\u7b49\u6548\u7684\u5355\u5c42\u7ebf\u6027\u53d8\u6362 W&#039;x &#043; b&#039;\u3002\u8fd9\u610f\u5473\u7740&#xff0c;\u5982\u679c\u6ca1\u6709\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570&#xff0c;\u4e00\u4e2a\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u8868\u8fbe\u80fd\u529b\u5c06\u88ab\u9000\u5316\u4e3a\u4e00\u4e2a\u5355\u5c42\u7684\u7ebf\u6027\u7f51\u7edc&#xff0c;\u4ece\u800c\u5b8c\u5168\u5931\u53bb\u4e86\u201c\u6df1\u5ea6\u201d\u7684\u610f\u4e49&#xff0c;\u4e5f\u65e0\u6cd5\u89e3\u51b3\u50cfXOR\u8fd9\u6837\u7684\u975e\u7ebf\u6027\u95ee\u9898\u3002<\/p>\n<p>\u8868\u8fbe\u80fd\u529b\u7684\u6e90\u6cc9 \u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u662f\u7f51\u7edc\u80fd\u591f\u62df\u5408\u4efb\u610f\u590d\u6742\u51fd\u6570\u7684\u5173\u952e\u3002\u5b83\u5728\u6bcf\u4e00\u5c42\u7ebf\u6027\u53d8\u6362&#xff08;w\u1d40x &#043; b&#xff09;\u4e4b\u540e&#xff0c;\u5bf9\u7279\u5f81\u7a7a\u95f4\u8fdb\u884c\u4e00\u6b21\u201c\u5f2f\u66f2\u201d\u6216\u201c\u6298\u53e0\u201d\u3002\u6b63\u662f\u8fd9\u4e9b\u8fde\u7eed\u7684\u3001\u975e\u7ebf\u6027\u7684\u626d\u66f2\u64cd\u4f5c&#xff0c;\u4f7f\u5f97\u795e\u7ecf\u7f51\u7edc\u6709\u80fd\u529b\u53bb\u903c\u8fd1&#xff08;fit&#xff09;\u6570\u636e\u4e2d\u4efb\u4f55\u590d\u6742\u7684\u3001\u975e\u7ebf\u6027\u7684\u6f5c\u5728\u89c4\u5f8b\u3002\u6839\u636e\u901a\u7528\u8fd1\u4f3c\u5b9a\u7406&#xff08;Universal Approximation Theorem&#xff09;&#xff0c;\u4e00\u4e2a\u5e26\u6709\u4e00\u4e2a\u9690\u85cf\u5c42\u548c\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u7684MLP&#xff0c;\u53ea\u8981\u9690\u85cf\u5c42\u795e\u7ecf\u5143\u6570\u91cf\u8db3\u591f\u591a&#xff0c;\u7406\u8bba\u4e0a\u5c31\u53ef\u4ee5\u4ee5\u4efb\u610f\u7cbe\u5ea6\u8fd1\u4f3c\u4efb\u4f55\u8fde\u7eed\u51fd\u6570\u3002<\/p>\n<h5>4.4.2 \u5e38\u7528\u6fc0\u6d3b\u51fd\u6570\u5de1\u793c<\/h5>\n<p>\u9009\u62e9\u5408\u9002\u7684\u6fc0\u6d3b\u51fd\u6570&#xff0c;\u5bf9\u7f51\u7edc\u7684\u8bad\u7ec3\u901f\u5ea6\u548c\u6027\u80fd\u81f3\u5173\u91cd\u8981\u3002\u4ee5\u4e0b\u662f\u51e0\u4f4d\u5728\u795e\u7ecf\u7f51\u7edc\u53d1\u5c55\u53f2\u4e2d\u626e\u6f14\u4e86\u91cd\u8981\u89d2\u8272\u7684\u6fc0\u6d3b\u51fd\u6570\u3002<\/p>\n<p>Sigmoid\u51fd\u6570&#xff1a;\u7ecf\u5178\u7684S\u5f62\u66f2\u7ebf<\/p>\n<ul>\n<li>\u51fd\u6570&#xff1a;\u03c3(x) &#061; 1 \/ (1 &#043; e\u207b\u02e3)<\/li>\n<li>\u7279\u6027&#xff1a;\u5c06\u4efb\u610f\u5b9e\u6570\u8f93\u5165\u5e73\u6ed1\u5730\u538b\u7f29\u5230\u00a0(0, 1)\u00a0\u533a\u95f4\u3002\u8fd9\u4f7f\u5f97\u5b83\u7684\u8f93\u51fa\u53ef\u4ee5\u88ab\u89e3\u91ca\u4e3a\u6982\u7387&#xff0c;\u5728\u65e9\u671f\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u4e8c\u5206\u7c7b\u95ee\u9898\u7684\u8f93\u51fa\u5c42\u3002<\/li>\n<li>\u7f3a\u70b9&#xff1a;\n<li>\u68af\u5ea6\u6d88\u5931&#xff08;Vanishing Gradients&#xff09;&#xff1a;\u5f53\u8f93\u5165\u503c\u975e\u5e38\u5927\u6216\u975e\u5e38\u5c0f\u65f6&#xff0c;Sigmoid\u51fd\u6570\u7684\u66f2\u7ebf\u53d8\u5f97\u975e\u5e38\u5e73\u5766&#xff0c;\u5176\u5bfc\u6570&#xff08;\u68af\u5ea6&#xff09;\u63a5\u8fd1\u4e8e0\u3002\u5728\u6df1\u5c42\u7f51\u7edc\u4e2d&#xff0c;\u8fd9\u4e9b\u63a5\u8fd10\u7684\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u65f6\u4f1a\u5c42\u5c42\u76f8\u4e58&#xff0c;\u5bfc\u81f4\u4f20\u5230\u6d45\u5c42\u7f51\u7edc\u7684\u68af\u5ea6\u4fe1\u53f7\u53d8\u5f97\u6781\u5176\u5fae\u5f31&#xff0c;\u4f7f\u5f97\u6d45\u5c42\u7f51\u7edc\u7684\u53c2\u6570\u51e0\u4e4e\u65e0\u6cd5\u66f4\u65b0\u3002<\/li>\n<li>\u8f93\u51fa\u975e\u96f6\u4e2d\u5fc3&#xff1a;\u5176\u8f93\u51fa\u6052\u4e3a\u6b63&#xff0c;\u8fd9\u4f1a\u5bfc\u81f4\u540e\u7eed\u5c42\u7684\u8f93\u5165\u90fd\u662f\u975e\u96f6\u4e2d\u5fc3\u7684&#xff0c;\u53ef\u80fd\u964d\u4f4e\u68af\u5ea6\u4e0b\u964d\u7684\u6536\u655b\u6548\u7387\u3002<\/li>\n<\/li>\n<\/ul>\n<p>Tanh\u51fd\u6570&#xff08;\u53cc\u66f2\u6b63\u5207&#xff09;&#xff1a;\u96f6\u4e2d\u5fc3\u5316\u7684S\u5f62\u66f2\u7ebf<\/p>\n<ul>\n<li>\u51fd\u6570&#xff1a;tanh(x) &#061; (e\u02e3 &#8211; e\u207b\u02e3) \/ (e\u02e3 &#043; e\u207b\u02e3)<\/li>\n<li>\u7279\u6027&#xff1a;\u5c06\u8f93\u5165\u538b\u7f29\u5230\u00a0(-1, 1)\u00a0\u533a\u95f4\u3002\u5176\u8f93\u51fa\u662f\u96f6\u4e2d\u5fc3\u7684&#xff0c;\u8fd9\u5728\u5b9e\u8df5\u4e2d\u901a\u5e38\u6bd4Sigmoid\u51fd\u6570\u6709\u66f4\u597d\u7684\u6027\u80fd&#xff0c;\u6536\u655b\u901f\u5ea6\u66f4\u5feb\u3002<\/li>\n<li>\u7f3a\u70b9&#xff1a;\u5b83\u540c\u6837\u662f\u4e00\u6761S\u5f62\u66f2\u7ebf&#xff0c;\u56e0\u6b64\u4e5f\u65e0\u6cd5\u907f\u514d\u5728\u9971\u548c\u533a\u7684\u68af\u5ea6\u6d88\u5931\u95ee\u9898\u3002<\/li>\n<\/ul>\n<p>ReLU (Rectified Linear Unit)&#xff1a;\u73b0\u4ee3\u7f51\u7edc\u7684\u9ed8\u8ba4\u9009\u62e9<\/p>\n<ul>\n<li>\u51fd\u6570&#xff1a;ReLU(x) &#061; max(0, x)<\/li>\n<li>\u7279\u6027&#xff1a;\u8fd9\u662f\u4e00\u4e2a\u6781\u5176\u7b80\u5355\u7684\u5206\u6bb5\u7ebf\u6027\u51fd\u6570\u3002\u5f53\u8f93\u5165\u4e3a\u6b63\u65f6&#xff0c;\u8f93\u51fa\u7b49\u4e8e\u8f93\u5165&#xff1b;\u5f53\u8f93\u5165\u4e3a\u8d1f\u65f6&#xff0c;\u8f93\u51fa\u4e3a0\u3002<\/li>\n<li>\u4f18\u70b9&#xff1a;\n<li>\u7f13\u89e3\u68af\u5ea6\u6d88\u5931&#xff1a;\u5728\u6b63\u6570\u533a\u95f4&#xff0c;\u5176\u5bfc\u6570\u6052\u4e3a1&#xff0c;\u8fd9\u6781\u5927\u5730\u7f13\u89e3\u4e86\u68af\u5ea6\u6d88\u5931\u95ee\u9898&#xff0c;\u4f7f\u5f97\u6df1\u5c42\u7f51\u7edc\u53ef\u4ee5\u5f97\u5230\u6709\u6548\u7684\u8bad\u7ec3\u3002<\/li>\n<li>\u8ba1\u7b97\u9ad8\u6548&#xff1a;\u76f8\u6bd4Sigmoid\u548cTanh\u6d89\u53ca\u7684\u6307\u6570\u8fd0\u7b97&#xff0c;ReLU\u7684\u8ba1\u7b97\u6210\u672c\u6781\u4f4e\u3002<\/li>\n<li>\u7a00\u758f\u6027&#xff1a;\u5b83\u4f1a\u4f7f\u4e00\u90e8\u5206\u795e\u7ecf\u5143\u7684\u8f93\u51fa\u4e3a0&#xff0c;\u8fd9\u9020\u6210\u4e86\u7f51\u7edc\u7684\u7a00\u758f\u6027&#xff0c;\u53ef\u80fd\u6709\u52a9\u4e8e\u63d0\u53d6\u66f4\u6709\u610f\u4e49\u7684\u7279\u5f81\u3002<\/li>\n<\/li>\n<li>\u7f3a\u70b9&#xff1a;\n<li>Dying ReLU&#xff08;\u6b7b\u4ea1ReLU&#xff09;\u95ee\u9898&#xff1a;\u5982\u679c\u4e00\u4e2a\u795e\u7ecf\u5143\u7684\u8f93\u5165\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u59cb\u7ec8\u4e3a\u8d1f&#xff0c;\u90a3\u4e48\u5b83\u7684\u8f93\u51fa\u5c06\u6c38\u8fdc\u662f0&#xff0c;\u68af\u5ea6\u4e5f\u6c38\u8fdc\u662f0\u3002\u8fd9\u4e2a\u795e\u7ecf\u5143\u5c06\u4e0d\u518d\u5bf9\u4efb\u4f55\u6570\u636e\u6709\u54cd\u5e94&#xff0c;\u4e5f\u65e0\u6cd5\u518d\u8fdb\u884c\u5b66\u4e60&#xff0c;\u5982\u540c\u201c\u6b7b\u4ea1\u201d\u4e86\u4e00\u822c\u3002<\/li>\n<\/li>\n<\/ul>\n<p>ReLU\u7684\u53d8\u4f53&#xff1a;\u5e94\u5bf9\u201c\u6b7b\u4ea1\u201d\u95ee\u9898 \u4e3a\u4e86\u89e3\u51b3Dying ReLU\u95ee\u9898&#xff0c;\u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e9b\u6539\u8fdb\u7248\u672c&#xff1a;<\/p>\n<ul>\n<li>Leaky ReLU&#xff1a;f(x) &#061; x\u00a0\u5982\u679c\u00a0x &gt; 0&#xff0c;f(x) &#061; \u03b1x\u00a0\u5982\u679c\u00a0x \u2264 0\u3002\u5b83\u4e3a\u8d1f\u533a\u95f4\u5f15\u5165\u4e00\u4e2a\u5fae\u5c0f\u7684\u3001\u56fa\u5b9a\u7684\u6b63\u659c\u7387\u00a0\u03b1&#xff08;\u59820.01&#xff09;&#xff0c;\u786e\u4fdd\u795e\u7ecf\u5143\u5728\u8d1f\u8f93\u5165\u65f6\u4e5f\u80fd\u6709\u975e\u96f6\u68af\u5ea6\u3002<\/li>\n<li>ELU (Exponential Linear Unit)&#xff1a;\u5728\u8d1f\u533a\u95f4&#xff0c;\u5b83\u662f\u4e00\u6761\u5e73\u6ed1\u7684\u6307\u6570\u66f2\u7ebf&#xff0c;\u800c\u4e0d\u662f\u76f4\u7ebf\u3002\u7406\u8bba\u4e0a\u5b83\u7ed3\u5408\u4e86ReLU\u548cLeaky ReLU\u7684\u4f18\u70b9&#xff0c;\u5bf9\u566a\u58f0\u66f4\u5177\u9c81\u68d2\u6027&#xff0c;\u4f46\u8ba1\u7b97\u6210\u672c\u7a0d\u9ad8\u3002<\/li>\n<\/ul>\n<h4>4.5 \u635f\u5931\u51fd\u6570&#xff1a;\u8861\u91cf\u201c\u7406\u60f3\u201d\u4e0e\u201c\u73b0\u5b9e\u201d\u7684\u5dee\u8ddd<\/h4>\n<p>\u6211\u4eec\u5df2\u7ecf\u6784\u5efa\u4e86\u7f51\u7edc\u7684\u7ed3\u6784&#xff0c;\u5e76\u4e3a\u5176\u6ce8\u5165\u4e86\u975e\u7ebf\u6027\u7684\u6d3b\u529b\u3002\u4f46\u6211\u4eec\u5982\u4f55\u544a\u8bc9\u7f51\u7edc\u5b83\u7684\u8868\u73b0\u662f\u597d\u662f\u574f\u5462&#xff1f;\u8fd9\u5c31\u9700\u8981\u635f\u5931\u51fd\u6570&#xff08;Loss Function&#xff09;\u3002<\/p>\n<h5>4.5.1 \u635f\u5931\u51fd\u6570&#xff08;Loss Function&#xff09;\u7684\u89d2\u8272<\/h5>\n<p>\u4f18\u5316\u7684\u201c\u5bfc\u822a\u5458\u201d \u635f\u5931\u51fd\u6570&#xff08;\u6709\u65f6\u4e5f\u79f0\u6210\u672c\u51fd\u6570\u6216\u76ee\u6807\u51fd\u6570&#xff09;\u662f\u4e00\u4e2a\u7528\u4e8e\u91cf\u5316\u6a21\u578b\u9884\u6d4b\u503c \u0176 \u4e0e\u771f\u5b9e\u6807\u7b7e Y \u4e4b\u95f4\u5dee\u8ddd\u7684\u51fd\u6570\u3002\u8fd9\u4e2a\u5dee\u8ddd\u503c&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u635f\u5931&#xff08;Loss&#xff09;\u3002 \u635f\u5931\u503c\u5c31\u662f\u4e00\u4e2a\u6807\u91cf&#xff0c;\u5b83\u4ee3\u8868\u4e86\u6a21\u578b\u5728\u5f53\u524d\u8bad\u7ec3\u6837\u672c\u4e0a\u201c\u72af\u9519\u201d\u7684\u7a0b\u5ea6\u3002\u635f\u5931\u8d8a\u5927&#xff0c;\u8bf4\u660e\u6a21\u578b\u9519\u5f97\u8d8a\u79bb\u8c31\u3002<\/p>\n<p>\u76ee\u6807 \u6574\u4e2a\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5176\u552f\u4e00\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u901a\u8fc7\u4e0d\u65ad\u8c03\u6574\u7f51\u7edc\u5185\u90e8\u7684\u6743\u91cd\u548c\u504f\u7f6e\u53c2\u6570&#xff0c;\u6765\u4f7f\u5f97\u5728\u6574\u4e2a\u8bad\u7ec3\u6570\u636e\u96c6\u4e0a\u7684\u603b\u635f\u5931\u51fd\u6570\u7684\u503c\u5c3d\u53ef\u80fd\u5c0f\u3002\u635f\u5931\u51fd\u6570\u4e3a\u6211\u4eec\u7684\u4f18\u5316\u8fc7\u7a0b&#xff08;\u5982\u68af\u5ea6\u4e0b\u964d&#xff09;\u63d0\u4f9b\u4e86\u660e\u786e\u7684\u3001\u53ef\u91cf\u5316\u7684\u76ee\u6807\u548c\u65b9\u5411\u3002<\/p>\n<h5>4.5.2 \u5e38\u89c1\u7684\u635f\u5931\u51fd\u6570<\/h5>\n<p>\u635f\u5931\u51fd\u6570\u7684\u9009\u62e9&#xff0c;\u53d6\u51b3\u4e8e\u6211\u4eec\u6240\u9762\u5bf9\u7684\u4efb\u52a1\u7c7b\u578b\u3002<\/p>\n<p>\u5747\u65b9\u8bef\u5dee&#xff08;MSE, Mean Squared Error&#xff09;<\/p>\n<p>\u9002\u7528\u573a\u666f&#xff1a;\u4e3b\u8981\u7528\u4e8e\u56de\u5f52\u4efb\u52a1&#xff0c;\u5373\u9884\u6d4b\u4e00\u4e2a\u8fde\u7eed\u503c&#xff08;\u5982\u623f\u4ef7\u3001\u6e29\u5ea6&#xff09;\u3002<\/p>\n<p>\u6570\u5b66\u5f62\u5f0f&#xff1a;L(Y, \u0176) &#061; (1\/n) * \u03a3(Y\u1d62 &#8211; \u0176\u1d62)\u00b2\u00a0\u5b83\u8ba1\u7b97\u7684\u662f\u9884\u6d4b\u503c\u4e0e\u771f\u5b9e\u503c\u4e4b\u5dee\u7684\u5e73\u65b9\u7684\u5e73\u5747\u503c\u3002<\/p>\n<p>\u4ea4\u53c9\u71b5\u635f\u5931&#xff08;Cross-Entropy Loss&#xff09;<\/p>\n<p>\u9002\u7528\u573a\u666f&#xff1a;\u662f\u5206\u7c7b\u4efb\u52a1\u7684\u6807\u51c6\u635f\u5931\u51fd\u6570\u3002<\/p>\n<p>\u4e8c\u5143\u4ea4\u53c9\u71b5&#xff08;Binary Cross-Entropy&#xff09;&#xff1a;\u7528\u4e8e\u4e8c\u5206\u7c7b\u95ee\u9898&#xff08;\u5982\u5224\u65ad\u90ae\u4ef6\u662f\u5426\u4e3a\u5783\u573e\u90ae\u4ef6&#xff09;\u3002\u5b83\u901a\u5e38\u4e0e\u8f93\u51fa\u5c42\u7684\u5355\u4e2aSigmoid\u6fc0\u6d3b\u51fd\u6570\u914d\u5408\u4f7f\u7528\u3002<\/p>\n<p>\u5206\u7c7b\u4ea4\u53c9\u71b5&#xff08;Categorical Cross-Entropy&#xff09;&#xff1a;\u7528\u4e8e\u591a\u5206\u7c7b\u95ee\u9898&#xff08;\u5982\u8bc6\u522b\u56fe\u7247\u662f\u732b\u3001\u72d7\u8fd8\u662f\u9e1f&#xff09;\u3002\u5b83\u901a\u5e38\u4e0e\u8f93\u51fa\u5c42\u7684Softmax\u6fc0\u6d3b\u51fd\u6570\u914d\u5408\u4f7f\u7528\u3002Softmax\u51fd\u6570\u80fd\u5c06\u4e00\u7ec4\u4efb\u610f\u5b9e\u6570\u8f6c\u6362\u4e3a\u4e00\u4e2a\u6982\u7387\u5206\u5e03&#xff08;\u6240\u6709\u8f93\u51fa\u503c\u5728(0,1)\u4e4b\u95f4\u4e14\u603b\u548c\u4e3a1&#xff09;\u3002<\/p>\n<p>\u4e3a\u4f55\u4f18\u4e8eMSE&#xff1a;\u5728\u5206\u7c7b\u4efb\u52a1\u4e2d&#xff0c;\u5982\u679c\u4f7f\u7528MSE&#xff0c;\u5f53\u6a21\u578b\u7684\u9884\u6d4b\u6982\u7387\u8fdc\u79bb\u771f\u5b9e\u6807\u7b7e&#xff08;\u5982\u771f\u5b9e\u4e3a1&#xff0c;\u9884\u6d4b\u4e3a0.001&#xff09;\u65f6&#xff0c;\u5176\u68af\u5ea6\u53ef\u80fd\u4f1a\u53d8\u5f97\u5f88\u5c0f&#xff0c;\u5bfc\u81f4\u5b66\u4e60\u975e\u5e38\u7f13\u6162\u3002\u800c\u4ea4\u53c9\u71b5\u635f\u5931\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\u80fd\u63d0\u4f9b\u4e00\u4e2a\u975e\u5e38\u5927\u7684\u68af\u5ea6\u4fe1\u53f7&#xff0c;\u4ece\u800c\u4fc3\u4f7f\u6a21\u578b\u66f4\u5feb\u5730\u4fee\u6b63\u9519\u8bef\u3002<\/p>\n<h4>4.6 \u53cd\u5411\u4f20\u64ad\u7b97\u6cd5&#xff1a;\u7f51\u7edc\u5b66\u4e60\u7684\u5f15\u64ce<\/h4>\n<p>\u6211\u4eec\u6709\u4e86\u7f51\u7edc\u7ed3\u6784\u3001\u6fc0\u6d3b\u51fd\u6570\u548c\u635f\u5931\u51fd\u6570\u3002\u73b0\u5728&#xff0c;\u6700\u5173\u952e\u7684\u95ee\u9898\u6765\u4e86&#xff1a;\u6211\u4eec\u7a76\u7adf\u8be5\u5982\u4f55\u7cfb\u7edf\u6027\u5730\u8c03\u6574\u90a3\u6210\u5343\u4e0a\u4e07\u4e2a\u53c2\u6570&#xff0c;\u6765\u6700\u5c0f\u5316\u635f\u5931\u5462&#xff1f;\u7b54\u6848\u5c31\u662f\u5927\u540d\u9f0e\u9f0e\u7684\u53cd\u5411\u4f20\u64ad&#xff08;Backpropagation&#xff09;\u7b97\u6cd5\u3002<\/p>\n<h5>4.6.1 \u7b97\u6cd5\u7684\u5b8f\u89c2\u7406\u89e3<\/h5>\n<p>\u76ee\u6807 \u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u7684\u552f\u4e00\u76ee\u6807&#xff0c;\u5c31\u662f\u9ad8\u6548\u5730\u8ba1\u7b97\u51fa\u635f\u5931\u51fd\u6570 L \u5173\u4e8e\u7f51\u7edc\u4e2d\u6240\u6709\u53c2\u6570&#xff08;\u6bcf\u4e00\u4e2a\u6743\u91cd w \u548c\u504f\u7f6e b&#xff09;\u7684\u68af\u5ea6&#xff08;\u504f\u5bfc\u6570&#xff09;\u3002<\/p>\n<p>\u4e24\u4e2a\u9636\u6bb5 \u4e00\u6b21\u5b8c\u6574\u7684\u8bad\u7ec3\u8fed\u4ee3&#xff0c;\u5305\u542b\u4e24\u4e2a\u9636\u6bb5&#xff1a;<\/p>\n<li>\u524d\u5411\u4f20\u64ad&#xff08;Forward Pass&#xff09;&#xff1a;\u5c06\u4e00\u4e2a\u6216\u4e00\u6279\u8bad\u7ec3\u6570\u636e\u8f93\u5165\u7f51\u7edc\u3002\u6570\u636e\u4ece\u8f93\u5165\u5c42\u5f00\u59cb&#xff0c;\u9010\u5c42\u6d41\u8fc7\u7f51\u7edc&#xff0c;\u7ecf\u8fc7\u6bcf\u4e00\u5c42\u7684\u7ebf\u6027\u53d8\u6362\u548c\u975e\u7ebf\u6027\u6fc0\u6d3b&#xff0c;\u76f4\u5230\u8f93\u51fa\u5c42\u4ea7\u751f\u6700\u7ec8\u7684\u9884\u6d4b\u503c\u00a0\u0176\u3002\u7136\u540e&#xff0c;\u7528\u635f\u5931\u51fd\u6570\u8ba1\u7b97\u51fa\u00a0\u0176\u00a0\u548c\u771f\u5b9e\u6807\u7b7e\u00a0Y\u00a0\u4e4b\u95f4\u7684\u635f\u5931\u503c\u00a0L\u3002<\/li>\n<li>\u53cd\u5411\u4f20\u64ad&#xff08;Backward Pass&#xff09;&#xff1a;\u8fd9\u662f\u7b97\u6cd5\u7684\u7cbe\u9ad3\u3002\u5b83\u4ece\u6700\u7ec8\u7684\u635f\u5931\u00a0L\u00a0\u51fa\u53d1&#xff0c;\u201c\u53cd\u5411\u201d\u5730\u3001\u9010\u5c42\u5730\u7a7f\u8d8a\u7f51\u7edc\u3002\u5728\u6bcf\u4e00\u5c42&#xff0c;\u5b83\u90fd\u5229\u7528\u6211\u4eec\u5728\u7b2c\u4e8c\u7ae0\u5b66\u8fc7\u7684\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u6765\u8ba1\u7b97\u635f\u5931\u00a0L\u00a0\u5bf9\u5f53\u524d\u5c42\u53c2\u6570\u7684\u68af\u5ea6&#xff0c;\u5e76\u5c06\u8bef\u5dee\u4fe1\u53f7\u7ee7\u7eed\u5411\u540e\u4f20\u9012\u3002<\/li>\n<h5>4.6.2 \u68af\u5ea6\u4e0b\u964d\u4e0e\u94fe\u5f0f\u6cd5\u5219\u7684\u5b8c\u7f8e\u7ed3\u5408<\/h5>\n<p>\u68af\u5ea6\u4e0b\u964d\u7684\u5e94\u7528 \u4e00\u65e6\u53cd\u5411\u4f20\u64ad\u5b8c\u6210\u4e86\u5b83\u7684\u4f7f\u547d&#xff0c;\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u6240\u6709\u53c2\u6570\u7684\u68af\u5ea6&#xff0c;\u68af\u5ea6\u4e0b\u964d&#xff08;\u6216\u5176\u66f4\u9ad8\u7ea7\u7684\u53d8\u4f53&#xff0c;\u5982Adam\u3001RMSprop&#xff0c;\u6211\u4eec\u5c06\u5728\u540e\u7eed\u7ae0\u8282\u8be6\u8ff0&#xff09;\u5c31\u4f1a\u63a5\u7ba1\u5de5\u4f5c\u3002\u5b83\u6839\u636e\u4ee5\u4e0b\u516c\u5f0f\u6765\u66f4\u65b0\u6bcf\u4e00\u4e2a\u53c2\u6570&#xff1a; \u65b0\u53c2\u6570 &#061; \u65e7\u53c2\u6570 &#8211; \u5b66\u4e60\u7387 * \u68af\u5ea6 \u8fd9\u4e2a\u7b80\u5355\u7684\u66f4\u65b0\u6b65\u9aa4&#xff0c;\u4f1a\u4f7f\u53c2\u6570\u5411\u7740\u80fd\u51cf\u5c0f\u635f\u5931\u7684\u65b9\u5411\u79fb\u52a8\u4e00\u5c0f\u6b65\u3002<\/p>\n<p>\u7b97\u6cd5\u7684\u76f4\u89c2\u611f\u53d7 \u6211\u4eec\u53ef\u4ee5\u5c06\u4e00\u6b21\u5b8c\u6574\u7684\u524d\u5411&#043;\u53cd\u5411\u4f20\u64ad\u8fc7\u7a0b&#xff0c;\u6bd4\u4f5c\u4e00\u6b21\u9ad8\u6548\u7684\u201c\u8d23\u4efb\u5206\u914d\u201d\u5927\u4f1a&#xff1a;<\/p>\n<ul>\n<li>\u524d\u5411\u4f20\u64ad\u662f\u201c\u4e1a\u7ee9\u6c47\u62a5\u201d&#xff1a;\u6a21\u578b\u6839\u636e\u73b0\u6709\u80fd\u529b&#xff08;\u53c2\u6570&#xff09;&#xff0c;\u5bf9\u8f93\u5165\u6570\u636e\u505a\u51fa\u9884\u6d4b&#xff0c;\u5e76\u8ba1\u7b97\u51fa\u4e0e\u76ee\u6807&#xff08;\u771f\u5b9e\u6807\u7b7e&#xff09;\u4e4b\u95f4\u7684\u201c\u4e1a\u7ee9\u5dee\u989d\u201d&#xff08;Loss&#xff09;\u3002<\/li>\n<li>\u53cd\u5411\u4f20\u64ad\u662f\u201c\u8d23\u4efb\u8ffd\u6eaf\u201d&#xff1a;\u4ece\u8fd9\u4e2a\u201c\u603b\u5dee\u989d\u201d\u5f00\u59cb&#xff0c;\u7b97\u6cd5\u50cf\u4e00\u4f4d\u660e\u5bdf\u79cb\u6beb\u7684\u5ba1\u8ba1\u5b98&#xff0c;\u5229\u7528\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u5c06\u603b\u8d23\u4efb\u7cbe\u51c6\u5730\u3001\u6309\u8d21\u732e\u5ea6\u5730\u5c42\u5c42\u5206\u89e3&#xff0c;\u6700\u7ec8\u843d\u5b9e\u5230\u6bcf\u4e00\u4e2a\u57fa\u5c42\u5458\u5de5&#xff08;\u53c2\u6570&#xff09;\u5934\u4e0a&#xff0c;\u660e\u786e\u5730\u544a\u8bc9\u5b83&#xff1a;\u201c\u60a8\u5bf9\u8fd9\u6b21\u7684\u9519\u8bef\u8d1f\u6709xx\u7a0b\u5ea6\u7684\u8d23\u4efb&#xff0c;\u60a8\u5e94\u8be5\u671d\u7740xx\u65b9\u5411\u6539\u6b63\u3002\u201d<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u5728\u6574\u4e2a\u6570\u636e\u96c6\u4e0a\u6210\u5343\u4e0a\u4e07\u6b21\u5730\u91cd\u590d\u8fd9\u4e2a\u201c\u6c47\u62a5-\u8ffd\u8d23-\u6539\u6b63\u201d\u7684\u5faa\u73af&#xff0c;\u795e\u7ecf\u7f51\u7edc\u8fd9\u4e2a\u5e9e\u5927\u7684\u7ec4\u7ec7&#xff0c;\u5c31\u80fd\u591f\u4ece\u4e00\u4e2a\u968f\u673a\u7684\u3001\u65e0\u5e8f\u7684\u72b6\u6001&#xff0c;\u9010\u6b65\u6f14\u5316\u4e3a\u4e00\u4e2a\u5206\u5de5\u660e\u786e\u3001\u914d\u5408\u9ed8\u5951\u3001\u80fd\u591f\u9ad8\u6548\u5b8c\u6210\u7279\u5b9a\u4efb\u52a1\u7684\u667a\u80fd\u7cfb\u7edf\u3002<\/p>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u4ece\u96f6\u5230\u4e00\u7684\u6784\u5efa\u3002\u4ece\u751f\u7269\u795e\u7ecf\u5143\u7684\u7075\u611f\u95ea\u73b0&#xff0c;\u5230\u4eba\u5de5\u795e\u7ecf\u5143\u7684\u6570\u5b66\u62bd\u8c61&#xff1b;\u4ece\u5355\u5c42\u611f\u77e5\u673a\u7684\u7ebf\u6027\u5c40\u9650&#xff0c;\u5230\u591a\u5c42\u611f\u77e5\u673a\u901a\u8fc7\u5f15\u5165\u9690\u85cf\u5c42\u548c\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\u6240\u5b9e\u73b0\u7684\u7a81\u7834&#xff1b;\u6211\u4eec\u8fd8\u4e3a\u7f51\u7edc\u88c5\u914d\u4e86\u8861\u91cf\u5bf9\u9519\u7684\u201c\u635f\u5931\u51fd\u6570\u201d&#xff0c;\u5e76\u6700\u7ec8\u63ed\u793a\u4e86\u9a71\u52a8\u5176\u5b66\u4e60\u7684\u5f15\u64ce\u2014\u2014\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u3002<\/p>\n<p>\u8bfb\u8005\u73b0\u5728\u5e94\u8be5\u5bf9\u4e00\u4e2a\u57fa\u672c\u7684\u795e\u7ecf\u7f51\u7edc\u662f\u5982\u4f55\u6784\u6210\u3001\u5982\u4f55\u5de5\u4f5c\u3001\u4ee5\u53ca\u5982\u4f55\u5b66\u4e60\u7684&#xff0c;\u6709\u4e86\u4e00\u4e2a\u6e05\u6670\u800c\u5b8c\u6574\u7684\u8ba4\u8bc6\u3002\u8fd9\u4e9b\u57fa\u7840\u6982\u5ff5&#xff0c;\u662f\u7406\u89e3\u540e\u7eed\u6240\u6709\u66f4\u9ad8\u7ea7\u3001\u66f4\u590d\u6742\u7684\u7f51\u7edc\u7ed3\u6784&#xff08;\u5982\u5377\u79ef\u795e\u7ecf\u7f51\u7edcCNN\u3001\u5faa\u73af\u795e\u7ecf\u7f51\u7edcRNN&#xff09;\u7684\u7edd\u5bf9\u57fa\u77f3\u3002\u5e26\u7740\u8fd9\u4efd\u575a\u5b9e\u7684\u57fa\u7840&#xff0c;\u6211\u4eec\u5373\u5c06\u5728\u4e0b\u4e00\u7ae0&#xff0c;\u771f\u6b63\u8e0f\u5165\u4f7f\u7528\u73b0\u4ee3\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u8fdb\u884c\u5b9e\u6218\u7684\u6fc0\u52a8\u4eba\u5fc3\u7684\u65c5\u7a0b\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u4e94\u7ae0&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u5165\u95e8\u4e0e\u5b9e\u6218<\/h3>\n<p>\u4ece\u7406\u8bba\u5230\u4ee3\u7801 \u2014\u2014 \u5de8\u4eba\u7684\u529b\u91cf<\/p>\n<p>\u5728\u524d\u9762\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u5df2\u7ecf\u6df1\u5165\u63a2\u7d22\u4e86\u795e\u7ecf\u7f51\u7edc\u7684\u7406\u8bba\u57fa\u77f3\u3002\u6211\u4eec\u7406\u89e3\u4e86\u795e\u7ecf\u5143\u7684\u6570\u5b66\u6a21\u578b&#xff0c;\u89c1\u8bc1\u4e86\u201c\u6df1\u5ea6\u201d\u5982\u4f55\u8d4b\u4e88\u7f51\u7edc\u5f3a\u5927\u7684\u8868\u8fbe\u80fd\u529b&#xff0c;\u4e5f\u63ed\u793a\u4e86\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u9a71\u52a8\u6a21\u578b\u5b66\u4e60\u7684\u5965\u79d8\u3002\u7136\u800c&#xff0c;\u5982\u679c\u5c06\u8fd9\u4e9b\u7406\u8bba\u4ece\u96f6\u5f00\u59cb&#xff0c;\u7528\u7eaf\u7cb9\u7684Python\u548cNumPy\u6765\u5b9e\u73b0\u4e00\u4e2a\u590d\u6742\u7684\u73b0\u4ee3\u795e\u7ecf\u7f51\u7edc&#xff0c;\u90a3\u5c06\u662f\u4e00\u9879\u6781\u5176\u7e41\u7410\u3001\u6613\u9519\u4e14\u6548\u7387\u4f4e\u4e0b\u7684\u5de5\u7a0b\u3002\u6211\u4eec\u5c06\u4e0d\u5f97\u4e0d\u624b\u52a8\u7ba1\u7406\u6210\u5343\u4e0a\u4e07\u7684\u53c2\u6570&#xff0c;\u4eb2\u624b\u7f16\u5199\u590d\u6742\u7684\u68af\u5ea6\u8ba1\u7b97\u94fe\u6761&#xff0c;\u5e76\u8d39\u5c3d\u5fc3\u529b\u53bb\u8c03\u7528\u5e95\u5c42\u786c\u4ef6\u8fdb\u884c\u52a0\u901f\u3002<\/p>\n<p>\u5e78\u8fd0\u7684\u662f&#xff0c;\u6211\u4eec\u4e0d\u5fc5\u91cd\u590d\u9020\u8f6e\u3002\u6211\u4eec\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u8fd9\u4e9b\u5de8\u4eba&#xff0c;\u5c31\u662f\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u3002<\/p>\n<p>\u4e00\u4e2a\u73b0\u4ee3\u7684\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6&#xff0c;\u5982\u540c\u4e00\u4e2a\u4e13\u4e3a\u795e\u7ecf\u7f51\u7edc\u6253\u9020\u7684\u5f3a\u5927\u201c\u64cd\u4f5c\u7cfb\u7edf\u201d\u3002\u5b83\u4e3a\u6211\u4eec\u5c01\u88c5\u4e86\u6240\u6709\u7e41\u6742\u7684\u5e95\u5c42\u7ec6\u8282&#xff1a;\u4ece\u9ad8\u6548\u7684\u5f20\u91cf\u8fd0\u7b97&#xff0c;\u5230\u795e\u5947\u7684\u81ea\u52a8\u6c42\u5bfc\u673a\u5236&#xff0c;\u518d\u5230\u5bf9GPU\/TPU\u7b49\u4e13\u7528\u786c\u4ef6\u7684\u65e0\u7f1d\u5229\u7528\u3002\u5b83\u5c06\u6211\u4eec\u4ece\u7eb7\u7e41\u7684\u5de5\u7a0b\u7ec6\u8282\u4e2d\u89e3\u653e\u51fa\u6765&#xff0c;\u8ba9\u6211\u4eec\u80fd\u591f\u5c06\u5b9d\u8d35\u7684\u7cbe\u529b\u4e13\u6ce8\u4e8e\u6a21\u578b\u67b6\u6784\u672c\u8eab\u7684\u8bbe\u8ba1\u3001\u5b9e\u9a8c\u4e0e\u521b\u65b0\u3002<\/p>\n<p>\u5728\u5f53\u4eca\u7684\u6df1\u5ea6\u5b66\u4e60\u4e16\u754c&#xff0c;\u7fa4\u96c4\u5e76\u8d77&#xff0c;\u4f46\u771f\u6b63\u7684\u821e\u53f0\u4e3b\u89d2&#xff0c;\u65e0\u7591\u662f\u4e24\u5927\u5de8\u5934&#xff1a;\u7531Google\u652f\u6301\u7684TensorFlow\u548c\u7531Facebook&#xff08;Meta&#xff09;\u4e3b\u5bfc\u7684PyTorch\u3002\u5b83\u4eec\u5982\u540c\u6b66\u6797\u4e2d\u7684\u5c11\u6797\u4e0e\u6b66\u5f53&#xff0c;\u5404\u81ea\u4ee3\u8868\u4e86\u4e00\u79cd\u72ec\u7279\u7684\u6c14\u8d28\u4e0e\u54f2\u5b66&#xff1a;TensorFlow\u4ee5\u5176\u65e0\u4e0e\u4f26\u6bd4\u7684\u751f\u6001\u7cfb\u7edf\u3001\u5de5\u4e1a\u7ea7\u7684\u90e8\u7f72\u80fd\u529b\u548c\u7a33\u5065\u6027\u8457\u79f0&#xff0c;\u662f\u4e1a\u754c\u5e94\u7528\u7684\u9996\u9009&#xff1b;\u800cPyTorch\u5219\u4ee5\u5176\u7075\u52a8\u3001\u4f18\u96c5\u3001\u4e0ePython\u6df1\u5ea6\u878d\u5408\u7684\u8bbe\u8ba1&#xff0c;\u8d62\u5f97\u4e86\u5e7f\u5927\u5b66\u672f\u7814\u7a76\u8005\u548c\u7b97\u6cd5\u5de5\u7a0b\u5e08\u7684\u9752\u7750\u3002<\/p>\n<p>\u7cbe\u901a\u5176\u4e2d\u4e4b\u4e00&#xff0c;\u8db3\u4ee5\u8ba9\u60a8\u5728\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u6e38\u5203\u6709\u4f59&#xff1b;\u800c\u4e24\u8005\u7686\u901a&#xff0c;\u5219\u80fd\u8ba9\u60a8\u6d1e\u5bdf\u4e0d\u540c\u8bbe\u8ba1\u54f2\u5b66\u4e0b\u7684\u5229\u5f0a\u6743\u8861&#xff0c;\u6839\u636e\u4e0d\u540c\u573a\u666f\u9009\u62e9\u6700\u5408\u9002\u7684\u5de5\u5177\u3002<\/p>\n<p>\u672c\u7ae0\u7684\u76ee\u6807&#xff0c;\u6b63\u662f\u8981\u5e26\u9886\u8bfb\u8005\u8de8\u8d8a\u4ece\u7406\u8bba\u5230\u4ee3\u7801\u7684\u9e3f\u6c9f\u3002\u6211\u4eec\u5c06\u5e76\u80a9\u63a2\u7d22TensorFlow 2.x\u4e0ePyTorch\u7684\u6838\u5fc3\u8bbe\u8ba1\u4e0e\u4f7f\u7528\u65b9\u6cd5&#xff0c;\u901a\u8fc7\u4eb2\u624b\u7f16\u5199\u4ee3\u7801&#xff0c;\u5b8c\u6210\u4ece\u6570\u636e\u51c6\u5907\u3001\u6a21\u578b\u642d\u5efa\u3001\u8bad\u7ec3\u5230\u8bc4\u4f30\u7684\u5b8c\u6574\u6d41\u7a0b\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u8fd8\u5c06\u63d0\u4f9b\u4e00\u4efd\u8be6\u5c3d\u7684\u73af\u5883\u642d\u5efa\u6307\u5357&#xff0c;\u5e2e\u52a9\u60a8\u6253\u9020\u4e00\u4e2a\u5c5e\u4e8e\u81ea\u5df1\u7684\u3001\u9ad8\u6548\u7a33\u5b9a\u7684\u6df1\u5ea6\u5b66\u4e60\u201c\u70bc\u4e39\u7089\u201d\u3002<\/p>\n<p>\u51c6\u5907\u597d\u5c06\u7406\u8bba\u5316\u4e3a\u6307\u5c16\u7684\u529b\u91cf\u4e86\u5417&#xff1f;\u8ba9\u6211\u4eec\u5373\u523b\u542f\u7a0b\u3002<\/p>\n<hr \/>\n<h4>5.1 TensorFlow 2.x \u4e0e Keras&#xff1a;Google\u7684\u5de5\u4e1a\u7ea7\u89e3\u51b3\u65b9\u6848<\/h4>\n<p>TensorFlow\u662f\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u65e0\u53ef\u4e89\u8bae\u7684\u91cd\u91cf\u7ea7\u9009\u624b\u3002\u5b83\u4e0d\u4ec5\u4ec5\u662f\u4e00\u4e2a\u5e93&#xff0c;\u66f4\u662f\u4e00\u4e2a\u5e9e\u5927\u7684\u751f\u6001\u7cfb\u7edf&#xff0c;\u6db5\u76d6\u4e86\u4ece\u7814\u7a76\u539f\u578b\u5230\u751f\u4ea7\u90e8\u7f72&#xff0c;\u518d\u5230\u79fb\u52a8\u7aef\u548c\u7269\u8054\u7f51\u8bbe\u5907\u7684\u5168\u94fe\u8def\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n<h5>5.1.1 \u6846\u67b6\u7684\u6f14\u8fdb\u4e0e\u54f2\u5b66<\/h5>\n<p>\u4eceTensorFlow 1.x\u52302.x&#xff1a;\u4e00\u573a\u6df1\u523b\u7684\u53d8\u9769 \u65e9\u671f\u7684TensorFlow 1.x\u91c7\u7528\u7684\u662f\u4e00\u79cd\u79f0\u4e3a**\u201c\u9759\u6001\u56fe\u201d&#xff08;Define-and-Run&#xff09;**\u7684\u6a21\u5f0f\u3002\u5f00\u53d1\u8005\u9700\u8981\u5148\u50cf\u7ed8\u5236\u7535\u8def\u56fe\u4e00\u6837&#xff0c;\u5b8c\u6574\u5730\u5b9a\u4e49\u51fa\u6574\u4e2a\u8ba1\u7b97\u56fe\u7684\u7ed3\u6784&#xff0c;\u7136\u540e\u518d\u5411\u8fd9\u4e2a\u56fa\u5b9a\u7684\u56fe\u4e2d\u201c\u6ce8\u5165\u201d\u6570\u636e\u6765\u6267\u884c\u8ba1\u7b97\u3002\u8fd9\u79cd\u6a21\u5f0f\u867d\u7136\u6709\u5229\u4e8e\u8fdb\u884c\u5168\u5c40\u4f18\u5316\u548c\u8de8\u5e73\u53f0\u90e8\u7f72&#xff0c;\u4f46\u5176\u7f16\u5199\u548c\u8c03\u8bd5\u8fc7\u7a0b\u5374\u76f8\u5f53\u7e41\u7410\u548c\u53cd\u76f4\u89c9&#xff0c;\u4ee4\u8bb8\u591a\u521d\u5b66\u8005\u671b\u800c\u5374\u6b65\u3002<\/p>\n<p>\u8ba4\u8bc6\u5230\u8fd9\u4e00\u70b9\u540e&#xff0c;Google\u5728TensorFlow 2.x\u4e2d\u8fdb\u884c\u4e86\u4e00\u573a\u6df1\u523b\u7684\u81ea\u6211\u9769\u547d&#xff0c;\u5176\u6838\u5fc3\u662f\u62e5\u62b1\u4e86\u201c\u52a8\u6001\u56fe\u201d&#xff08;Define-by-Run&#xff09;\u4f5c\u4e3a\u9ed8\u8ba4\u6267\u884c\u6a21\u5f0f\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7801\u4f1a\u50cf\u666e\u901a\u7684Python\u7a0b\u5e8f\u4e00\u6837&#xff0c;\u6309\u987a\u5e8f\u7acb\u5373\u6267\u884c&#xff0c;\u8ba1\u7b97\u56fe\u5728\u8fd0\u884c\u65f6\u52a8\u6001\u6784\u5efa\u3002\u8fd9\u4f7f\u5f97\u8c03\u8bd5\u53d8\u5f97\u5f02\u5e38\u7b80\u5355&#xff0c;\u4ee3\u7801\u4e5f\u66f4\u52a0\u76f4\u89c2\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;TF 2.x\u505a\u51fa\u4e86\u4e00\u4e2a\u660e\u667a\u7684\u51b3\u5b9a&#xff1a;\u5c06Keras\u63d0\u5347\u4e3a\u5b98\u65b9\u552f\u4e00\u6307\u5b9a\u7684\u9ad8\u7ea7API\u3002<\/p>\n<p>Keras&#xff1a;\u4e3a\u4eba\u7c7b\u800c\u975e\u673a\u5668\u8bbe\u8ba1\u7684API Keras\u662f\u7531Fran\u00e7ois Chollet\u521b\u5efa\u7684\u4e00\u4e2a\u6df1\u5ea6\u5b66\u4e60\u5e93&#xff0c;\u5176\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66\u662f**\u201c\u6613\u7528\u6027\u4f18\u5148\u201d**\u3002\u5b83\u8ffd\u6c42\u4ee5\u6700\u5c11\u7684\u4ee3\u7801\u3001\u6700\u6e05\u6670\u7684\u903b\u8f91\u6765\u6784\u5efa\u795e\u7ecf\u7f51\u7edc&#xff0c;\u8ba9\u5f00\u53d1\u8005\u80fd\u591f\u5feb\u901f\u5730\u5c06\u60f3\u6cd5\u8f6c\u5316\u4e3a\u5b9e\u9a8c\u7ed3\u679c\u3002Keras\u7684API\u7b80\u6d01\u3001\u9ad8\u5ea6\u6a21\u5757\u5316\u4e14\u6613\u4e8e\u6269\u5c55&#xff0c;\u5982\u540c\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u201c\u4e50\u9ad8\u79ef\u6728\u201d\u3002\u5728TF 2.x\u4e2d&#xff0c;Keras\u4e0d\u518d\u662f\u4e00\u4e2a\u72ec\u7acb\u7684\u5e93&#xff0c;\u800c\u662f\u6df1\u5ea6\u6574\u5408\u4e3atf.keras&#xff0c;\u6210\u4e3a\u4e86\u4e0eTensorFlow\u4ea4\u4e92\u7684\u9996\u9009\u65b9\u5f0f\u3002<\/p>\n<h5>5.1.2 TensorFlow\u6838\u5fc3\u6982\u5ff5&#xff1a;\u6df1\u5165\u5f15\u64ce\u5ba4<\/h5>\n<p>\u5c3d\u7ba1Keras\u4e3a\u6211\u4eec\u9690\u85cf\u4e86\u8bb8\u591a\u5e95\u5c42\u7ec6\u8282&#xff0c;\u4f46\u7406\u89e3\u5176\u80cc\u540e\u7684\u6838\u5fc3\u7ec4\u4ef6&#xff0c;\u80fd\u8ba9\u6211\u4eec\u5728\u4f7f\u7528\u65f6\u66f4\u52a0\u5f97\u5fc3\u5e94\u624b\u3002<\/p>\n<p>\u5f20\u91cf&#xff08;tf.Tensor&#xff09;&#xff1a;\u4e0d\u53ef\u53d8\u7684\u57fa\u77f3 \u6b63\u5982\u6211\u4eec\u5728\u7b2c\u4e8c\u7ae0\u6240\u5b66&#xff0c;\u5f20\u91cf\u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u57fa\u672c\u6570\u636e\u5355\u5143\u3002\u5728TensorFlow\u4e2d&#xff0c;tf.Tensor\u5bf9\u8c61\u662f\u627f\u8f7d\u6570\u636e\u7684\u6838\u5fc3\u3002\u5b83\u4e0eNumPy\u7684ndarray\u975e\u5e38\u76f8\u4f3c&#xff0c;\u53ef\u4ee5\u5b58\u50a8\u6807\u91cf\u3001\u5411\u91cf\u3001\u77e9\u9635\u7b49\u3002\u4e00\u4e2a\u5173\u952e\u7279\u6027\u662f&#xff0c;tf.Tensor\u662f\u4e0d\u53ef\u53d8\u7684&#xff08;immutable&#xff09;\u3002\u4e00\u65e6\u521b\u5efa&#xff0c;\u60a8\u65e0\u6cd5\u6539\u53d8\u5b83\u7684\u503c&#xff0c;\u53ea\u80fd\u901a\u8fc7\u8fd0\u7b97\u521b\u5efa\u65b0\u7684\u5f20\u91cf\u3002<\/p>\n<p>import tensorflow as tf<\/p>\n<p># \u521b\u5efa\u4e00\u4e2a\u5e38\u91cf\u5f20\u91cf<br \/>\na &#061; tf.constant([[1.0, 2.0], [3.0, 4.0]])<br \/>\nprint(a)<\/p>\n<p># \u5f20\u91cf\u8fd0\u7b97\u4f1a\u521b\u5efa\u65b0\u7684\u5f20\u91cf<br \/>\nb &#061; a &#043; 10<br \/>\nprint(b)<\/p>\n<p># \u4e0eNumPy\u7684\u65e0\u7f1d\u8f6c\u6362<br \/>\nnumpy_array &#061; b.numpy()<br \/>\nprint(type(numpy_array))<\/p>\n<p>\u53d8\u91cf&#xff08;tf.Variable&#xff09;&#xff1a;\u53ef\u53d8\u7684\u6a21\u578b\u53c2\u6570 \u65e2\u7136\u5f20\u91cf\u662f\u4e0d\u53ef\u53d8\u7684&#xff0c;\u90a3\u6211\u4eec\u5982\u4f55\u5b58\u50a8\u548c\u66f4\u65b0\u9700\u8981\u5728\u8bad\u7ec3\u4e2d\u4e0d\u65ad\u53d8\u5316\u7684**\u6a21\u578b\u53c2\u6570&#xff08;\u6743\u91cd\u548c\u504f\u7f6e&#xff09;**\u5462&#xff1f;\u7b54\u6848\u662f\u4f7f\u7528tf.Variable\u3002\u5b83\u662f\u4e00\u4e2a\u7279\u6b8a\u7684\u3001\u53ef\u53d8\u7684\u5f20\u91cf&#xff0c;\u4e13\u95e8\u7528\u4e8e\u5b58\u50a8\u6a21\u578b\u72b6\u6001\u3002<\/p>\n<p># \u521b\u5efa\u4e00\u4e2a\u53d8\u91cf<br \/>\nv &#061; tf.Variable([[1.0, 2.0], [3.0, 4.0]])<br \/>\nprint(f&#034;Is Variable: {tf.is_tensor(v)}&#034;) # Variable\u4e5f\u662f\u4e00\u79cdTensor<\/p>\n<p># \u53ef\u4ee5\u4f7f\u7528 .assign() \u65b9\u6cd5\u5c31\u5730\u4fee\u6539\u503c<br \/>\nv.assign(v &#043; 10)<br \/>\nprint(v)<\/p>\n<p>\u81ea\u52a8\u6c42\u5bfc&#xff08;tf.GradientTape&#xff09;&#xff1a;\u53cd\u5411\u4f20\u64ad\u7684\u9b54\u672f\u5e08 tf.GradientTape\u662fTensorFlow 2.x\u5b9e\u73b0\u81ea\u52a8\u6c42\u5bfc\u7684\u6838\u5fc3\u5de5\u5177\u3002\u5b83\u7684\u5de5\u4f5c\u539f\u7406\u53ef\u4ee5\u7528\u4e00\u4e2a\u751f\u52a8\u7684\u6bd4\u55bb\u6765\u7406\u89e3&#xff1a;<\/p>\n<p>\u78c1\u5e26\u5f55\u97f3\u673a\u6bd4\u55bb&#xff1a; \u5f53\u60a8\u521b\u5efa\u4e00\u4e2atf.GradientTape\u7684\u4e0a\u4e0b\u6587\u73af\u5883\u65f6&#xff0c;\u5c31\u597d\u6bd4\u6309\u4e0b\u4e86\u4e00\u53f0\u8001\u5f0f\u78c1\u5e26\u5f55\u97f3\u673a\u7684\u201c\u5f55\u5236\u201d\u6309\u94ae\u3002\u5728\u6b64\u73af\u5883\u4e2d\u53d1\u751f\u7684\u6240\u6709\u6d89\u53catf.Variable\u6216\u88ab\u76d1\u89c6\u7684tf.Tensor\u7684\u8fd0\u7b97&#xff0c;\u90fd\u4f1a\u88ab\u8fd9\u53f0\u201c\u5f55\u97f3\u673a\u201d\u8bb0\u5f55\u5728\u4e00\u6761\u865a\u62df\u7684\u201c\u78c1\u5e26\u201d\u4e0a\u3002<\/p>\n<p>\u5f53\u60a8\u5b8c\u6210\u4e86\u524d\u5411\u4f20\u64ad\u7684\u8ba1\u7b97&#xff08;\u6bd4\u5982\u7b97\u51fa\u4e86\u635f\u5931&#xff09;&#xff0c;\u5c31\u53ef\u4ee5\u8c03\u7528\u78c1\u5e26\u7684.gradient()\u65b9\u6cd5\u3002\u8fd9\u597d\u6bd4\u6309\u4e0b\u4e86\u201c\u5012\u5e26\u201d\u5e76\u201c\u64ad\u653e\u201d&#xff0c;TensorFlow\u4f1a\u6cbf\u7740\u8bb0\u5f55\u7684\u8def\u5f84\u53cd\u5411\u8ffd\u6eaf&#xff0c;\u5e76\u5229\u7528\u94fe\u5f0f\u6cd5\u5219\u81ea\u52a8\u8ba1\u7b97\u51fa\u60a8\u6307\u5b9a\u7684\u76ee\u6807&#xff08;\u5982\u635f\u5931&#xff09;\u5173\u4e8e\u6240\u6709\u88ab\u8bb0\u5f55\u7684\u53d8\u91cf\u7684\u68af\u5ea6\u3002<\/p>\n<p># \u6f14\u793a\u8ba1\u7b97 y &#061; x\u00b2 \u5728 x&#061;3 \u65f6\u7684\u68af\u5ea6<br \/>\nx &#061; tf.Variable(3.0)<\/p>\n<p>with tf.GradientTape() as tape:<br \/>\n  y &#061; x ** 2<\/p>\n<p># \u8ba1\u7b97\u68af\u5ea6 dy\/dx<br \/>\ngrad &#061; tape.gradient(y, x)<br \/>\nprint(grad) # \u8f93\u51fa tf.Tensor(6.0, shape&#061;(), dtype&#061;float32)<\/p>\n<h5>5.1.3 Keras\u5b9e\u6218&#xff1a;\u4e09\u6b65\u5b8c\u6210\u6a21\u578b\u642d\u5efa<\/h5>\n<p>\u638c\u63e1\u4e86Keras&#xff0c;\u6784\u5efa\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\u5c31\u50cf\u642d\u79ef\u6728\u4e00\u6837\u7b80\u5355&#xff0c;\u901a\u5e38\u9075\u5faa\u201c\u5b9a\u4e49-\u7f16\u8bd1-\u8bad\u7ec3\u201d\u4e09\u90e8\u66f2\u3002<\/p>\n<p>\u5e8f\u8d2f\u6a21\u578b&#xff08;Sequential API&#xff09;&#xff1a;\u5feb\u901f\u642d\u5efa\u7ebf\u6027\u5806\u53e0 Sequential\u6a21\u578b\u662f\u6700\u7b80\u5355\u3001\u6700\u5e38\u7528\u7684\u6a21\u578b\u7c7b\u578b&#xff0c;\u9002\u7528\u4e8e\u7edd\u5927\u591a\u6570\u5c42\u4e0e\u5c42\u4e4b\u95f4\u5448\u7ebf\u6027\u987a\u5e8f\u5806\u53e0\u7684\u7f51\u7edc\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218&#xff1a;\u6784\u5efa\u4e00\u4e2a\u7b80\u5355\u7684MLP\u7528\u4e8e\u5206\u7c7b<\/p>\n<p>import tensorflow as tf<br \/>\nfrom tensorflow import keras<br \/>\nfrom tensorflow.keras import layers<\/p>\n<p># 1. \u5b9a\u4e49\u6a21\u578b&#xff1a;\u50cf\u642d\u79ef\u6728\u4e00\u6837\u6dfb\u52a0\u5c42<br \/>\nmodel &#061; keras.Sequential([<br \/>\n    # \u8f93\u5165\u5c42&#xff1a;\u5c0628&#215;28\u7684\u56fe\u7247\u5c55\u5e73\u4e3a784\u7ef4\u7684\u5411\u91cf<br \/>\n    layers.Flatten(input_shape&#061;(28, 28)),<br \/>\n    # \u7b2c\u4e00\u4e2a\u9690\u85cf\u5c42&#xff1a;128\u4e2a\u795e\u7ecf\u5143&#xff0c;\u4f7f\u7528ReLU\u6fc0\u6d3b\u51fd\u6570<br \/>\n    layers.Dense(128, activation&#061;&#039;relu&#039;),<br \/>\n    # \u7b2c\u4e8c\u4e2a\u9690\u85cf\u5c42&#xff1a;64\u4e2a\u795e\u7ecf\u5143&#xff0c;\u4f7f\u7528ReLU\u6fc0\u6d3b\u51fd\u6570<br \/>\n    layers.Dense(64, activation&#061;&#039;relu&#039;),<br \/>\n    # \u8f93\u51fa\u5c42&#xff1a;10\u4e2a\u795e\u7ecf\u5143&#xff08;\u5bf9\u5e9410\u4e2a\u7c7b\u522b&#xff09;&#xff0c;\u4f7f\u7528Softmax\u8f93\u51fa\u6982\u7387<br \/>\n    layers.Dense(10, activation&#061;&#039;softmax&#039;)<br \/>\n])<\/p>\n<p># \u6253\u5370\u6a21\u578b\u7ed3\u6784<br \/>\nmodel.summary()<\/p>\n<p># 2. \u7f16\u8bd1\u6a21\u578b&#xff1a;\u914d\u7f6e\u4f18\u5316\u5668\u3001\u635f\u5931\u51fd\u6570\u548c\u8bc4\u4f30\u6307\u6807<br \/>\nmodel.compile(optimizer&#061;&#039;adam&#039;,<br \/>\n              loss&#061;&#039;sparse_categorical_crossentropy&#039;,<br \/>\n              metrics&#061;[&#039;accuracy&#039;])<\/p>\n<p># \u5047\u8bbe\u6211\u4eec\u6709\u51c6\u5907\u597d\u7684\u8bad\u7ec3\u6570\u636e (x_train, y_train) \u548c\u6d4b\u8bd5\u6570\u636e (x_test, y_test)<br \/>\n# (x_train, y_train), (x_test, y_test) &#061; keras.datasets.mnist.load_data()<br \/>\n# x_train, x_test &#061; x_train \/ 255.0, x_test \/ 255.0 # \u5f52\u4e00\u5316<\/p>\n<p># 3. \u8bad\u7ec3\u6a21\u578b&#xff1a;\u5c06\u6570\u636e\u201c\u5582\u201d\u7ed9\u6a21\u578b<br \/>\n# history &#061; model.fit(x_train, y_train, epochs&#061;5, batch_size&#061;32, validation_split&#061;0.2)<\/p>\n<p># \u8bc4\u4f30\u6a21\u578b<br \/>\n# test_loss, test_acc &#061; model.evaluate(x_test, y_test)<br \/>\n# print(f&#039;\\\\nTest accuracy: {test_acc}&#039;)<\/p>\n<p>\u51fd\u6570\u5f0fAPI&#xff08;Functional API&#xff09;&#xff1a;\u6784\u5efa\u590d\u6742\u7684\u975e\u7ebf\u6027\u7f51\u7edc \u5f53\u6a21\u578b\u7ed3\u6784\u53d8\u5f97\u590d\u6742&#xff0c;\u4f8b\u5982\u9700\u8981\u5904\u7406\u591a\u8f93\u5165\u3001\u591a\u8f93\u51fa&#xff0c;\u6216\u8005\u7f51\u7edc\u4e2d\u5b58\u5728\u5171\u4eab\u5c42\u3001\u6b8b\u5dee\u8fde\u63a5\u7b49\u975e\u7ebf\u6027\u62d3\u6251\u7ed3\u6784\u65f6&#xff0c;Sequential\u6a21\u578b\u4fbf\u65e0\u80fd\u4e3a\u529b\u3002\u8fd9\u65f6&#xff0c;\u66f4\u7075\u6d3b\u7684\u51fd\u6570\u5f0fAPI\u5c31\u6d3e\u4e0a\u4e86\u7528\u573a\u3002\u5b83\u5c06\u7f51\u7edc\u5c42\u89c6\u4e3a\u53ef\u4ee5\u8c03\u7528\u7684\u51fd\u6570&#xff0c;\u5141\u8bb8\u60a8\u6784\u5efa\u4efb\u610f\u7684\u6709\u5411\u65e0\u73af\u56fe\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218&#xff1a;\u6784\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u591a\u8f93\u5165\u6a21\u578b \u5047\u8bbe\u6211\u4eec\u6709\u4e00\u4e2a\u6a21\u578b&#xff0c;\u540c\u65f6\u63a5\u6536\u4e00\u4e2a\u4e3b\u8f93\u5165\u548c\u4e00\u4e2a\u8f85\u52a9\u8f93\u5165\u3002<\/p>\n<p># \u5b9a\u4e49\u8f93\u5165\u5c42<br \/>\nmain_input &#061; keras.Input(shape&#061;(100,), name&#061;&#039;main_input&#039;)<br \/>\naux_input &#061; keras.Input(shape&#061;(10,), name&#061;&#039;aux_input&#039;)<\/p>\n<p># \u4e3b\u6d41\u7a0b<br \/>\nx &#061; layers.Dense(64, activation&#061;&#039;relu&#039;)(main_input)<br \/>\nx &#061; layers.Dense(32, activation&#061;&#039;relu&#039;)(x)<\/p>\n<p># \u8f85\u52a9\u8f93\u5165\u4e0e\u4e3b\u6d41\u7a0b\u5408\u5e76<br \/>\nmerged &#061; layers.concatenate([x, aux_input])<\/p>\n<p># \u5b9a\u4e49\u8f93\u51fa\u5c42<br \/>\nmain_output &#061; layers.Dense(1, name&#061;&#039;main_output&#039;)(merged)<br \/>\naux_output &#061; layers.Dense(1, name&#061;&#039;aux_output&#039;)(x) # \u8f85\u52a9\u8f93\u51fa<\/p>\n<p># \u4f7f\u7528\u8f93\u5165\u548c\u8f93\u51fa\u5f20\u91cf\u5217\u8868\u6765\u5b9e\u4f8b\u5316\u6a21\u578b<br \/>\ncomplex_model &#061; keras.Model(inputs&#061;[main_input, aux_input],<br \/>\n                            outputs&#061;[main_output, aux_output])<\/p>\n<p>complex_model.summary()<\/p>\n<hr \/>\n<h4>5.2 PyTorch&#xff1a;Facebook\u7684\u52a8\u6001\u56fe\u4e0e\u7814\u7a76\u8005\u9996\u9009<\/h4>\n<p>\u5982\u679c\u8bf4TensorFlow\u662f\u4e25\u8c28\u7684\u5de5\u7a0b\u5e08&#xff0c;\u90a3\u4e48PyTorch\u5c31\u662f\u4e00\u4f4d\u7075\u52a8\u7684\u827a\u672f\u5bb6\u3002\u5b83\u4ee5\u5176\u4f18\u96c5\u548c\u5bf9Python\u8bed\u8a00\u7684\u6df1\u5ea6\u878d\u5408&#xff0c;\u6210\u4e3a\u4e86\u5b66\u672f\u754c\u548c\u5feb\u901f\u539f\u578b\u5f00\u53d1\u9886\u57df\u7684\u5ba0\u513f\u3002<\/p>\n<h5>5.2.1 \u6846\u67b6\u7684\u54f2\u5b66\u4e0e\u6c14\u8d28<\/h5>\n<ul>\n<li>\n<p>Pythonic\u7684\u8bbe\u8ba1 PyTorch\u7684\u8bbe\u8ba1\u54f2\u5b66\u662f\u201c\u5c3d\u53ef\u80fd\u5730\u50cfPython\u201d\u3002\u5b83\u7684API\u548c\u7528\u6cd5\u975e\u5e38\u7b26\u5408Python\u7a0b\u5e8f\u5458\u7684\u76f4\u89c9\u3002\u5728PyTorch\u4e2d&#xff0c;\u60a8\u53ef\u4ee5\u50cf\u8c03\u8bd5\u666e\u901aPython\u4ee3\u7801\u4e00\u6837&#xff0c;\u4f7f\u7528print()\u6216\u4efb\u4f55\u8c03\u8bd5\u5de5\u5177\u6765\u68c0\u67e5\u4efb\u610f\u4e2d\u95f4\u53d8\u91cf&#xff0c;\u8fd9\u5f97\u76ca\u4e8e\u5176\u52a8\u6001\u56fe\u7684\u672c\u8d28\u3002<\/p>\n<\/li>\n<li>\n<p>\u52a8\u6001\u56fe\u7684\u7075\u6d3b\u6027 PyTorch\u4ece\u8bde\u751f\u4e4b\u521d\u5c31\u91c7\u7528\u52a8\u6001\u8ba1\u7b97\u56fe&#xff08;Define-by-Run&#xff09;\u3002\u8ba1\u7b97\u56fe\u662f\u5728\u4ee3\u7801\u8fd0\u884c\u65f6&#xff0c;\u968f\u7740\u8fd0\u7b97\u7684\u53d1\u751f\u800c\u52a8\u6001\u5efa\u7acb\u7684\u3002\u8fd9\u610f\u5473\u7740\u60a8\u53ef\u4ee5\u4f7f\u7528Python\u6240\u6709\u539f\u751f\u7684\u63a7\u5236\u6d41\u8bed\u53e5&#xff08;\u5982if-else\u3001for\u5faa\u73af&#xff09;\u6765\u81ea\u7531\u5730\u6539\u53d8\u7f51\u7edc\u7684\u884c\u4e3a\u3002\u8fd9\u79cd\u7075\u6d3b\u6027\u5bf9\u4e8e\u5904\u7406\u53d8\u957f\u8f93\u5165\u7684\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;\u4efb\u52a1&#xff0c;\u6216\u8fdb\u884c\u590d\u6742\u7684\u7b97\u6cd5\u7814\u7a76\u81f3\u5173\u91cd\u8981\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>5.2.2 PyTorch\u6838\u5fc3\u6982\u5ff5&#xff1a;\u4e09\u5927\u652f\u67f1<\/h5>\n<ul>\n<li>\n<p>\u5f20\u91cf&#xff08;torch.Tensor&#xff09;&#xff1a;\u6570\u636e\u4e0e\u68af\u5ea6\u7684\u8f7d\u4f53 torch.Tensor\u662fPyTorch\u7684\u6838\u5fc3\u6570\u636e\u7ed3\u6784&#xff0c;\u4e0eNumPy\u7684ndarray\u6781\u4e3a\u76f8\u4f3c&#xff0c;\u5e76\u4e14\u4e24\u8005\u53ef\u4ee5\u9ad8\u6548\u3001\u65e0\u7f1d\u5730\u5171\u4eab\u5e95\u5c42\u5185\u5b58\u8fdb\u884c\u8f6c\u6362&#xff08;\u5728CPU\u4e0a&#xff09;&#xff0c;\u907f\u514d\u4e86\u6570\u636e\u62f7\u8d1d\u7684\u5f00\u9500\u3002<\/p>\n<\/li>\n<li>\n<p>\u81ea\u52a8\u6c42\u5bfc&#xff08;torch.autograd&#xff09;&#xff1a;\u52a8\u6001\u56fe\u7684\u5f15\u64ce autograd\u662fPyTorch\u7684\u81ea\u52a8\u6c42\u5bfc\u5f15\u64ce\u3002\u5b83\u7684\u5de5\u4f5c\u65b9\u5f0f\u4e0eGradientTape\u7c7b\u4f3c&#xff0c;\u4f46\u96c6\u6210\u5f97\u66f4\u52a0\u65e0\u7f1d&#xff1a;<\/p>\n<ul>\n<li>requires_grad\u5c5e\u6027&#xff1a;\u5f53\u60a8\u521b\u5efa\u4e00\u4e2a\u5f20\u91cf\u65f6&#xff0c;\u53ef\u4ee5\u8bbe\u7f6erequires_grad&#061;True\u3002\u8fd9\u4f1a\u544a\u8bc9PyTorch&#xff0c;\u9700\u8981\u8ffd\u8e2a\u6240\u6709\u53d1\u751f\u5728\u8be5\u5f20\u91cf\u4e0a\u7684\u64cd\u4f5c\u3002<\/li>\n<li>.backward()\u65b9\u6cd5&#xff1a;\u5728\u4e00\u4e2a\u6807\u91cf&#xff08;\u901a\u5e38\u662f\u8ba1\u7b97\u51fa\u7684loss&#xff09;\u4e0a\u8c03\u7528.backward()\u65b9\u6cd5&#xff0c;PyTorch\u4f1a\u81ea\u52a8\u8ba1\u7b97\u6240\u6709requires_grad&#061;True\u7684\u5f20\u91cf\u76f8\u5bf9\u4e8e\u8be5\u6807\u91cf\u7684\u68af\u5ea6\u3002<\/li>\n<li>.grad\u5c5e\u6027&#xff1a;\u8ba1\u7b97\u51fa\u7684\u68af\u5ea6\u4f1a\u7d2f\u79ef\u5230\u5bf9\u5e94\u5f20\u91cf\u7684.grad\u5c5e\u6027\u4e2d\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u795e\u7ecf\u7f51\u7edc\u6a21\u5757&#xff08;torch.nn&#xff09;&#xff1a;\u6a21\u578b\u7ec4\u4ef6\u7684\u84dd\u56fe torch.nn\u662fPyTorch\u4e2d\u7528\u4e8e\u6784\u5efa\u795e\u7ecf\u7f51\u7edc\u7684\u6838\u5fc3\u6a21\u5757\u3002<\/p>\n<ul>\n<li>nn.Module&#xff1a;\u6240\u6709\u795e\u7ecf\u7f51\u7edc\u5c42&#xff08;\u5982nn.Linear,\u00a0nn.Conv2d&#xff09;\u548c\u6a21\u578b\u5bb9\u5668\u7684\u57fa\u7c7b\u3002\u6784\u5efa\u81ea\u5b9a\u4e49\u6a21\u578b\u65f6&#xff0c;\u60a8\u9700\u8981\u521b\u5efa\u4e00\u4e2a\u7ee7\u627f\u81eann.Module\u7684\u7c7b\u3002\u5728__init__\u65b9\u6cd5\u4e2d\u5b9a\u4e49\u597d\u6240\u6709\u9700\u8981\u7684\u5c42&#xff0c;\u7136\u540e\u5728forward\u65b9\u6cd5\u4e2d&#xff0c;\u660e\u786e\u5730\u5b9a\u4e49\u6570\u636e\u662f\u5982\u4f55\u4ece\u8f93\u5165\u6d41\u5411\u8f93\u51fa\u7684&#xff08;\u5373\u524d\u5411\u4f20\u64ad\u7684\u903b\u8f91&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>5.2.3 PyTorch\u5b9e\u6218&#xff1a;\u6807\u51c6\u7684\u4e94\u6b65\u8bad\u7ec3\u6d41\u7a0b<\/h5>\n<p>PyTorch\u4e0d\u50cfKeras\u90a3\u6837\u63d0\u4f9b\u4e00\u4e2a\u9ad8\u5ea6\u5c01\u88c5\u7684.fit()\u65b9\u6cd5\u3002\u5b83\u9f13\u52b1&#xff08;\u6216\u8005\u8bf4&#xff0c;\u8981\u6c42&#xff09;\u7528\u6237\u81ea\u5df1\u7f16\u5199\u8bad\u7ec3\u5faa\u73af\u3002\u8fd9\u867d\u7136\u4ee3\u7801\u91cf\u7a0d\u591a&#xff0c;\u4f46\u8d4b\u4e88\u4e86\u7528\u6237\u5bf9\u8bad\u7ec3\u8fc7\u7a0b\u6bcf\u4e00\u4e2a\u7ec6\u8282\u7684\u5b8c\u5168\u63a7\u5236\u529b\u3002<\/p>\n<p>\u4ee3\u7801\u5b9e\u6218&#xff1a;\u4ece\u96f6\u6784\u5efaMLP\u8bad\u7ec3\u6d41\u7a0b<\/p>\n<p>import torch<br \/>\nimport torch.nn as nn<br \/>\nimport torch.optim as optim<br \/>\nfrom torch.utils.data import DataLoader, TensorDataset<\/p>\n<p># 1. \u51c6\u5907\u6570\u636e (\u5047\u8bbe\u6211\u4eec\u6709Numpy\u6570\u7ec4 x_train_np, y_train_np)<br \/>\n# x_train_tensor &#061; torch.from_numpy(x_train_np).float()<br \/>\n# y_train_tensor &#061; torch.from_numpy(y_train_np).long()<br \/>\n# train_dataset &#061; TensorDataset(x_train_tensor, y_train_tensor)<br \/>\n# train_loader &#061; DataLoader(dataset&#061;train_dataset, batch_size&#061;32, shuffle&#061;True)<\/p>\n<p># 2. \u5b9a\u4e49\u6a21\u578b&#xff1a;\u521b\u5efa\u4e00\u4e2a\u7ee7\u627f\u81ea nn.Module \u7684\u7c7b<br \/>\nclass MLP(nn.Module):<br \/>\n    def __init__(self):<br \/>\n        super(MLP, self).__init__()<br \/>\n        self.flatten &#061; nn.Flatten()<br \/>\n        self.layers &#061; nn.Sequential(<br \/>\n            nn.Linear(28*28, 128),<br \/>\n            nn.ReLU(),<br \/>\n            nn.Linear(128, 64),<br \/>\n            nn.ReLU(),<br \/>\n            nn.Linear(64, 10)<br \/>\n        )<\/p>\n<p>    def forward(self, x):<br \/>\n        x &#061; self.flatten(x)<br \/>\n        logits &#061; self.layers(x)<br \/>\n        return logits<\/p>\n<p>device &#061; &#034;cuda&#034; if torch.cuda.is_available() else &#034;cpu&#034;<br \/>\nmodel &#061; MLP().to(device)<br \/>\nprint(model)<\/p>\n<p># 3. \u5b9a\u4e49\u635f\u5931\u51fd\u6570\u548c\u4f18\u5316\u5668<br \/>\ncriterion &#061; nn.CrossEntropyLoss() # CrossEntropyLoss\u5185\u90e8\u5df2\u5305\u542bSoftmax<br \/>\noptimizer &#061; optim.Adam(model.parameters(), lr&#061;0.001)<\/p>\n<p># 4. \u7f16\u5199\u8bad\u7ec3\u5faa\u73af<br \/>\nnum_epochs &#061; 5<br \/>\n# for epoch in range(num_epochs):<br \/>\n#     model.train() # \u8bbe\u7f6e\u4e3a\u8bad\u7ec3\u6a21\u5f0f<br \/>\n#     for inputs, labels in train_loader:<br \/>\n#         inputs, labels &#061; inputs.to(device), labels.to(device)<br \/>\n#<br \/>\n#         # a. \u6e05\u7a7a\u68af\u5ea6<br \/>\n#         optimizer.zero_grad()<br \/>\n#<br \/>\n#         # b. \u524d\u5411\u4f20\u64ad<br \/>\n#         outputs &#061; model(inputs)<br \/>\n#<br \/>\n#         # c. \u8ba1\u7b97\u635f\u5931<br \/>\n#         loss &#061; criterion(outputs, labels)<br \/>\n#<br \/>\n#         # d. \u53cd\u5411\u4f20\u64ad<br \/>\n#         loss.backward()<br \/>\n#<br \/>\n#         # e. \u66f4\u65b0\u6743\u91cd<br \/>\n#         optimizer.step()<br \/>\n#<br \/>\n#     print(f&#039;Epoch [{epoch&#043;1}\/{num_epochs}], Loss: {loss.item():.4f}&#039;)<\/p>\n<p># 5. \u7f16\u5199\u8bc4\u4f30\u903b\u8f91<br \/>\n# model.eval() # \u8bbe\u7f6e\u4e3a\u8bc4\u4f30\u6a21\u5f0f<br \/>\n# with torch.no_grad(): # \u5728\u6b64\u4e0a\u4e0b\u6587\u4e2d&#xff0c;\u4e0d\u8ba1\u7b97\u68af\u5ea6&#xff0c;\u8282\u7701\u5185\u5b58\u548c\u8ba1\u7b97<br \/>\n#     correct &#061; 0<br \/>\n#     total &#061; 0<br \/>\n#     for inputs, labels in test_loader:<br \/>\n#         inputs, labels &#061; inputs.to(device), labels.to(device)<br \/>\n#         outputs &#061; model(inputs)<br \/>\n#         _, predicted &#061; torch.max(outputs.data, 1)<br \/>\n#         total &#043;&#061; labels.size(0)<br \/>\n#         correct &#043;&#061; (predicted &#061;&#061; labels).sum().item()<br \/>\n#     print(f&#039;Accuracy on test set: {100 * correct \/ total} %&#039;)<\/p>\n<hr \/>\n<h4>5.3 \u73af\u5883\u642d\u5efa&#xff1a;\u6253\u9020\u81ea\u5df1\u7684\u201c\u70bc\u4e39\u7089\u201d<\/h4>\n<p>\u5de5\u6b32\u5584\u5176\u4e8b&#xff0c;\u5fc5\u5148\u5229\u5176\u5668\u3002\u4e00\u4e2a\u7a33\u5b9a\u3001\u9694\u79bb\u3001\u53ef\u590d\u73b0\u7684\u5f00\u53d1\u73af\u5883&#xff0c;\u662f\u9ad8\u6548\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60\u7814\u7a76\u4e0e\u5b9e\u8df5\u7684\u57fa\u77f3\u3002<\/p>\n<h5>5.3.1 \u5305\u4e0e\u73af\u5883\u7ba1\u7406&#xff1a;Conda\u7684\u827a\u672f<\/h5>\n<p>\u4e3a\u4f55\u9700\u8981\u73af\u5883\u7ba1\u7406 \u5728\u8f6f\u4ef6\u5f00\u53d1\u4e2d&#xff0c;\u6211\u4eec\u7ecf\u5e38\u4f1a\u9047\u5230\u201c\u4f9d\u8d56\u5730\u72f1\u201d&#xff1a;\u9879\u76eeA\u9700\u8981\u5e93X\u76841.0\u7248\u672c&#xff0c;\u800c\u9879\u76eeB\u9700\u8981\u5e93X\u76842.0\u7248\u672c\u3002\u5982\u679c\u5728\u5168\u5c40\u73af\u5883\u4e2d\u5b89\u88c5&#xff0c;\u8fd9\u4e24\u4e2a\u9879\u76ee\u5c06\u65e0\u6cd5\u540c\u65f6\u5de5\u4f5c\u3002\u865a\u62df\u73af\u5883\u5c31\u662f\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u7684\u826f\u65b9\u3002\u5b83\u4e3a\u6bcf\u4e2a\u9879\u76ee\u521b\u5efa\u4e00\u4e2a\u72ec\u7acb\u7684\u3001\u9694\u79bb\u7684Python\u73af\u5883&#xff0c;\u60a8\u53ef\u4ee5\u5728\u5176\u4e2d\u5b89\u88c5\u4efb\u610f\u7248\u672c\u7684\u5e93&#xff0c;\u800c\u4e0d\u4f1a\u5f71\u54cd\u5230\u5176\u4ed6\u9879\u76ee\u3002<\/p>\n<p>Conda\u6838\u5fc3\u547d\u4ee4 Conda\u662f\u76ee\u524d\u6700\u6d41\u884c\u3001\u6700\u5f3a\u5927\u7684Python\u73af\u5883\u548c\u5305\u7ba1\u7406\u5668\u4e4b\u4e00\u3002<\/p>\n<ul>\n<li>\u521b\u5efa\u65b0\u73af\u5883&#xff1a;conda create &#8211;name my_dl_env python&#061;3.9<\/li>\n<li>\u6fc0\u6d3b\u73af\u5883&#xff1a;conda activate my_dl_env<\/li>\n<li>\u5b89\u88c5\u5305&#xff1a;conda install numpy pandas matplotlib\u00a0\u6216\u00a0pip install tensorflow<\/li>\n<li>\u67e5\u770b\u5df2\u5b89\u88c5\u7684\u5305&#xff1a;conda list<\/li>\n<li>\u9000\u51fa\u73af\u5883&#xff1a;conda deactivate<\/li>\n<\/ul>\n<h5>5.3.2 \u4ea4\u4e92\u5f0f\u7f16\u7a0b\u73af\u5883&#xff1a;Jupyter Notebook\/Lab<\/h5>\n<ul>\n<li>\n<p>Jupyter\u7684\u4f18\u52bf Jupyter Notebook\/Lab\u662f\u4e00\u4e2a\u57fa\u4e8eWeb\u7684\u4ea4\u4e92\u5f0f\u8ba1\u7b97\u73af\u5883\u3002\u5b83\u5141\u8bb8\u60a8\u521b\u5efa\u548c\u5171\u4eab\u5305\u542b\u5b9e\u65f6\u4ee3\u7801\u3001\u516c\u5f0f\u3001\u53ef\u89c6\u5316\u548c\u53d9\u8ff0\u6027\u6587\u672c\u7684\u6587\u6863\u3002\u8fd9\u79cd\u201c\u6587\u5b66\u5f0f\u7f16\u7a0b\u201d\u7684\u98ce\u683c&#xff0c;\u4f7f\u5176\u6210\u4e3a\u6570\u636e\u79d1\u5b66\u63a2\u7d22\u3001\u5feb\u901f\u539f\u578b\u9a8c\u8bc1\u548c\u6559\u5b66\u6f14\u793a\u7684\u7406\u60f3\u5de5\u5177\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b89\u88c5\u4e0e\u542f\u52a8 \u5728\u6fc0\u6d3b\u4e86\u60a8\u7684Conda\u73af\u5883\u540e&#xff0c;\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5&#xff1a; conda install -c conda-forge jupyterlab \u7136\u540e&#xff0c;\u5728\u60a8\u7684\u9879\u76ee\u6587\u4ef6\u5939\u4e0b&#xff0c;\u901a\u8fc7\u547d\u4ee4\u884c\u542f\u52a8&#xff1a; jupyter lab<\/p>\n<\/li>\n<\/ul>\n<h5>5.3.3 GPU\u52a0\u901f\u914d\u7f6e\u6307\u5357&#xff1a;\u4e3a\u6a21\u578b\u63d2\u4e0a\u7fc5\u8180<\/h5>\n<p>CPU vs. GPU CPU&#xff08;\u4e2d\u592e\u5904\u7406\u5668&#xff09;\u548cGPU&#xff08;\u56fe\u5f62\u5904\u7406\u5668&#xff09;\u7684\u8bbe\u8ba1\u76ee\u6807\u4e0d\u540c\u3002<\/p>\n<ul>\n<li>CPU&#xff1a;\u50cf\u4e00\u4f4d\u535a\u5b66\u7684\u6559\u6388&#xff0c;\u62e5\u6709\u5c11\u6570\u51e0\u4e2a\u5f3a\u5927\u800c\u590d\u6742\u7684\u8ba1\u7b97\u6838\u5fc3&#xff0c;\u64c5\u957f\u5904\u7406\u903b\u8f91\u590d\u6742\u3001\u9700\u8981\u6309\u987a\u5e8f\u6267\u884c\u7684\u4efb\u52a1\u3002<\/li>\n<li>GPU&#xff1a;\u50cf\u4e00\u4e2a\u5c0f\u5b66\u751f\u519b\u56e2&#xff0c;\u62e5\u6709\u6210\u767e\u4e0a\u5343\u4e2a\u7b80\u5355\u4f46\u9ad8\u6548\u7684\u8ba1\u7b97\u6838\u5fc3\u3002\u5b83\u4e0d\u64c5\u957f\u590d\u6742\u903b\u8f91&#xff0c;\u4f46\u6781\u5176\u64c5\u957f\u6267\u884c\u5927\u89c4\u6a21\u3001\u53ef\u5e76\u884c\u7684\u7b80\u5355\u8ba1\u7b97\u4efb\u52a1\u3002 \u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u6838\u5fc3\u8fd0\u7b97\u2014\u2014\u5927\u89c4\u6a21\u7684\u77e9\u9635\u4e58\u6cd5\u548c\u52a0\u6cd5\u2014\u2014\u6b63\u662fGPU\u7684\u7528\u6b66\u4e4b\u5730\u3002\u4f7f\u7528GPU\u8fdb\u884c\u8bad\u7ec3&#xff0c;\u901a\u5e38\u80fd\u6bd4\u4f7f\u7528CPU\u5e26\u6765\u6570\u5341\u500d\u751a\u81f3\u4e0a\u767e\u500d\u7684\u901f\u5ea6\u63d0\u5347\u3002<\/li>\n<\/ul>\n<p>NVIDIA\u9a71\u52a8\u4e0eCUDA \u8981\u4f7f\u7528GPU\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u9700\u8981\u4e00\u5757NVIDIA\u663e\u5361&#xff0c;\u5e76\u6b63\u786e\u5b89\u88c5\u4e24\u6837\u4e1c\u897f&#xff1a;<\/p>\n<li>NVIDIA\u9a71\u52a8&#xff1a;\u786e\u4fdd\u663e\u5361\u9a71\u52a8\u7a0b\u5e8f\u66f4\u65b0\u5230\u6700\u65b0\u7248\u672c&#xff0c;\u53ef\u4ee5\u4eceNVIDIA\u5b98\u7f51\u4e0b\u8f7d\u3002<\/li>\n<li>CUDA Toolkit&#xff1a;CUDA\u662fNVIDIA\u63a8\u51fa\u7684\u5e76\u884c\u8ba1\u7b97\u5e73\u53f0\u548c\u7f16\u7a0b\u6a21\u578b\u3002\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u901a\u8fc7CUDA\u6765\u8c03\u7528GPU\u7684\u8ba1\u7b97\u80fd\u529b\u3002\u9700\u8981\u6839\u636e\u6240\u4f7f\u7528\u7684TensorFlow\u6216PyTorch\u7248\u672c\u7684\u8981\u6c42&#xff0c;\u4eceNVIDIA\u5b98\u7f51\u4e0b\u8f7d\u5e76\u5b89\u88c5\u5bf9\u5e94\u7248\u672c\u7684CUDA Toolkit\u3002<\/li>\n<p>\u9a8c\u8bc1\u5b89\u88c5 \u5b89\u88c5\u5b8c\u6846\u67b6\u540e&#xff0c;\u53ef\u4ee5\u7528\u4ee5\u4e0b\u4ee3\u7801\u7247\u6bb5\u6765\u9a8c\u8bc1GPU\u662f\u5426\u88ab\u6210\u529f\u8bc6\u522b&#xff1a;<\/p>\n<ul>\n<li>TensorFlow: import tensorflow as tf<br \/>\nprint(&#034;Num GPUs Available: &#034;, len(tf.config.list_physical_devices(&#039;GPU&#039;)))\n <\/li>\n<li>PyTorch: import torch<br \/>\nprint(f&#034;Is CUDA available: {torch.cuda.is_available()}&#034;)<br \/>\nif torch.cuda.is_available():<br \/>\n    print(f&#034;Device name: {torch.cuda.get_device_name(0)}&#034;)\n <\/li>\n<\/ul>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u4ece\u7406\u8bba\u5230\u5b9e\u8df5\u7684\u5173\u952e\u4e00\u8dc3\u3002\u6211\u4eec\u6df1\u5165\u63a2\u7d22\u4e86\u5f53\u4eca\u6700\u4e3b\u6d41\u7684\u4e24\u5927\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u2014\u2014TensorFlow\u4e0ePyTorch\u3002\u6211\u4eec\u4e0d\u4ec5\u5b66\u4e60\u4e86\u5b83\u4eec\u9ad8\u7ea7API\u7684\u4fbf\u6377\u7528\u6cd5&#xff0c;\u66f4\u7406\u89e3\u4e86\u5176\u80cc\u540e\u5173\u4e8e\u5f20\u91cf\u3001\u81ea\u52a8\u6c42\u5bfc\u548c\u6a21\u578b\u5c01\u88c5\u7684\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66\u3002\u901a\u8fc7\u5e76\u6392\u6bd4\u8f83Keras\u7684\u201c\u4e09\u90e8\u66f2\u201d\u548cPyTorch\u7684\u201c\u4e94\u6b65\u6cd5\u201d&#xff0c;\u8bfb\u8005\u5e94\u80fd\u4f53\u4f1a\u5230\u4e24\u79cd\u6846\u67b6\u5728\u6613\u7528\u6027\u4e0e\u7075\u6d3b\u6027\u4e4b\u95f4\u7684\u4e0d\u540c\u53d6\u820d\u3002<\/p>\n<p>\u6700\u540e&#xff0c;\u6211\u4eec\u8be6\u7ec6\u4ecb\u7ecd\u4e86\u5982\u4f55\u4f7f\u7528Conda\u3001Jupyter\u548cNVIDIA\u5de5\u5177\u94fe&#xff0c;\u6765\u642d\u5efa\u4e00\u4e2a\u4e13\u4e1a\u3001\u9ad8\u6548\u7684\u6df1\u5ea6\u5b66\u4e60\u5f00\u53d1\u73af\u5883\u3002\u8fd9\u5957\u201c\u70bc\u4e39\u7089\u201d\u5c06\u662f\u60a8\u672a\u6765\u63a2\u7d22\u66f4\u5e7f\u9614\u6df1\u5ea6\u5b66\u4e60\u4e16\u754c\u7684\u575a\u5b9e\u57fa\u7840\u3002<\/p>\n<p>\u624b\u63e1\u795e\u5175&#xff0c;\u7089\u706b\u5df2\u65fa\u3002\u73b0\u5728&#xff0c;\u60a8\u5df2\u7ecf\u5177\u5907\u4e86\u5c06\u4efb\u4f55\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u4ed8\u8bf8\u5b9e\u8df5\u7684\u80fd\u529b\u3002\u4ece\u4e0b\u4e00\u7ae0\u5f00\u59cb&#xff0c;\u6211\u4eec\u5c06\u8fd0\u7528\u8fd9\u4e9b\u5de5\u5177&#xff0c;\u53bb\u63a2\u7d22\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u4e2d\u90a3\u4e9b\u66f4\u4e13\u95e8\u3001\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u67b6\u6784\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u516d\u7ae0&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u70bc\u4e39\u672f\u201d \u2014\u2014 \u8bad\u7ec3\u4e0e\u4f18\u5316<\/h3>\n<p>\u4ece\u201c\u80fd\u8dd1\u201d\u5230\u201c\u8dd1\u5f97\u597d\u201d\u7684\u827a\u672f<\/p>\n<p>\u5728\u4e0a\u4e00\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5df2\u7ecf\u638c\u63e1\u4e86\u4f7f\u7528\u73b0\u4ee3\u6846\u67b6\u6784\u5efa\u795e\u7ecf\u7f51\u7edc\u5e76\u4f7f\u5176\u8fd0\u884c\u8d77\u6765\u7684\u201c\u795e\u5175\u5229\u5668\u201d\u3002\u7136\u800c&#xff0c;\u8ba9\u6a21\u578b\u201c\u80fd\u8dd1\u201d\u4ec5\u4ec5\u662f\u4e07\u91cc\u957f\u5f81\u7684\u7b2c\u4e00\u6b65\u3002\u4e00\u4e2a\u672a\u7ecf\u7cbe\u5fc3\u8c03\u6821\u7684\u795e\u7ecf\u7f51\u7edc&#xff0c;\u5f80\u5f80\u8868\u73b0\u5e73\u5e73&#xff0c;\u751a\u81f3\u65e0\u6cd5\u6536\u655b\u3002\u4ece\u201c\u80fd\u8dd1\u201d\u5230\u201c\u8dd1\u5f97\u597d\u201d&#xff0c;\u8fd9\u4e2d\u95f4\u7684\u9e3f\u6c9f&#xff0c;\u9700\u8981\u4e00\u95e8\u7cbe\u6df1\u7684\u6280\u827a\u6765\u8de8\u8d8a\u2014\u2014\u6211\u4eec\u79f0\u4e4b\u4e3a\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u70bc\u4e39\u672f\u201d\u3002<\/p>\n<p>\u5c06\u6a21\u578b\u8bad\u7ec3\u6bd4\u4f5c\u4e00\u573a\u53e4\u4ee3\u65b9\u58eb\u7684\u70bc\u4e39&#xff0c;\u5e76\u975e\u6545\u5f04\u7384\u865a\u3002\u56e0\u4e3a\u8fd9\u4e2a\u8fc7\u7a0b\u5145\u6ee1\u4e86\u79d1\u5b66\u7684\u4e25\u8c28\u4e0e\u827a\u672f\u7684\u76f4\u89c9\u3002\u6211\u4eec\u9700\u8981\u9762\u5bf9\u4e24\u5927\u6838\u5fc3\u6311\u6218&#xff1a;<\/p>\n<li>\u4f18\u5316&#xff08;Optimization&#xff09;&#xff1a;\u6211\u4eec\u7684\u635f\u5931\u51fd\u6570&#xff0c;\u662f\u4e00\u4e2a\u5750\u843d\u5728\u4ebf\u4e07\u7ef4\u5ea6\u53c2\u6570\u7a7a\u95f4\u4e2d\u7684\u3001\u6781\u5176\u5d0e\u5c96\u590d\u6742\u7684\u201c\u5c71\u8109\u201d\u3002\u5982\u4f55\u627e\u5230\u4e00\u6761\u6700\u9ad8\u6548\u3001\u6700\u7a33\u5065\u7684\u8def\u5f84&#xff0c;\u4ece\u968f\u673a\u7684\u8d77\u70b9\u51fa\u53d1&#xff0c;\u987a\u5229\u62b5\u8fbe\u5c71\u8109\u7684\u6700\u4f4e\u8c37&#xff08;\u5168\u5c40\u6700\u4f18\u89e3\u6216\u4e00\u4e2a\u8db3\u591f\u597d\u7684\u5c40\u90e8\u6700\u4f18\u89e3&#xff09;&#xff1f;\u8fd9\u4fbf\u662f\u4f18\u5316\u7684\u827a\u672f\u3002<\/li>\n<li>\u6cdb\u5316&#xff08;Generalization&#xff09;&#xff1a;\u5373\u4fbf\u6211\u4eec\u5728\u8bad\u7ec3\u8fd9\u5ea7\u201c\u5c71\u8109\u201d\u4e0a\u627e\u5230\u4e86\u6700\u4f4e\u70b9&#xff0c;\u53c8\u5982\u4f55\u4fdd\u8bc1\u6211\u4eec\u627e\u5230\u7684\u4e0d\u662f\u4e00\u4e2a\u53ea\u9002\u7528\u4e8e\u8fd9\u5ea7\u5c71\u7684\u201c\u6295\u673a\u53d6\u5de7\u201d\u7684\u6377\u5f84&#xff0c;\u800c\u662f\u4e00\u6761\u80fd\u591f\u9002\u7528\u4e8e\u4e16\u754c\u4e0a\u6240\u6709\u7c7b\u4f3c\u5c71\u8109\u7684\u201c\u666e\u9002\u6cd5\u5219\u201d&#xff1f;\u6362\u8a00\u4e4b&#xff0c;\u5982\u4f55\u8ba9\u6a21\u578b\u5728\u672a\u66fe\u89c1\u8fc7\u7684\u201c\u6d4b\u8bd5\u6570\u636e\u201d\u4e0a\u4f9d\u7136\u8868\u73b0\u51fa\u8272&#xff1f;\u8fd9\u4fbf\u662f\u6cdb\u5316\u7684\u667a\u6167\u3002<\/li>\n<p>\u672c\u7ae0\u7684\u4f7f\u547d&#xff0c;\u5c31\u662f\u4e3a\u8bfb\u8005\u7cfb\u7edf\u5730\u4f20\u6388\u5e94\u5bf9\u8fd9\u4e24\u5927\u6311\u6218\u7684\u4e00\u7cfb\u5217\u5173\u952e\u6280\u672f\u4e0e\u7b56\u7565\u3002\u6211\u4eec\u5c06\u4e00\u540c\u63a2\u7d22\u5404\u79cd\u5148\u8fdb\u7684\u4f18\u5316\u5668&#xff0c;\u5b83\u4eec\u5982\u540c\u6027\u80fd\u5404\u5f02\u7684\u201c\u4ea4\u901a\u5de5\u5177\u201d&#xff0c;\u5e2e\u52a9\u6211\u4eec\u5728\u635f\u5931\u51fd\u6570\u7684\u590d\u6742\u5730\u5f62\u4e2d\u9ad8\u6548\u7a7f\u884c\u3002\u6211\u4eec\u5c06\u5b66\u4e60\u591a\u79cd\u6b63\u5219\u5316\u6280\u672f&#xff0c;\u5b83\u4eec\u662f\u9632\u6b62\u6a21\u578b\u201c\u8d70\u706b\u5165\u9b54\u201d\u3001\u9677\u5165\u201c\u8fc7\u62df\u5408\u201d\u7684\u5f3a\u5927\u201c\u6212\u5f8b\u201d\u3002\u6211\u4eec\u8fd8\u4f1a\u63ed\u793a\u6279\u5f52\u4e00\u5316\u3001\u8d85\u53c2\u6570\u8c03\u4f18\u548c\u6743\u91cd\u521d\u59cb\u5316\u7b49\u4e00\u7cfb\u5217\u201c\u79d8\u6cd5\u201d&#xff0c;\u5b83\u4eec\u80fd\u6781\u5927\u5730\u52a0\u901f\u8bad\u7ec3\u8fdb\u7a0b&#xff0c;\u5e76\u63d0\u5347\u6a21\u578b\u7684\u6700\u7ec8\u6027\u80fd\u3002<\/p>\n<p>\u638c\u63e1\u4e86\u8fd9\u4e9b\u201c\u70bc\u4e39\u672f\u201d&#xff0c;\u8bfb\u8005\u5c06\u4e0d\u518d\u662f\u4e00\u4e2a\u53ea\u4f1a\u6309\u90e8\u5c31\u73ed\u7684\u201c\u5b66\u5f92\u201d&#xff0c;\u800c\u5c06\u6210\u957f\u4e3a\u4e00\u4f4d\u80fd\u591f\u6d1e\u6089\u8bad\u7ec3\u52a8\u6001\u3001\u8bca\u65ad\u6a21\u578b\u95ee\u9898\u3001\u5e76\u5bf9\u75c7\u4e0b\u836f\u7684\u201c\u70bc\u4e39\u5927\u5e08\u201d\u3002\u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u4ece\u9009\u62e9\u4e0b\u5c71\u7684\u4ea4\u901a\u5de5\u5177\u5f00\u59cb&#xff0c;\u8e0f\u4e0a\u8fd9\u6bb5\u5145\u6ee1\u6311\u6218\u4e0e\u667a\u6167\u7684\u65c5\u7a0b\u3002<\/p>\n<h4>6.1 \u4f18\u5316\u5668\u8be6\u89e3&#xff1a;\u5bfb\u627e\u6700\u4f18\u89e3\u7684\u201c\u4ea4\u901a\u5de5\u5177\u201d<\/h4>\n<p>\u4f18\u5316\u5668\u662f\u6df1\u5ea6\u5b66\u4e60\u8bad\u7ec3\u8fc7\u7a0b\u7684\u5fc3\u810f\u3002\u5b83\u7684\u804c\u8d23&#xff0c;\u662f\u6839\u636e\u53cd\u5411\u4f20\u64ad\u8ba1\u7b97\u51fa\u7684\u68af\u5ea6\u4fe1\u606f&#xff0c;\u6765\u6307\u5bfc\u6a21\u578b\u53c2\u6570\u5982\u4f55\u66f4\u65b0&#xff0c;\u4ee5\u671f\u6700\u5c0f\u5316\u635f\u5931\u51fd\u6570\u3002\u9009\u62e9\u4e0d\u540c\u7684\u4f18\u5316\u5668&#xff0c;\u5c31\u5982\u540c\u5728\u63a2\u7d22\u635f\u5931\u51fd\u6570\u8fd9\u7247\u5e7f\u88a4\u800c\u5d0e\u5c96\u7684\u201c\u5c71\u8109\u201d\u65f6&#xff0c;\u9009\u62e9\u4e86\u4e0d\u540c\u7684\u4ea4\u901a\u5de5\u5177\u3002<\/p>\n<h5>6.1.1 \u68af\u5ea6\u4e0b\u964d\u7684\u6311\u6218&#xff1a;\u5728\u5d0e\u5c96\u5c71\u8def\u4e2d\u524d\u884c<\/h5>\n<p>\u635f\u5931\u51fd\u6570\u7684\u201c\u5730\u5f62\u201d \u6211\u4eec\u9700\u8981\u5728\u8111\u6d77\u4e2d\u5efa\u7acb\u4e00\u5e45\u753b\u9762&#xff1a;\u4e00\u4e2a\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u635f\u5931\u51fd\u6570&#xff0c;\u5176\u201c\u5730\u5f62\u201d\u8fdc\u6bd4\u6211\u4eec\u60f3\u8c61\u7684\u8981\u590d\u6742\u3002\u5b83\u4e0d\u662f\u4e00\u4e2a\u5e73\u6ed1\u7684\u7897&#xff0c;\u800c\u662f\u4e00\u4e2a\u5750\u843d\u5728\u6781\u9ad8\u7ef4\u5ea6\u7a7a\u95f4\u4e2d\u7684\u3001\u5d0e\u5c96\u65e0\u6bd4\u7684\u5c71\u8109\u3002\u8fd9\u7247\u5c71\u8109\u4e2d&#xff0c;\u904d\u5e03\u7740\u72ed\u957f\u800c\u9661\u5ced\u7684\u5ce1\u8c37&#xff0c;\u68af\u5ea6\u5728\u5ce1\u8c37\u4e24\u4fa7\u65b9\u5411\u5267\u70c8\u53d8\u5316&#xff0c;\u4f46\u5728\u5ce1\u8c37\u5ef6\u4f38\u65b9\u5411\u4e0a\u5374\u5f88\u5e73\u7f13&#xff1b;\u8fd8\u5b58\u5728\u7740\u978d\u70b9&#xff08;Saddle Points&#xff09;&#xff0c;\u5b83\u5728\u67d0\u4e2a\u7ef4\u5ea6\u4e0a\u662f\u6700\u5c0f\u503c&#xff0c;\u4f46\u5728\u53e6\u4e00\u7ef4\u5ea6\u4e0a\u5374\u662f\u6700\u5927\u503c&#xff0c;\u68af\u5ea6\u5728\u6b64\u5904\u4e3a\u96f6&#xff0c;\u6781\u6613\u56f0\u4f4f\u4f18\u5316\u5668&#xff1b;\u6b64\u5916&#xff0c;\u8fd8\u6709\u65e0\u6570\u7684\u5c40\u90e8\u6700\u5c0f\u503c&#xff08;Local Minima&#xff09;\u3002<\/p>\n<p>\u6734\u7d20\u68af\u5ea6\u4e0b\u964d&#xff08;SGD&#xff09;\u7684\u56f0\u5883 \u6211\u4eec\u6700\u57fa\u7840\u7684\u4f18\u5316\u7b97\u6cd5\u2014\u2014\u968f\u673a\u68af\u5ea6\u4e0b\u964d&#xff08;SGD&#xff09;&#xff0c;\u5176\u7b56\u7565\u975e\u5e38\u6734\u7d20&#xff1a;\u53ea\u770b\u811a\u4e0b&#xff0c;\u54ea\u8fb9\u6700\u9661\u5c31\u5f80\u54ea\u8fb9\u8d70\u4e00\u6b65\u3002\u8fd9\u79cd\u7b56\u7565\u5728\u590d\u6742\u5730\u5f62\u4e2d\u4f1a\u9047\u5230\u8bf8\u591a\u9ebb\u70e6&#xff1a;<\/p>\n<ul>\n<li>\u632f\u8361&#xff1a;\u5728\u72ed\u957f\u7684\u5ce1\u8c37\u4e2d&#xff0c;SGD\u4f1a\u5728\u5ce1\u8c37\u4e24\u4fa7\u7684\u5ced\u58c1\u95f4\u6765\u56de\u632f\u8361&#xff0c;\u5bfc\u81f4\u6536\u655b\u901f\u5ea6\u6781\u6162\u3002<\/li>\n<li>\u88ab\u56f0&#xff1a;\u5728\u978d\u70b9\u6216\u5e73\u7f13\u7684\u5c40\u90e8\u6700\u5c0f\u503c\u533a\u57df&#xff0c;\u68af\u5ea6\u53d8\u5f97\u975e\u5e38\u5c0f\u751a\u81f3\u4e3a\u96f6&#xff0c;SGD\u4f1a\u8ba4\u4e3a\u5df2\u7ecf\u5230\u8fbe\u4e86\u8c37\u5e95&#xff0c;\u4ece\u800c\u8fc7\u65e9\u5730\u505c\u6b62\u66f4\u65b0\u3002<\/li>\n<\/ul>\n<p>\u4e3a\u4e86\u514b\u670d\u8fd9\u4e9b\u56f0\u5883&#xff0c;\u4e00\u7cfb\u5217\u66f4\u667a\u80fd\u3001\u66f4\u5f3a\u5927\u7684\u4f18\u5316\u5668\u88ab\u63d0\u4e86\u51fa\u6765\u3002<\/p>\n<h5>6.1.2 \u52a8\u91cf&#xff08;Momentum&#xff09;&#xff1a;\u51b2\u8fc7\u978d\u70b9\u7684\u201c\u60ef\u6027\u201d<\/h5>\n<p>\u7269\u7406\u7c7b\u6bd4 Momentum\u4f18\u5316\u5668\u7684\u7075\u611f&#xff0c;\u6e90\u4e8e\u4e00\u4e2a\u7b80\u5355\u7684\u7269\u7406\u7c7b\u6bd4\u3002\u60f3\u8c61\u4e00\u4e2a\u6709\u8d28\u91cf\u7684\u94c1\u7403\u4ece\u5c71\u4e0a\u6eda\u4e0b&#xff0c;\u5b83\u7684\u8fd0\u52a8\u8f68\u8ff9\u4e0d\u4ec5\u53d6\u51b3\u4e8e\u5f53\u524d\u4f4d\u7f6e\u7684\u5761\u5ea6&#xff08;\u68af\u5ea6&#xff09;&#xff0c;\u66f4\u53d7\u5230\u5176\u81ea\u8eab**\u60ef\u6027&#xff08;\u52a8\u91cf&#xff09;**\u7684\u5f71\u54cd\u3002\u5373\u4f7f\u5b83\u6eda\u5230\u4e00\u4e2a\u5e73\u5766\u7684\u533a\u57df&#xff08;\u68af\u5ea6\u4e3a\u96f6&#xff09;&#xff0c;\u7531\u4e8e\u60ef\u6027\u7684\u5b58\u5728&#xff0c;\u5b83\u4f9d\u7136\u4f1a\u7ee7\u7eed\u524d\u884c&#xff0c;\u4ece\u800c\u6709\u53ef\u80fd\u51b2\u8fc7\u978d\u70b9\u6216\u6d45\u7684\u5c40\u90e8\u6700\u5c0f\u503c\u3002<\/p>\n<p>\u5de5\u4f5c\u539f\u7406 Momentum\u5728SGD\u7684\u57fa\u7840\u4e0a&#xff0c;\u5f15\u5165\u4e86\u4e00\u4e2a\u201c\u52a8\u91cf\u201d\u9879 v&#xff0c;\u5b83\u662f\u8fc7\u53bb\u6240\u6709\u68af\u5ea6\u7684\u4e00\u4e2a\u6307\u6570\u52a0\u6743\u79fb\u52a8\u5e73\u5747\u3002\u53c2\u6570\u7684\u66f4\u65b0\u65b9\u5411&#xff0c;\u4e0d\u518d\u4ec5\u4ec5\u662f\u5f53\u524d\u7684\u68af\u5ea6 g&#xff0c;\u800c\u662f\u52a8\u91cf v&#xff1a; v &#061; \u03b2 * v &#043; (1-\u03b2) * g (\u66f4\u65b0\u52a8\u91cf&#xff0c;\u03b2\u901a\u5e38\u53d60.9) w &#061; w &#8211; learning_rate * v (\u7528\u52a8\u91cf\u6765\u66f4\u65b0\u6743\u91cd)<\/p>\n<p>\u8fd9\u79cd\u673a\u5236\u5e26\u6765\u4e86\u4e24\u4e2a\u597d\u5904&#xff1a;<\/p>\n<li>\u52a0\u901f\u6536\u655b&#xff1a;\u5728\u68af\u5ea6\u65b9\u5411\u57fa\u672c\u4e00\u81f4\u7684\u533a\u57df&#xff08;\u5982\u5ce1\u8c37\u7684\u5ef6\u4f38\u65b9\u5411&#xff09;&#xff0c;\u52a8\u91cf\u4f1a\u6301\u7eed\u7d2f\u79ef&#xff0c;\u4f7f\u5f97\u66f4\u65b0\u6b65\u4f10\u8d8a\u6765\u8d8a\u5927&#xff0c;\u4ece\u800c\u52a0\u901f\u524d\u8fdb\u3002<\/li>\n<li>\u6291\u5236\u632f\u8361&#xff1a;\u5728\u68af\u5ea6\u65b9\u5411\u6765\u56de\u53d8\u5316\u7684\u533a\u57df&#xff08;\u5982\u5ce1\u8c37\u7684\u4e24\u4fa7&#xff09;&#xff0c;\u52a8\u91cf\u9879\u4e2d\u7684\u6b63\u8d1f\u68af\u5ea6\u4f1a\u76f8\u4e92\u62b5\u6d88&#xff0c;\u4ece\u800c\u6291\u5236\u4e86\u632f\u8361&#xff0c;\u4f7f\u4f18\u5316\u8def\u5f84\u66f4\u5e73\u6ed1\u3002<\/li>\n<h5>6.1.3 \u81ea\u9002\u5e94\u5b66\u4e60\u7387&#xff1a;\u4e3a\u6bcf\u4e2a\u53c2\u6570\u5b9a\u5236\u201c\u6b65\u957f\u201d<\/h5>\n<p>\u53e6\u4e00\u5927\u7c7b\u4f18\u5316\u601d\u60f3&#xff0c;\u662f\u8ba9\u5b66\u4e60\u7387\u80fd\u591f\u81ea\u9002\u5e94&#xff08;Adaptive&#xff09;\u3002\u5176\u6838\u5fc3\u6d1e\u89c1\u662f&#xff1a;\u5bf9\u4e8e\u6a21\u578b\u4e2d\u4e0d\u540c\u7684\u53c2\u6570&#xff0c;\u6211\u4eec\u6216\u8bb8\u5e94\u8be5\u4f7f\u7528\u4e0d\u540c\u7684\u5b66\u4e60\u7387\u3002<\/p>\n<p>Adagrad (Adaptive Gradient Algorithm) \u6838\u5fc3\u601d\u60f3&#xff1a;Adagrad\u8ba4\u4e3a&#xff0c;\u90a3\u4e9b\u5728\u8bad\u7ec3\u4e2d\u66f4\u65b0\u4e0d\u9891\u7e41\u7684\u53c2\u6570&#xff08;\u901a\u5e38\u5bf9\u5e94\u7a00\u758f\u7279\u5f81&#xff0c;\u5176\u68af\u5ea6\u7d2f\u79ef\u8f83\u5c0f&#xff09;&#xff0c;\u53ef\u80fd\u8574\u542b\u7740\u91cd\u8981\u7684\u4fe1\u606f&#xff0c;\u6211\u4eec\u5e94\u8be5\u7ed9\u5b83\u4eec\u4e00\u4e2a\u8f83\u5927\u7684\u5b66\u4e60\u7387\u4ee5\u9f13\u52b1\u5176\u66f4\u65b0&#xff1b;\u800c\u90a3\u4e9b\u66f4\u65b0\u9891\u7e41\u7684\u53c2\u6570&#xff08;\u68af\u5ea6\u7d2f\u79ef\u8f83\u5927&#xff09;&#xff0c;\u5219\u5e94\u8be5\u7ed9\u5b83\u4eec\u4e00\u4e2a\u8f83\u5c0f\u7684\u5b66\u4e60\u7387\u4ee5\u6c42\u7a33\u5b9a\u3002\u5b83\u901a\u8fc7\u7d2f\u79ef\u6bcf\u4e2a\u53c2\u6570\u68af\u5ea6\u7684\u5e73\u65b9\u548c\u6765\u5b9e\u73b0\u8fd9\u4e00\u70b9\u3002<\/p>\n<p>\u4f18\u70b9\u4e0e\u7f3a\u70b9&#xff1a;Adagrad\u5728\u5904\u7406\u7a00\u758f\u6570\u636e&#xff08;\u5982NLP\u4e2d\u7684\u8bcd\u5d4c\u5165&#xff09;\u65f6\u8868\u73b0\u51fa\u8272\u3002\u4f46\u5b83\u6709\u4e00\u4e2a\u81f4\u547d\u7f3a\u9677&#xff1a;\u7531\u4e8e\u5206\u6bcd\u4e2d\u7684\u68af\u5ea6\u5e73\u65b9\u548c\u662f\u5355\u8c03\u9012\u589e\u7684&#xff0c;\u5b66\u4e60\u7387\u4f1a\u968f\u7740\u8bad\u7ec3\u7684\u8fdb\u884c\u800c\u6301\u7eed\u3001\u4e0d\u53ef\u9006\u5730\u4e0b\u964d&#xff0c;\u6700\u7ec8\u53ef\u80fd\u53d8\u5f97\u8fc7\u5c0f&#xff0c;\u5bfc\u81f4\u8bad\u7ec3\u5728\u8fd8\u672a\u5145\u5206\u6536\u655b\u65f6\u5c31\u63d0\u524d\u505c\u6b62\u4e86\u3002<\/p>\n<p>RMSprop (Root Mean Square Propagation) \u5bf9Adagrad\u7684\u6539\u8fdb&#xff1a;RMSprop\u662fGeoff Hinton\u63d0\u51fa\u7684\u4e00\u79cd\u5bf9Adagrad\u7684\u6539\u8fdb\u7b97\u6cd5&#xff0c;\u65e8\u5728\u89e3\u51b3\u5176\u5b66\u4e60\u7387\u8fc7\u65e9\u8870\u51cf\u7684\u95ee\u9898\u3002\u5b83\u4e0d\u518d\u662f\u7b80\u5355\u5730\u7d2f\u79ef\u6240\u6709\u5386\u53f2\u68af\u5ea6&#xff0c;\u800c\u662f\u4f7f\u7528\u68af\u5ea6\u7684\u6307\u6570\u52a0\u6743\u79fb\u52a8\u5e73\u5747\u3002\u8fd9\u6837&#xff0c;\u53ea\u6709\u8fd1\u671f\u68af\u5ea6\u7684\u4fe1\u606f\u4f1a\u88ab\u91cd\u70b9\u8003\u8651&#xff0c;\u4f7f\u5f97\u5b66\u4e60\u7387\u53ef\u4ee5\u6839\u636e\u6700\u8fd1\u7684\u68af\u5ea6\u60c5\u51b5\u52a8\u6001\u8c03\u6574&#xff0c;\u800c\u4e0d\u4f1a\u5355\u8c03\u9012\u51cf\u3002<\/p>\n<h5>6.1.4 Adam (Adaptive Moment Estimation)&#xff1a;\u52a8\u91cf\u4e0e\u81ea\u9002\u5e94\u7684\u96c6\u5927\u6210\u8005<\/h5>\n<p>\u5dc5\u5cf0\u4e4b\u4f5c Adam\u4f18\u5316\u5668&#xff0c;\u662f\u76ee\u524d\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u6700\u6d41\u884c\u3001\u6700\u6210\u529f\u7684\u4f18\u5316\u5668\u4e4b\u4e00\u3002\u5b83\u7684\u540d\u5b57\u5df2\u7ecf\u63ed\u793a\u4e86\u5176\u672c\u8d28&#xff1a;\u81ea\u9002\u5e94\u77e9\u4f30\u8ba1\u3002Adam\u5de7\u5999\u5730\u5c06\u6211\u4eec\u524d\u9762\u8ba8\u8bba\u7684\u4e24\u5927\u4e3b\u6d41\u601d\u60f3\u2014\u2014Momentum\u7684\u60ef\u6027\u601d\u60f3\u548cRMSprop\u7684\u81ea\u9002\u5e94\u5b66\u4e60\u7387\u601d\u60f3\u2014\u2014\u5b8c\u7f8e\u5730\u7ed3\u5408\u5728\u4e86\u4e00\u8d77\u3002<\/p>\n<p>\u5b83\u540c\u65f6\u7ef4\u62a4\u4e86\u4e24\u4e2a\u6307\u6570\u52a0\u6743\u79fb\u52a8\u5e73\u5747&#xff1a;<\/p>\n<li>\u4e00\u9636\u77e9\u4f30\u8ba1&#xff08;\u68af\u5ea6\u7684\u5e73\u5747\u503c&#xff09;&#xff1a;\u8fd9\u90e8\u5206\u4e0eMomentum\u975e\u5e38\u76f8\u4f3c&#xff0c;\u8d1f\u8d23\u63a7\u5236\u66f4\u65b0\u7684\u65b9\u5411\u548c\u60ef\u6027\u3002<\/li>\n<li>\u4e8c\u9636\u77e9\u4f30\u8ba1&#xff08;\u68af\u5ea6\u5e73\u65b9\u7684\u5e73\u5747\u503c&#xff09;&#xff1a;\u8fd9\u90e8\u5206\u4e0eRMSprop\u975e\u5e38\u76f8\u4f3c&#xff0c;\u8d1f\u8d23\u63a7\u5236\u6bcf\u4e2a\u53c2\u6570\u81ea\u9002\u5e94\u7684\u5b66\u4e60\u7387\u3002<\/li>\n<p>\u4e3a\u4f55\u6210\u4e3a\u201c\u9ed8\u8ba4\u9009\u9879\u201d \u901a\u8fc7\u8fd9\u79cd\u53cc\u91cd\u673a\u5236&#xff0c;Adam\u51e0\u4e4e\u5728\u6240\u6709\u7c7b\u578b\u7684\u6df1\u5ea6\u5b66\u4e60\u4efb\u52a1\u548c\u6a21\u578b\u67b6\u6784\u4e2d&#xff0c;\u90fd\u8868\u73b0\u51fa\u4e86\u6781\u5176\u51fa\u8272\u548c\u7a33\u5065\u7684\u6027\u80fd\u3002\u5b83\u901a\u5e38\u80fd\u591f\u5feb\u901f\u6536\u655b&#xff0c;\u5e76\u4e14\u5bf9\u8d85\u53c2\u6570\u7684\u9009\u62e9&#xff08;\u5982\u5b66\u4e60\u7387&#xff09;\u76f8\u5bf9\u4e0d\u90a3\u4e48\u654f\u611f\u3002\u56e0\u6b64&#xff0c;\u5728\u4e0d\u786e\u5b9a\u4f7f\u7528\u54ea\u79cd\u4f18\u5316\u5668\u65f6&#xff0c;Adam\u901a\u5e38\u662f\u90a3\u4e2a\u6700\u5b89\u5168\u3001\u6700\u9ad8\u6548\u7684\u9ed8\u8ba4\u9009\u9879\u3002<\/p>\n<hr \/>\n<h4>6.2 \u6b63\u5219\u5316&#xff1a;\u9632\u6b62\u6a21\u578b\u201c\u8d70\u706b\u5165\u9b54\u201d\u7684\u201c\u6212\u5f8b\u201d<\/h4>\n<p>\u5982\u679c\u6211\u4eec\u53ea\u4e13\u6ce8\u4e8e\u4f18\u5316&#xff0c;\u53ef\u80fd\u4f1a\u8bad\u7ec3\u51fa\u4e00\u4e2a\u5728\u8bad\u7ec3\u96c6\u4e0a\u5f97\u5206\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u6a21\u578b\u3002\u4f46\u8fd9\u5f80\u5f80\u662f\u4e00\u79cd\u201c\u5047\u8c61\u201d&#xff0c;\u662f\u6a21\u578b\u201c\u8d70\u706b\u5165\u9b54\u201d\u7684\u5f00\u59cb\u3002\u8fd9\u4e2a\u9b54&#xff0c;\u5c31\u662f\u8fc7\u62df\u5408&#xff08;Overfitting&#xff09;\u3002\u6b63\u5219\u5316&#xff0c;\u5c31\u662f\u6211\u4eec\u4e3a\u6a21\u578b\u8bbe\u5b9a\u7684\u4e00\u7cfb\u5217\u201c\u6212\u5f8b\u201d&#xff0c;\u9632\u6b62\u5b83\u8bef\u5165\u6b67\u9014\u3002<\/p>\n<h5>6.2.1 \u8fc7\u62df\u5408&#xff08;Overfitting&#xff09;&#xff1a;\u667a\u6167\u7684\u201c\u8bc5\u5492\u201d<\/h5>\n<p>\u73b0\u8c61\u4e0e\u672c\u8d28 \u8fc7\u62df\u5408\u7684\u5178\u578b\u73b0\u8c61\u662f&#xff1a;\u6a21\u578b\u5728\u8bad\u7ec3\u96c6&#xff08;Training Set&#xff09;\u4e0a\u8868\u73b0\u6781\u4f73&#xff08;\u635f\u5931\u5f88\u4f4e&#xff0c;\u7cbe\u5ea6\u5f88\u9ad8&#xff09;&#xff0c;\u4f46\u5728\u4ece\u672a\u89c1\u8fc7\u7684\u6d4b\u8bd5\u96c6&#xff08;Test Set&#xff09;\u6216\u9a8c\u8bc1\u96c6&#xff08;Validation Set&#xff09;\u4e0a\u5374\u8868\u73b0\u7cdf\u7cd5\u3002 \u5176\u672c\u8d28\u662f&#xff0c;\u4e00\u4e2a\u8fc7\u4e8e\u5f3a\u5927\u7684\u6a21\u578b&#xff08;\u53c2\u6570\u8fc7\u591a&#xff09;&#xff0c;\u5728\u6709\u9650\u7684\u8bad\u7ec3\u6570\u636e\u4e0a&#xff0c;\u4e0d\u4ec5\u5b66\u5230\u4e86\u6570\u636e\u4e2d\u666e\u9002\u7684\u3001\u6f5c\u5728\u7684\u89c4\u5f8b&#xff0c;\u66f4\u5b66\u5230\u4e86\u6570\u636e\u4e2d\u72ec\u6709\u7684\u566a\u58f0\u548c\u5076\u7136\u7684\u5de7\u5408\u3002\u5b83\u5b8c\u7f8e\u5730\u201c\u8bb0\u5fc6\u201d\u4e86\u8bad\u7ec3\u6837\u672c&#xff0c;\u5374\u5931\u53bb\u4e86**\u6cdb\u5316&#xff08;Generalize&#xff09;**\u5230\u65b0\u6837\u672c\u7684\u80fd\u529b\u3002<\/p>\n<p>\u5965\u5361\u59c6\u5243\u5200\u539f\u5219 \u6240\u6709\u6b63\u5219\u5316\u6280\u672f&#xff0c;\u5176\u80cc\u540e\u7684\u54f2\u5b66\u601d\u60f3\u90fd\u53ef\u4ee5\u8ffd\u6eaf\u5230\u8457\u540d\u7684\u5965\u5361\u59c6\u5243\u5200\u539f\u5219&#xff1a;\u201c\u5982\u65e0\u5fc5\u8981&#xff0c;\u52ff\u589e\u5b9e\u4f53\u201d&#xff08;Entities should not be multiplied without necessity&#xff09;\u3002\u5728\u673a\u5668\u5b66\u4e60\u4e2d&#xff0c;\u8fd9\u610f\u5473\u7740&#xff1a;\u5982\u679c\u6709\u591a\u4e2a\u6a21\u578b\u90fd\u80fd\u5f88\u597d\u5730\u89e3\u91ca\u6570\u636e&#xff0c;\u6211\u4eec\u5e94\u8be5\u9009\u62e9\u90a3\u4e2a\u6700\u7b80\u5355\u7684\u6a21\u578b\u3002\u56e0\u4e3a\u7b80\u5355\u7684\u6a21\u578b\u66f4\u53ef\u80fd\u6293\u4f4f\u95ee\u9898\u7684\u672c\u8d28&#xff0c;\u800c\u975e\u8868\u9762\u7684\u566a\u58f0\u3002\u6b63\u5219\u5316&#xff0c;\u5c31\u662f\u901a\u8fc7\u5404\u79cd\u624b\u6bb5\u5bf9\u6a21\u578b\u7684\u590d\u6742\u5ea6\u65bd\u52a0\u201c\u60e9\u7f5a\u201d\u6216\u201c\u9650\u5236\u201d&#xff0c;\u5f15\u5bfc\u5b83\u53bb\u5bfb\u627e\u90a3\u4e2a\u66f4\u7b80\u5355\u7684\u89e3\u3002<\/p>\n<h5>6.2.2 L1\/L2\u6b63\u5219\u5316&#xff1a;\u4e3a\u53c2\u6570\u6234\u4e0a\u201c\u7d27\u7b8d\u5492\u201d<\/h5>\n<p>L1\u548cL2\u6b63\u5219\u5316\u662f\u6700\u53e4\u8001\u3001\u6700\u7ecf\u5178\u7684\u6b63\u5219\u5316\u65b9\u6cd5\u3002\u5b83\u4eec\u901a\u8fc7\u5728\u539f\u59cb\u7684\u635f\u5931\u51fd\u6570\u4e0a&#xff0c;\u589e\u52a0\u4e00\u4e2a\u5173\u4e8e\u6a21\u578b\u6743\u91cd\u7684\u201c\u60e9\u7f5a\u9879\u201d\u6765\u5b9e\u73b0\u3002<\/p>\n<p>L2\u6b63\u5219\u5316&#xff08;\u6743\u91cd\u8870\u51cf&#xff09; \u539f\u7406&#xff1a;\u5728\u635f\u5931\u51fd\u6570\u540e&#xff0c;\u52a0\u4e0a\u4e00\u4e2a\u4e0e\u6240\u6709\u6743\u91cd\u53c2\u6570\u5e73\u65b9\u548c\u6210\u6b63\u6bd4\u7684\u60e9\u7f5a\u9879&#xff1a;Loss_new &#061; Loss_original &#043; \u03bb * \u03a3(w\u00b2) \u3002\u5176\u4e2d \u03bb \u662f\u6b63\u5219\u5316\u5f3a\u5ea6\u7684\u8d85\u53c2\u6570\u3002 \u6548\u679c&#xff1a;\u8fd9\u4e2a\u60e9\u7f5a\u9879\u4f1a\u4fc3\u4f7f\u4f18\u5316\u5668\u5728\u51cf\u5c0f\u539f\u59cb\u635f\u5931\u7684\u540c\u65f6&#xff0c;\u4e5f\u5c3d\u91cf\u51cf\u5c0f\u6743\u91cd\u7684\u5927\u5c0f\u3002\u5b83\u503e\u5411\u4e8e\u8ba9\u6a21\u578b\u7684\u6743\u91cd\u503c\u53d8\u5f97\u66f4\u5c0f\u3001\u66f4\u5206\u6563&#xff0c;\u4f7f\u5f97\u6a21\u578b\u7684\u51b3\u7b56\u8fb9\u754c\u66f4\u5e73\u6ed1&#xff0c;\u4ece\u800c\u964d\u4f4e\u4e86\u5bf9\u8bad\u7ec3\u6570\u636e\u4e2d\u4e2a\u522b\u566a\u58f0\u70b9\u7684\u654f\u611f\u5ea6\u3002\u5728\u5f88\u591a\u6846\u67b6\u4e2d&#xff0c;\u5b83\u4e5f\u88ab\u79f0\u4e3a\u6743\u91cd\u8870\u51cf&#xff08;Weight Decay&#xff09;\u3002<\/p>\n<p>L1\u6b63\u5219\u5316 \u539f\u7406&#xff1a;\u5728\u635f\u5931\u51fd\u6570\u540e&#xff0c;\u52a0\u4e0a\u4e00\u4e2a\u4e0e\u6240\u6709\u6743\u91cd\u53c2\u6570\u7edd\u5bf9\u503c\u4e4b\u548c\u6210\u6b63\u6bd4\u7684\u60e9\u7f5a\u9879&#xff1a;Loss_new &#061; Loss_original &#043; \u03bb * \u03a3|w|\u3002 \u6548\u679c&#xff1a;L1\u6b63\u5219\u5316\u6709\u4e00\u4e2a\u975e\u5e38\u72ec\u7279\u7684\u7279\u6027\u3002\u7531\u4e8e\u7edd\u5bf9\u503c\u51fd\u6570\u5728\u539f\u70b9\u5904\u7684\u5c16\u70b9&#xff0c;\u5b83\u4f1a\u5f3a\u70c8\u5730\u9a71\u4f7f\u90a3\u4e9b\u5bf9\u6a21\u578b\u8d21\u732e\u4e0d\u5927\u7684\u53c2\u6570\u6743\u91cd&#xff0c;\u7cbe\u786e\u5730\u53d8\u4e3a\u96f6\u3002\u8fd9\u76f8\u5f53\u4e8e\u81ea\u52a8\u5730\u8fdb\u884c\u4e86\u4e00\u6b21\u7279\u5f81\u9009\u62e9&#xff0c;\u5254\u9664\u4e86\u65e0\u6548\u7279\u5f81&#xff0c;\u4f7f\u5f97\u6700\u7ec8\u7684\u6a21\u578b\u53d8\u5f97\u7a00\u758f&#xff08;Sparse&#xff09;\u3002<\/p>\n<h5>6.2.3 Dropout&#xff1a;\u968f\u673a\u201c\u5931\u5fc6\u201d\u7684\u667a\u6167<\/h5>\n<p>Dropout\u662f\u6df1\u5ea6\u5b66\u4e60\u65f6\u4ee3\u63d0\u51fa\u7684\u4e00\u79cd\u6781\u5176\u5f3a\u5927\u4e14\u7b80\u5355\u6709\u6548\u7684\u6b63\u5219\u5316\u6280\u672f\u3002<\/p>\n<p>\u5de5\u4f5c\u539f\u7406 \u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d&#xff0c;\u5bf9\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u67d0\u4e00\u5c42&#xff0c;Dropout\u4f1a\u4ee5\u4e00\u4e2a\u9884\u8bbe\u7684\u6982\u7387 p&#xff08;\u4f8b\u5982 p&#061;0.5&#xff09;&#xff0c;\u968f\u673a\u5730\u5c06\u5176\u4e2d\u7684\u4e00\u90e8\u5206\u795e\u7ecf\u5143\u7684\u8f93\u51fa\u6682\u65f6\u7f6e\u4e3a\u96f6\u3002\u8fd9\u610f\u5473\u7740&#xff0c;\u5728\u6bcf\u4e00\u6b21\u524d\u5411\u4f20\u64ad\u4e2d&#xff0c;\u7f51\u7edc\u90fd\u597d\u50cf\u88ab\u201c\u9609\u5272\u201d\u6210\u4e86\u4e00\u4e2a\u4e0d\u540c\u7684\u3001\u66f4\u5c0f\u7684\u201c\u5b50\u7f51\u7edc\u201d\u3002\u800c\u5728\u6d4b\u8bd5\u8fc7\u7a0b\u4e2d&#xff0c;\u5219\u4f1a\u4f7f\u7528\u5b8c\u6574\u7684\u3001\u672a\u7ecf\u4e22\u5f03\u7684\u7f51\u7edc\u3002<\/p>\n<p>\u6548\u679c\u4e0e\u9690\u55bb Dropout\u7684\u5f3a\u5927\u6548\u679c&#xff0c;\u53ef\u4ee5\u4ece\u4e24\u4e2a\u89d2\u5ea6\u6765\u7406\u89e3&#xff1a;<\/p>\n<li>\u6253\u7834\u534f\u540c\u9002\u5e94&#xff1a;\u5b83\u5f3a\u5236\u4e00\u4e2a\u795e\u7ecf\u5143\u4e0d\u80fd\u8fc7\u5ea6\u4f9d\u8d56\u4e8e\u5176\u4ed6\u67d0\u51e0\u4e2a\u7279\u5b9a\u7684\u795e\u7ecf\u5143\u3002\u56e0\u4e3a\u5b83\u201c\u8eab\u8fb9\u201d\u7684\u4efb\u4f55\u4e00\u4e2a\u201c\u540c\u4e8b\u201d&#xff0c;\u90fd\u968f\u65f6\u53ef\u80fd\u201c\u7f62\u5de5\u201d\u3002\u8fd9\u8feb\u4f7f\u7f51\u7edc\u53bb\u5b66\u4e60\u66f4\u52a0\u9c81\u68d2\u3001\u66f4\u52a0\u72ec\u7acb\u7684\u7279\u5f81\u8868\u793a\u3002<\/li>\n<li>\u96c6\u6210\u5b66\u4e60\u7684\u8fd1\u4f3c&#xff1a;\u4ece\u5b8f\u89c2\u4e0a\u770b&#xff0c;\u6bcf\u4e00\u6b21\u8fed\u4ee3\u4f7f\u7528\u4e00\u4e2a\u4e0d\u540c\u7684\u5b50\u7f51\u7edc\u8fdb\u884c\u8bad\u7ec3&#xff0c;\u6574\u4e2aDropout\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c31\u597d\u50cf\u5728\u540c\u65f6\u8bad\u7ec3\u6210\u5343\u4e0a\u4e07\u4e2a\u5171\u4eab\u6743\u91cd\u7684\u3001\u4e0d\u540c\u7ed3\u6784\u7684\u7f51\u7edc\u3002\u800c\u5728\u6d4b\u8bd5\u65f6\u4f7f\u7528\u5b8c\u6574\u7684\u7f51\u7edc&#xff0c;\u5219\u8fd1\u4f3c\u4e8e\u5c06\u8fd9\u4e9b\u6d77\u91cf\u7684\u5b50\u7f51\u7edc\u8fdb\u884c**\u6a21\u578b\u96c6\u6210&#xff08;Ensemble&#xff09;**\u6765\u505a\u9884\u6d4b\u3002\u8fd9\u662f\u4e00\u79cd\u6781\u5176\u5ec9\u4ef7\u800c\u9ad8\u6548\u7684Bagging\u96c6\u6210\u8fd1\u4f3c&#xff0c;\u80fd\u663e\u8457\u63d0\u5347\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<h5>6.2.4 \u65e9\u505c&#xff08;Early Stopping&#xff09;&#xff1a;\u89c1\u597d\u5c31\u6536\u7684\u201c\u7985\u5b9a\u201d<\/h5>\n<p>\u539f\u7406 \u65e9\u505c\u662f\u4e00\u79cd\u5145\u6ee1\u5b9e\u8df5\u667a\u6167\u7684\u6b63\u5219\u5316\u7b56\u7565\u3002\u5b83\u7684\u505a\u6cd5\u662f&#xff1a;\u5728\u8bad\u7ec3\u6a21\u578b\u7684\u540c\u65f6&#xff0c;\u6211\u4eec\u5e76\u4e0d\u53ea\u5173\u5fc3\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u4e0a\u7684\u635f\u5931&#xff0c;\u800c\u662f\u6301\u7eed\u5730\u76d1\u63a7\u5b83\u5728\u4e00\u4e2a\u72ec\u7acb\u7684\u9a8c\u8bc1\u96c6\u4e0a\u7684\u6027\u80fd&#xff08;\u4f8b\u5982&#xff0c;\u9a8c\u8bc1\u96c6\u4e0a\u7684\u635f\u5931\u6216\u51c6\u786e\u7387&#xff09;\u3002 \u901a\u5e38&#xff0c;\u8bad\u7ec3\u521d\u671f\u7684\u51e0\u4e2aepoch&#xff0c;\u8bad\u7ec3\u96c6\u548c\u9a8c\u8bc1\u96c6\u7684\u635f\u5931\u90fd\u4f1a\u4e0b\u964d\u3002\u4f46\u5230\u67d0\u4e2a\u65f6\u95f4\u70b9\u4e4b\u540e&#xff0c;\u8bad\u7ec3\u96c6\u7684\u635f\u5931\u4ecd\u5728\u7ee7\u7eed\u4e0b\u964d&#xff0c;\u800c\u9a8c\u8bc1\u96c6\u7684\u635f\u5931\u5374\u5f00\u59cb\u505c\u6b62\u4e0b\u964d&#xff0c;\u751a\u81f3\u4e0d\u964d\u53cd\u5347\u3002\u8fd9\u4e2a\u201c\u62d0\u70b9\u201d&#xff0c;\u6b63\u662f\u6a21\u578b\u5f00\u59cb\u8fc7\u62df\u5408\u7684\u4fe1\u53f7\u3002\u65e9\u505c\u7b56\u7565&#xff0c;\u5c31\u662f\u5728\u63a2\u6d4b\u5230\u8fd9\u4e2a\u62d0\u70b9\u540e&#xff0c;\u7acb\u5373\u505c\u6b62\u8bad\u7ec3&#xff0c;\u5e76\u5c06\u6a21\u578b\u6062\u590d\u5230\u5728\u9a8c\u8bc1\u96c6\u4e0a\u6027\u80fd\u6700\u597d\u7684\u90a3\u4e2a\u72b6\u6001\u3002<\/p>\n<p>\u5b9e\u8df5\u667a\u6167 \u65e9\u505c\u7684\u7f8e\u5999\u4e4b\u5904\u5728\u4e8e\u5176\u7b80\u5355\u548c\u9ad8\u6548\u3002\u5b83\u4e0d\u9700\u8981\u4fee\u6539\u635f\u5931\u51fd\u6570&#xff0c;\u4e5f\u4e0d\u9700\u8981\u5f15\u5165\u989d\u5916\u7684\u8d85\u53c2\u6570&#xff08;\u9664\u4e86\u8010\u5fc3\u7b49\u5f85\u7684epoch\u6570&#xff09;\u3002\u5b83\u76f4\u51fb\u95ee\u9898\u7684\u6838\u5fc3&#xff0c;\u5728\u6a21\u578b\u5373\u5c06\u201c\u8d70\u706b\u5165\u9b54\u201d\u7684\u90a3\u4e00\u523b&#xff0c;\u679c\u65ad\u5730\u8ba9\u5b83\u201c\u6536\u529f\u7985\u5b9a\u201d&#xff0c;\u4ece\u800c\u4ee5\u6700\u5c0f\u7684\u4ee3\u4ef7\u83b7\u5f97\u4e86\u6781\u4f73\u7684\u6b63\u5219\u5316\u6548\u679c\u3002<\/p>\n<hr \/>\n<h4>6.3 \u6279\u5f52\u4e00\u5316&#xff08;Batch Normalization&#xff09;&#xff1a;\u91cd\u5851\u201c\u5730\u5f62\u201d\u7684\u201c\u98ce\u6c34\u672f\u201d<\/h4>\n<p>\u6279\u5f52\u4e00\u5316&#xff08;BN&#xff09;\u662f\u6df1\u5ea6\u5b66\u4e60\u8bad\u7ec3\u6280\u672f\u4e2d\u7684\u4e00\u9879\u91cc\u7a0b\u7891\u5f0f\u7684\u53d1\u660e\u3002\u5b83\u4e0d\u4ec5\u80fd\u6781\u5927\u5730\u52a0\u901f\u6a21\u578b\u7684\u6536\u655b\u901f\u5ea6&#xff0c;\u8fd8\u517c\u5177\u4e00\u5b9a\u7684\u6b63\u5219\u5316\u6548\u679c\u3002\u5b83\u5c31\u50cf\u4e00\u4f4d\u9ad8\u660e\u7684\u201c\u98ce\u6c34\u5927\u5e08\u201d&#xff0c;\u901a\u8fc7\u8c03\u6574\u7f51\u7edc\u5185\u90e8\u7684\u6570\u636e\u201c\u98ce\u6c34\u201d&#xff0c;\u8ba9\u8bad\u7ec3\u8fc7\u7a0b\u53d8\u5f97\u5f02\u5e38\u987a\u7545\u3002<\/p>\n<h5>6.3.1 \u5185\u90e8\u534f\u53d8\u91cf\u504f\u79fb&#xff08;Internal Covariate Shift&#xff09;<\/h5>\n<p>\u95ee\u9898\u63cf\u8ff0\u00a0\u60f3\u8c61\u4e00\u4e0b\u6df1\u5ea6\u7f51\u7edc\u7684\u8bad\u7ec3\u8fc7\u7a0b\u3002\u5f53\u7b2c\u4e00\u5c42\u7684\u53c2\u6570\u901a\u8fc7\u68af\u5ea6\u4e0b\u964d\u8fdb\u884c\u66f4\u65b0\u540e&#xff0c;\u5b83\u8f93\u51fa\u7684\u6570\u636e\u7684\u5206\u5e03&#xff08;\u5747\u503c\u3001\u65b9\u5dee\u7b49&#xff09;\u5c31\u53d1\u751f\u4e86\u6539\u53d8\u3002\u5bf9\u4e8e\u7b2c\u4e8c\u5c42\u6765\u8bf4&#xff0c;\u5b83\u4e0a\u4e00\u8f6e\u521a\u5b66\u4f1a\u5982\u4f55\u5904\u7406\u65e7\u7684\u8f93\u5165\u5206\u5e03&#xff0c;\u4e0b\u4e00\u8f6e\u5c31\u8981\u9762\u5bf9\u4e00\u4e2a\u65b0\u7684\u3001\u964c\u751f\u7684\u8f93\u5165\u5206\u5e03\u3002\u8fd9\u79cd\u7f51\u7edc\u5185\u90e8&#xff0c;\u5c42\u4e0e\u5c42\u4e4b\u95f4\u8f93\u5165\u5206\u5e03\u4e0d\u65ad\u53d8\u5316\u7684\u73b0\u8c61&#xff0c;\u5c31\u88ab\u79f0\u4e3a\u5185\u90e8\u534f\u53d8\u91cf\u504f\u79fb\u3002 \u8fd9\u8feb\u4f7f\u7f51\u7edc\u7684\u6bcf\u4e00\u5c42\u90fd\u9700\u8981\u4e0d\u65ad\u5730\u53bb\u9002\u5e94\u5176\u4e0a\u6e38\u5c42\u8f93\u5165\u5206\u5e03\u7684\u53d8\u5316&#xff0c;\u5c31\u50cf\u5728\u6d41\u6c99\u4e0a\u76d6\u697c\u4e00\u6837&#xff0c;\u6781\u5927\u5730\u62d6\u6162\u4e86\u6574\u4f53\u7684\u8bad\u7ec3\u6548\u7387\u3002<\/p>\n<h5>6.3.2 BN\u7684\u5de5\u4f5c\u539f\u7406<\/h5>\n<p>\u5f3a\u5236\u201c\u6c34\u571f\u201d\u7a33\u5b9a BN\u7684\u89e3\u51b3\u65b9\u6848\u7b80\u5355\u800c\u7c97\u66b4&#xff1a;\u5728\u6bcf\u4e00\u5c42\u7684\u7ebf\u6027\u53d8\u6362\u4e4b\u540e\u3001\u6fc0\u6d3b\u51fd\u6570\u4e4b\u524d&#xff0c;\u5b83\u5f3a\u884c\u5c06\u8f93\u5165\u7684\u6570\u636e&#xff08;\u5728\u4e00\u4e2amini-batch\u7684\u8303\u56f4\u5185&#xff09;\u8fdb\u884c\u4e00\u6b21\u6807\u51c6\u5316\u5904\u7406&#xff0c;\u4f7f\u5176\u5747\u503c\u6062\u590d\u4e3a0&#xff0c;\u65b9\u5dee\u6062\u590d\u4e3a1\u3002 \u8fd9\u6837\u4e00\u6765&#xff0c;\u65e0\u8bba\u4e0a\u6e38\u5c42\u7684\u53c2\u6570\u5982\u4f55\u53d8\u5316&#xff0c;\u6d41\u5230\u4e0b\u6e38\u5c42\u7684\u6570\u636e&#xff0c;\u5176\u5206\u5e03\u90fd\u88ab\u7a33\u5b9a\u5728\u4e86\u8fd9\u4e2a\u6807\u51c6\u72b6\u6001&#xff0c;\u6781\u5927\u5730\u7a33\u5b9a\u4e86\u201c\u5730\u57fa\u201d\u3002<\/p>\n<p>\u53ef\u5b66\u4e60\u7684\u201c\u5fae\u8c03\u201d \u4f46\u65b0\u7684\u95ee\u9898\u6765\u4e86&#xff1a;\u5f3a\u884c\u5c06\u6570\u636e\u90fd\u626d\u66f2\u6210\u6807\u51c6\u6b63\u6001\u5206\u5e03&#xff0c;\u4f1a\u4e0d\u4f1a\u7834\u574f\u6389\u7f51\u7edc\u597d\u4e0d\u5bb9\u6613\u5b66\u5230\u7684\u6709\u7528\u7279\u5f81\u5462&#xff1f;\u6bd4\u5982&#xff0c;\u67d0\u4e2a\u7279\u5f81\u7684\u5206\u5e03\u8303\u56f4\u672c\u8eab\u5c31\u8574\u542b\u7740\u91cd\u8981\u4fe1\u606f\u3002 \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898&#xff0c;BN\u5f15\u5165\u4e86\u4e24\u4e2a\u53ef\u4ee5\u50cf\u6743\u91cd\u4e00\u6837\u901a\u8fc7\u5b66\u4e60\u6765\u66f4\u65b0\u7684\u53c2\u6570&#xff1a;\u7f29\u653e\u56e0\u5b50 \u03b3 (gamma) \u548c \u5e73\u79fb\u56e0\u5b50 \u03b2 (beta)\u3002\u5728\u6807\u51c6\u5316\u4e4b\u540e&#xff0c;BN\u4f1a\u7528\u8fd9\u4e24\u4e2a\u53c2\u6570\u5bf9\u6570\u636e\u518d\u8fdb\u884c\u4e00\u6b21\u7ebf\u6027\u7684\u7f29\u653e\u548c\u5e73\u79fb&#xff1a;y &#061; \u03b3 * x_normalized &#043; \u03b2\u3002 \u8fd9\u7ed9\u4e86\u7f51\u7edc\u4e00\u4e2a\u201c\u53cd\u6094\u201d\u7684\u673a\u4f1a\u3002\u5728\u6700\u574f\u7684\u60c5\u51b5\u4e0b&#xff0c;\u7f51\u7edc\u53ef\u4ee5\u901a\u8fc7\u5b66\u4e60\u8ba9 \u03b3 \u7b49\u4e8e\u539f\u59cb\u6570\u636e\u7684\u6807\u51c6\u5dee&#xff0c;\u03b2 \u7b49\u4e8e\u539f\u59cb\u6570\u636e\u7684\u5747\u503c&#xff0c;\u4ece\u800c\u5b8c\u5168\u62b5\u6d88\u6389\u6807\u51c6\u5316\u64cd\u4f5c&#xff0c;\u6062\u590d\u51fa\u539f\u59cb\u7684\u7279\u5f81\u5206\u5e03\u3002\u8fd9\u786e\u4fdd\u4e86BN\u5728\u5e26\u6765\u597d\u5904\u7684\u540c\u65f6&#xff0c;\u4e0d\u4f1a\u524a\u5f31\u6a21\u578b\u7684\u8868\u8fbe\u80fd\u529b\u3002<\/p>\n<h5>6.3.3 BN\u5e26\u6765\u7684\u597d\u5904<\/h5>\n<p>\u52a0\u901f\u6536\u655b&#xff1a;\u8fd9\u662fBN\u6700\u4e3b\u8981\u7684\u597d\u5904\u3002\u7531\u4e8e\u5185\u90e8\u6570\u636e\u5206\u5e03\u88ab\u7a33\u5b9a&#xff0c;\u4f18\u5316\u8fc7\u7a0b\u53d8\u5f97\u66f4\u52a0\u5e73\u6ed1&#xff0c;\u4f7f\u5f97\u6211\u4eec\u53ef\u4ee5\u653e\u5fc3\u5730\u4f7f\u7528\u66f4\u9ad8\u7684\u5b66\u4e60\u7387&#xff0c;\u4ece\u800c\u6781\u5927\u5730\u52a0\u901f\u6a21\u578b\u7684\u6536\u655b\u3002<\/p>\n<p>\u7f13\u89e3\u68af\u5ea6\u6d88\u5931&#xff1a;\u901a\u8fc7\u5c06\u6570\u636e\u62c9\u56de\u5230\u6fc0\u6d3b\u51fd\u6570&#xff08;\u5982Sigmoid&#xff09;\u7684\u7ebf\u6027\u533a&#xff08;\u4e2d\u5fc3\u533a&#xff09;&#xff0c;BN\u6709\u6548\u5730\u7f13\u89e3\u4e86\u68af\u5ea6\u9971\u548c\u548c\u68af\u5ea6\u6d88\u5931\u7684\u95ee\u9898\u3002<\/p>\n<p>\u81ea\u5e26\u6b63\u5219\u5316\u6548\u679c&#xff1a;\u7531\u4e8eBN\u662f\u57fa\u4e8e\u6bcf\u4e2amini-batch\u7684\u5747\u503c\u548c\u65b9\u5dee\u8fdb\u884c\u8ba1\u7b97\u7684&#xff0c;\u800c\u6bcf\u4e2abatch\u7684\u6570\u636e\u90fd\u7565\u6709\u4e0d\u540c&#xff0c;\u8fd9\u76f8\u5f53\u4e8e\u4e3a\u7f51\u7edc\u7684\u6bcf\u4e00\u5c42\u90fd\u5f15\u5165\u4e86\u8f7b\u5fae\u7684\u968f\u673a\u566a\u58f0\u3002\u8fd9\u79cd\u566a\u58f0&#xff0c;\u8d77\u5230\u4e86\u7c7b\u4f3cDropout\u7684\u6b63\u5219\u5316\u6548\u679c&#xff0c;\u6709\u52a9\u4e8e\u63d0\u5347\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<hr \/>\n<h4>6.4 \u8d85\u53c2\u6570\u8c03\u4f18&#xff1a;\u5bfb\u627e\u6700\u4f73\u201c\u4e39\u65b9\u201d\u7684\u827a\u672f<\/h4>\n<p>\u5982\u679c\u8bf4\u524d\u9762\u4ecb\u7ecd\u7684\u6280\u672f\u662f\u201c\u70bc\u4e39\u201d\u4e2d\u7684\u5177\u4f53\u201c\u624b\u6cd5\u201d&#xff0c;\u90a3\u4e48\u8d85\u53c2\u6570\u8c03\u4f18&#xff0c;\u5c31\u662f\u5bfb\u627e\u6700\u4f73\u201c\u4e39\u65b9\u201d\u7684\u827a\u672f\u3002<\/p>\n<h5>6.4.1 \u8d85\u53c2\u6570 vs. \u53c2\u6570<\/h5>\n<p>\u53c2\u6570&#xff08;Parameters&#xff09;&#xff1a;\u662f\u6a21\u578b\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d&#xff0c;\u901a\u8fc7\u4f18\u5316\u7b97\u6cd5\u81ea\u52a8\u5b66\u4e60\u5f97\u5230\u7684\u503c\u3002\u4f8b\u5982&#xff0c;\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u6743\u91cd\u00a0w\u00a0\u548c\u504f\u7f6e\u00a0b\u3002<\/p>\n<p>\u8d85\u53c2\u6570&#xff08;Hyperparameters&#xff09;&#xff1a;\u662f\u6211\u4eec\u5728\u8bad\u7ec3\u5f00\u59cb\u4e4b\u524d&#xff0c;\u9700\u8981\u624b\u52a8\u8bbe\u5b9a\u7684\u914d\u7f6e\u3002\u4f8b\u5982&#xff0c;\u5b66\u4e60\u7387\u7684\u5927\u5c0f\u3001\u7f51\u7edc\u7684\u5c42\u6570\u3001\u6bcf\u4e2a\u9690\u85cf\u5c42\u7684\u795e\u7ecf\u5143\u6570\u91cf\u3001Dropout\u7684\u4e22\u5f03\u7387\u3001\u6b63\u5219\u5316\u5f3a\u5ea6\u00a0\u03bb\u00a0\u7b49\u3002\u8fd9\u4e9b\u8d85\u53c2\u6570\u5171\u540c\u51b3\u5b9a\u4e86\u6a21\u578b\u7684\u67b6\u6784\u548c\u8bad\u7ec3\u7684\u65b9\u5f0f&#xff0c;\u662f\u6a21\u578b\u7684\u201c\u57fa\u56e0\u201d\u548c\u201c\u57f9\u517b\u65b9\u6848\u201d\u3002<\/p>\n<h5>6.4.2 \u7ecf\u5178\u7684\u8c03\u4f18\u7b56\u7565<\/h5>\n<p>\u7f51\u683c\u641c\u7d22&#xff08;Grid Search&#xff09; \u8fd9\u662f\u4e00\u79cd\u6700\u66b4\u529b\u3001\u6700\u76f4\u63a5\u7684\u5730\u6bef\u5f0f\u641c\u7d22\u65b9\u6cd5\u3002\u9700\u8981\u4e3a\u6bcf\u4e00\u4e2a\u6240\u5173\u5fc3\u7684\u8d85\u53c2\u6570&#xff0c;\u8bbe\u5b9a\u4e00\u4e2a\u5019\u9009\u503c\u7684\u5217\u8868\u3002\u7f51\u683c\u641c\u7d22\u4f1a\u7a77\u5c3d\u8fd9\u4e9b\u5217\u8868\u503c\u7684\u6240\u6709\u53ef\u80fd\u7ec4\u5408&#xff0c;\u4e3a\u6bcf\u4e00\u79cd\u7ec4\u5408\u90fd\u8bad\u7ec3\u4e00\u4e2a\u6a21\u578b&#xff0c;\u5e76\u6700\u7ec8\u9009\u51fa\u5728\u9a8c\u8bc1\u96c6\u4e0a\u8868\u73b0\u6700\u597d\u7684\u90a3\u4e2a\u7ec4\u5408\u3002 \u5b83\u7684\u4f18\u70b9\u662f\u5b8c\u5907&#xff0c;\u53ea\u8981\u5019\u9009\u5217\u8868\u8db3\u591f\u597d&#xff0c;\u7406\u8bba\u4e0a\u80fd\u627e\u5230\u6700\u4f18\u89e3\u3002\u7f3a\u70b9\u662f\u5176\u8ba1\u7b97\u6210\u672c\u968f\u7740\u8d85\u53c2\u6570\u6570\u91cf\u7684\u589e\u52a0\u5448\u6307\u6570\u7ea7\u589e\u957f&#xff0c;\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d\u51e0\u4e4e\u4e0d\u53ef\u884c\u3002<\/p>\n<p>\u968f\u673a\u641c\u7d22&#xff08;Random Search&#xff09; \u968f\u673a\u641c\u7d22\u7684\u505a\u6cd5\u662f&#xff0c;\u4e0d\u518d\u5c1d\u8bd5\u6240\u6709\u7ec4\u5408&#xff0c;\u800c\u662f\u5728\u4e3a\u6bcf\u4e2a\u8d85\u53c2\u6570\u8bbe\u5b9a\u7684\u4e00\u4e2a\u8303\u56f4\u5185&#xff0c;\u968f\u673a\u5730\u91c7\u6837\u6307\u5b9a\u6b21\u6570\u7684\u7ec4\u5408\u6765\u8fdb\u884c\u5b9e\u9a8c\u3002 \u5b9e\u8df5\u548c\u7406\u8bba\u90fd\u8bc1\u660e&#xff0c;\u968f\u673a\u641c\u7d22\u901a\u5e38\u6bd4\u7f51\u683c\u641c\u7d22\u66f4\u9ad8\u6548\u3002\u56e0\u4e3a\u5bf9\u4e8e\u5927\u591a\u6570\u6a21\u578b\u6765\u8bf4&#xff0c;\u771f\u6b63\u5bf9\u6027\u80fd\u6709\u51b3\u5b9a\u6027\u5f71\u54cd\u7684&#xff0c;\u5f80\u5f80\u53ea\u662f\u5c11\u6570\u51e0\u4e2a\u201c\u5173\u952e\u201d\u8d85\u53c2\u6570\u3002\u968f\u673a\u641c\u7d22\u66f4\u6709\u53ef\u80fd\u5728\u8fd9\u4e9b\u5173\u952e\u8d85\u53c2\u6570\u4e0a\u63a2\u7d22\u5230\u66f4\u4f18\u7684\u503c&#xff0c;\u800c\u4e0d\u4f1a\u628a\u5927\u91cf\u7684\u8ba1\u7b97\u8d44\u6e90\u6d6a\u8d39\u5728\u90a3\u4e9b\u4e0d\u91cd\u8981\u7684\u8d85\u53c2\u6570\u7684\u7cbe\u7ec6\u5212\u5206\u4e0a\u3002<\/p>\n<h5>6.4.3 \u66f4\u667a\u80fd\u7684\u8c03\u4f18&#xff1a;\u8d1d\u53f6\u65af\u4f18\u5316<\/h5>\n<p>\u5b66\u4e60\u5982\u4f55\u641c\u7d22\u00a0\u7f51\u683c\u641c\u7d22\u548c\u968f\u673a\u641c\u7d22\u90fd\u662f\u201c\u76f2\u76ee\u201d\u7684&#xff0c;\u6bcf\u4e00\u6b21\u5b9e\u9a8c\u90fd\u662f\u72ec\u7acb\u7684&#xff0c;\u4e0d\u4f1a\u4ece\u8fc7\u53bb\u7684\u5931\u8d25\u6216\u6210\u529f\u4e2d\u5b66\u5230\u4efb\u4f55\u4e1c\u897f\u3002\u8d1d\u53f6\u65af\u4f18\u5316\u5219\u662f\u4e00\u79cd\u66f4\u667a\u80fd\u7684\u7b56\u7565\u3002 \u5b83\u7684\u6838\u5fc3\u601d\u60f3\u662f&#xff1a;\u6839\u636e\u6240\u6709\u5df2\u7ecf\u5b8c\u6210\u7684\u5b9e\u9a8c\u7ed3\u679c&#xff08;\u5373\u201c&#xff08;\u8d85\u53c2\u6570\u7ec4\u5408&#xff0c;\u6a21\u578b\u6027\u80fd&#xff09;\u201d\u8fd9\u4e2a\u6570\u636e\u70b9&#xff09;&#xff0c;\u5efa\u7acb\u4e00\u4e2a\u5173\u4e8e\u201c\u54ea\u4e2a\u8d85\u53c2\u6570\u7ec4\u5408\u53ef\u80fd\u4f1a\u5e26\u6765\u66f4\u597d\u6027\u80fd\u201d\u7684\u6982\u7387\u6a21\u578b&#xff08;\u4ee3\u7406\u6a21\u578b&#xff09;\u3002\u7136\u540e&#xff0c;\u5b83\u5229\u7528\u8fd9\u4e2a\u6a21\u578b&#xff0c;\u53bb\u667a\u80fd\u5730\u3001\u6709\u9009\u62e9\u5730\u6311\u9009\u4e0b\u4e00\u4e2a\u6700\u6709\u5e0c\u671b&#xff08;\u6216\u8005\u8bf4&#xff0c;\u4e0d\u786e\u5b9a\u6027\u6700\u5927\u4e14\u6f5c\u529b\u6700\u9ad8&#xff09;\u7684\u5019\u9009\u70b9\u8fdb\u884c\u5c1d\u8bd5\u3002 \u901a\u8fc7\u8fd9\u79cd\u201c\u8fb9\u5b66\u8fb9\u731c\u201d\u7684\u65b9\u5f0f&#xff0c;\u8d1d\u53f6\u65af\u4f18\u5316\u901a\u5e38\u80fd\u7528\u6bd4\u968f\u673a\u641c\u7d22\u5c11\u5f97\u591a\u7684\u5b9e\u9a8c\u6b21\u6570&#xff0c;\u627e\u5230\u4e00\u4e2a\u66f4\u597d\u6216\u76f8\u5f53\u7684\u89e3&#xff0c;\u662f\u76ee\u524d\u8fdb\u884c\u590d\u6742\u6a21\u578b\u8d85\u53c2\u6570\u8c03\u4f18\u7684\u4e3b\u6d41\u9ad8\u7ea7\u65b9\u6cd5\u3002<\/p>\n<hr \/>\n<h4>6.5 \u6743\u91cd\u521d\u59cb\u5316&#xff1a;\u8d62\u5728\u201c\u8d77\u8dd1\u7ebf\u201d<\/h4>\n<p>\u5728\u201c\u70bc\u4e39\u201d\u7684\u6700\u540e&#xff0c;\u6211\u4eec\u56de\u5230\u8d77\u70b9\u3002\u4e39\u7089\u70b9\u706b\u7684\u90a3\u4e00\u523b&#xff0c;\u7089\u5185\u7269\u8d28\u7684\u521d\u59cb\u72b6\u6001&#xff0c;\u5f80\u5f80\u5bf9\u6700\u7ec8\u6210\u4e39\u7684\u54c1\u8d28\u6709\u51b3\u5b9a\u6027\u5f71\u54cd\u3002\u6743\u91cd\u521d\u59cb\u5316&#xff0c;\u5c31\u662f\u8fd9\u95e8\u201c\u8d62\u5728\u8d77\u8dd1\u7ebf\u201d\u7684\u5b66\u95ee\u3002<\/p>\n<h5>6.5.1 \u521d\u59cb\u5316\u4e3a\u4f55\u91cd\u8981<\/h5>\n<p>\u6253\u7834\u5bf9\u79f0\u6027 \u4e00\u4e2a\u7edd\u5bf9\u4e0d\u80fd\u72af\u7684\u9519\u8bef&#xff0c;\u662f\u5c06\u6240\u6709\u6743\u91cd\u90fd\u521d\u59cb\u5316\u4e3a0\u3002\u5982\u679c\u8fd9\u6837\u505a&#xff0c;\u90a3\u4e48\u5728\u540c\u4e00\u5c42\u7684\u6240\u6709\u795e\u7ecf\u5143&#xff0c;\u5b83\u4eec\u7684\u8f93\u5165\u3001\u8f93\u51fa\u3001\u4ee5\u53ca\u63a5\u6536\u5230\u7684\u68af\u5ea6\u5c06\u6c38\u8fdc\u662f\u5b8c\u5168\u76f8\u540c\u7684\u3002\u5b83\u4eec\u4f1a\u50cf\u88ab\u6346\u7ed1\u5728\u4e00\u8d77\u4e00\u6837&#xff0c;\u5b66\u4e60\u5230\u5b8c\u5168\u76f8\u540c\u7684\u7279\u5f81\u3002\u8fd9\u88ab\u79f0\u4e3a\u5bf9\u79f0\u6027\u95ee\u9898&#xff0c;\u5b83\u4f1a\u4f7f\u5f97\u6df1\u5ea6\u7f51\u7edc\u7684\u591a\u5c42\u7ed3\u6784\u5b8c\u5168\u5931\u53bb\u610f\u4e49\u3002<\/p>\n<p>\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8 \u4e0d\u6070\u5f53\u7684\u968f\u673a\u521d\u59cb\u5316&#xff0c;\u4e5f\u53ef\u80fd\u5e26\u6765\u707e\u96be\u3002\u5982\u679c\u6743\u91cd\u521d\u59cb\u503c\u8fc7\u5c0f&#xff0c;\u4fe1\u53f7\u5728\u524d\u5411\u4f20\u64ad\u4e2d\u4f1a\u9010\u5c42\u8870\u51cf&#xff0c;\u5bfc\u81f4\u8f93\u51fa\u63a5\u8fd1\u4e8e0&#xff1b;\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u4e2d\u4e5f\u4f1a\u9010\u5c42\u6d88\u5931&#xff0c;\u4f7f\u7f51\u7edc\u96be\u4ee5\u8bad\u7ec3\u3002\u53cd\u4e4b&#xff0c;\u5982\u679c\u6743\u91cd\u521d\u59cb\u503c\u8fc7\u5927&#xff0c;\u5219\u53ef\u80fd\u5bfc\u81f4\u4fe1\u53f7\u548c\u68af\u5ea6\u5728\u524d\u5411\u548c\u53cd\u5411\u4f20\u64ad\u4e2d\u88ab\u9010\u5c42\u6307\u6570\u7ea7\u5730\u653e\u5927&#xff0c;\u9020\u6210\u68af\u5ea6\u7206\u70b8&#xff0c;\u4f7f\u8bad\u7ec3\u8fc7\u7a0b\u53d1\u6563\u3002<\/p>\n<h5>6.5.2 \u73b0\u4ee3\u521d\u59cb\u5316\u7b56\u7565<\/h5>\n<p>\u73b0\u4ee3\u7684\u6743\u91cd\u521d\u59cb\u5316\u7b56\u7565&#xff0c;\u5176\u6838\u5fc3\u76ee\u6807\u90fd\u662f\u4e3a\u4e86\u8ba9\u4fe1\u53f7\u80fd\u591f\u5728\u7f51\u7edc\u4e2d\u66f4\u7a33\u5b9a\u5730\u6d41\u52a8&#xff0c;\u5373\u4fdd\u6301\u6bcf\u4e00\u5c42\u8f93\u51fa\u7684\u65b9\u5dee\u5927\u81f4\u7a33\u5b9a\u3002<\/p>\n<p>Xavier\/Glorot\u521d\u59cb\u5316 \u6838\u5fc3\u601d\u60f3&#xff1a;Xavier\u521d\u59cb\u5316\u63a8\u5bfc\u51fa\u7684\u4e00\u4e2a\u5173\u952e\u7ed3\u8bba\u662f&#xff0c;\u4e3a\u4e86\u4fdd\u6301\u524d\u5411\u4f20\u64ad\u548c\u53cd\u5411\u4f20\u64ad\u4e2d\u4fe1\u53f7\u65b9\u5dee\u7684\u7a33\u5b9a&#xff0c;\u6743\u91cd\u7684\u521d\u59cb\u5316\u65b9\u5dee\u5e94\u8be5\u4e0e\u8be5\u5c42\u7684\u8f93\u5165\u8282\u70b9\u6570 n_in \u548c\u8f93\u51fa\u8282\u70b9\u6570 n_out\u90fd\u6709\u5173\u3002\u5b83\u901a\u5e38\u4ece\u4e00\u4e2a\u5747\u503c\u4e3a0&#xff0c;\u65b9\u5dee\u4e3a 2 \/ (n_in &#043; n_out) \u7684\u5747\u5300\u5206\u5e03\u6216\u6b63\u6001\u5206\u5e03\u4e2d\u8fdb\u884c\u91c7\u6837\u3002 \u9002\u7528\u573a\u666f&#xff1a;\u8fd9\u79cd\u521d\u59cb\u5316\u65b9\u6cd5&#xff0c;\u5728\u5176\u63a8\u5bfc\u4e2d\u5047\u8bbe\u4e86\u6fc0\u6d3b\u51fd\u6570\u662f\u7ebf\u6027\u7684&#xff0c;\u56e0\u6b64\u5b83\u5728\u90a3\u4e9b\u5173\u4e8e\u539f\u70b9\u5bf9\u79f0\u7684\u6fc0\u6d3b\u51fd\u6570&#xff08;\u5982Sigmoid\u548cTanh&#xff09;\u4e0a\u8868\u73b0\u5f97\u975e\u5e38\u597d\u3002<\/p>\n<p>He\u521d\u59cb\u5316 \u5bf9Xavier\u7684\u4fee\u6b63&#xff1a;\u5f53\u6fc0\u6d3b\u51fd\u6570\u6362\u6210\u73b0\u4ee3\u7f51\u7edc\u4e2d\u6700\u5e38\u7528\u7684ReLU\u65f6&#xff0c;\u60c5\u51b5\u53d1\u751f\u4e86\u53d8\u5316\u3002ReLU\u4f1a\u5c06\u6240\u6709\u8d1f\u7684\u8f93\u5165\u90fd\u7f6e\u4e3a\u96f6&#xff0c;\u8fd9\u76f8\u5f53\u4e8e\u76f4\u63a5\u201c\u780d\u6389\u201d\u4e86\u4e00\u534a\u7684\u4fe1\u53f7\u3002He\u521d\u59cb\u5316\u654f\u9510\u5730\u8003\u8651\u5230\u4e86\u8fd9\u4e00\u70b9&#xff0c;\u5b83\u5728\u63a8\u5bfc\u4e2d\u4fee\u6b63\u4e86\u65b9\u5dee\u7684\u8ba1\u7b97&#xff0c;\u6700\u7ec8\u5f97\u5230\u7684\u91c7\u6837\u65b9\u5dee\u4e3a 2 \/ n_in\u3002 \u9002\u7528\u573a\u666f&#xff1a;\u7531\u4e8e\u5b83\u4e13\u4e3aReLU\u53ca\u5176\u53d8\u4f53&#xff08;\u5982Leaky ReLU&#xff09;\u8bbe\u8ba1&#xff0c;He\u521d\u59cb\u5316\u662f\u6240\u6709\u4f7f\u7528ReLU\u6fc0\u6d3b\u51fd\u6570\u7684\u73b0\u4ee3\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u6807\u51c6\u6743\u91cd\u521d\u59cb\u5316\u65b9\u6cd5\u3002<\/p>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u5df2\u7ecf\u5b8c\u6574\u5730\u5b66\u4e60\u4e86\u6df1\u5ea6\u5b66\u4e60\u201c\u70bc\u4e39\u672f\u201d\u4e2d\u7684\u4e94\u5927\u6838\u5fc3\u79d8\u7b08\u3002\u4ece\u9009\u62e9\u5408\u9002\u7684\u4f18\u5316\u5668\u6765\u9ad8\u6548\u5bfc\u822a&#xff0c;\u5230\u8fd0\u7528\u6b63\u5219\u5316\u6280\u672f\u4e3a\u6a21\u578b\u8bbe\u5b9a\u6212\u5f8b\u4ee5\u9632\u8fc7\u62df\u5408&#xff1b;\u4ece\u5229\u7528\u6279\u5f52\u4e00\u5316\u6765\u7a33\u5b9a\u5185\u90e8\u73af\u5883\u4ee5\u52a0\u901f\u6536\u655b&#xff0c;\u5230\u63a2\u7d22\u8d85\u53c2\u6570\u8c03\u4f18\u7684\u827a\u672f\u6765\u5bfb\u627e\u6700\u4f73\u4e39\u65b9&#xff1b;\u6700\u540e&#xff0c;\u6211\u4eec\u8fd8\u638c\u63e1\u4e86\u901a\u8fc7\u6743\u91cd\u521d\u59cb\u5316\u6765\u8d62\u5f97\u4e00\u4e2a\u826f\u597d\u5f00\u5c40\u7684\u667a\u6167\u3002<\/p>\n<p>\u8fd9\u4e9b\u6280\u672f&#xff0c;\u5e76\u975e\u76f8\u4e92\u72ec\u7acb&#xff0c;\u800c\u662f\u76f8\u8f85\u76f8\u6210&#xff0c;\u5171\u540c\u6784\u6210\u4e86\u4e00\u4e2a\u6210\u529f\u7684\u6df1\u5ea6\u5b66\u4e60\u5b9e\u8df5\u8005\u6240\u5fc5\u987b\u5177\u5907\u7684\u77e5\u8bc6\u4f53\u7cfb\u3002\u5b83\u4eec\u5c06\u7406\u8bba\u4e0e\u5de5\u7a0b\u3001\u79d1\u5b66\u4e0e\u827a\u672f\u5b8c\u7f8e\u5730\u7ed3\u5408\u5728\u4e00\u8d77\u3002<\/p>\n<p>\u638c\u63e1\u4e86\u8fd9\u4e9b\u201c\u70bc-\u4e39-\u672f\u201d&#xff0c;\u8bfb\u8005\u4fbf\u62e5\u6709\u4e86\u5c06\u4e00\u4e2a\u5e73\u5eb8\u6a21\u578b&#xff0c;\u6253\u78e8\u6210\u4e00\u4e2a\u9ad8\u6027\u80fd\u6a21\u578b\u7684\u5173\u952e\u80fd\u529b\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u7684\u201c\u5185\u529f\u201d\u4e0e\u201c\u62db\u5f0f\u201d\u5747\u5df2\u5927\u6210&#xff0c;\u662f\u65f6\u5019\u53bb\u6311\u6218\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u4e2d\u90a3\u4e9b\u66f4\u4e13\u95e8\u3001\u66f4\u5f3a\u5927\u7684\u201c\u795e\u529f\u201d\u4e86\u3002\u5728\u63a5\u4e0b\u6765\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u5c06\u8fdb\u5165\u8ba1\u7b97\u673a\u89c6\u89c9\u548c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u8fd9\u4e24\u4e2a\u6700\u6fc0\u52a8\u4eba\u5fc3\u7684\u5e94\u7528\u9886\u57df\u3002<\/p>\n<hr \/>\n<h2>\u7b2c\u4e09\u90e8\u5206&#xff1a;\u8fdb\u9636\u7bc7 \u2014\u2014 \u638c\u63e1\u6838\u5fc3\u7f51\u7edc\u67b6\u6784<\/h2>\n<hr \/>\n<h3>\u7b2c\u4e03\u7ae0&#xff1a;\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09; \u2014\u2014 \u6d1e\u6089\u56fe\u50cf\u7684\u5965\u79d8<\/h3>\n<p>\u8d85\u8d8a\u50cf\u7d20 \u2014\u2014 \u8ba9\u673a\u5668\u201c\u770b\u201d\u61c2\u4e16\u754c<\/p>\n<p>\u6b22\u8fce\u6765\u5230\u6df1\u5ea6\u5b66\u4e60\u6700\u4ee4\u4eba\u6fc0\u52a8\u3001\u4e5f\u6700\u7b26\u5408\u6211\u4eec\u4eba\u7c7b\u76f4\u89c9\u7684\u9886\u57df\u2014\u2014\u8ba1\u7b97\u673a\u89c6\u89c9\u3002\u6211\u4eec\u7684\u4e16\u754c&#xff0c;\u662f\u4e00\u4e2a\u7531\u56fe\u50cf\u548c\u89c6\u9891\u6784\u6210\u7684\u89c6\u89c9\u4fe1\u606f\u7684\u6d77\u6d0b\u3002\u8ba9\u673a\u5668\u80fd\u591f\u50cf\u6211\u4eec\u4e00\u6837\u201c\u770b\u201d\u61c2\u8fd9\u4e2a\u4e16\u754c&#xff0c;\u662f\u4eba\u5de5\u667a\u80fd\u957f\u4e45\u4ee5\u6765\u7684\u68a6\u60f3\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u6d6a\u6f6e\u5174\u8d77\u4e4b\u524d&#xff0c;\u8fd9\u4e2a\u68a6\u60f3\u9065\u4e0d\u53ef\u53ca\u3002\u800c\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u7684\u51fa\u73b0&#xff0c;\u5219\u5982\u540c\u4e00\u9053\u5212\u7834\u957f\u591c\u7684\u95ea\u7535&#xff0c;\u5f7b\u5e95\u6539\u53d8\u4e86\u8fd9\u4e00\u5207\u3002<\/p>\n<p>\u5728\u524d\u9762\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u5df2\u7ecf\u719f\u6089\u4e86\u591a\u5c42\u611f\u77e5\u673a&#xff08;MLP&#xff09;\u3002\u5982\u679c\u6211\u4eec\u8bd5\u56fe\u7528MLP\u6765\u5904\u7406\u4e00\u5f20\u56fe\u50cf&#xff0c;\u4f1a\u9047\u5230\u4e24\u4e2a\u96be\u4ee5\u903e\u8d8a\u7684\u969c\u788d\u3002\u9996\u5148&#xff0c;\u6211\u4eec\u9700\u8981\u5c06\u4e8c\u7ef4\u7684\u56fe\u50cf\u77e9\u9635\u5c55\u5e73&#xff08;Flatten&#xff09;\u6210\u4e00\u4e2a\u957f\u957f\u7684\u4e00\u7ef4\u5411\u91cf\u3002\u8fd9\u4e2a\u7c97\u66b4\u7684\u64cd\u4f5c&#xff0c;\u4f1a\u5f7b\u5e95\u4e22\u5931\u56fe\u50cf\u5b9d\u8d35\u7684\u7a7a\u95f4\u7ed3\u6784\u4fe1\u606f\u2014\u2014\u4e00\u4e2a\u50cf\u7d20\u4e0e\u5176\u4e0a\u4e0b\u5de6\u53f3\u90bb\u5c45\u7684\u5173\u7cfb&#xff0c;\u5bf9\u4e8e\u7406\u89e3\u56fe\u50cf\u81f3\u5173\u91cd\u8981\u3002\u5176\u6b21&#xff0c;\u5373\u4f7f\u662f\u4e00\u5f20\u5f88\u5c0f\u7684\u5f69\u8272\u56fe\u7247&#xff0c;\u5c55\u5e73\u540e\u7684\u5411\u91cf\u7ef4\u5ea6\u4e5f\u6781\u5176\u5de8\u5927&#xff0c;\u8fd9\u610f\u5473\u7740MLP\u7684\u8f93\u5165\u5c42\u548c\u7b2c\u4e00\u9690\u85cf\u5c42\u4e4b\u95f4\u5c06\u4ea7\u751f\u5929\u6587\u6570\u5b57\u822c\u7684\u53c2\u6570\u91cf&#xff0c;\u8fd9\u4e0d\u4ec5\u8ba9\u6a21\u578b\u6781\u6613\u8fc7\u62df\u5408&#xff0c;\u4e5f\u4f7f\u5176\u5728\u8ba1\u7b97\u4e0a\u96be\u4ee5\u627f\u53d7\u3002<\/p>\n<p>CNN&#xff0c;\u6b63\u662f\u4e3a\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\u800c\u751f\u7684\u3001\u4e00\u79cd\u4e13\u4e3a\u5904\u7406\u50cf\u56fe\u50cf\u8fd9\u7c7b\u7f51\u683c\u72b6\u6570\u636e\u800c\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\u3002\u5b83\u7684\u8bbe\u8ba1\u4e2d&#xff0c;\u8574\u542b\u7740\u6df1\u523b\u7684\u4eff\u751f\u5b66\u667a\u6167&#xff0c;\u5176\u6838\u5fc3\u601d\u60f3\u6e90\u4e8e\u5bf9\u751f\u7269\u89c6\u89c9\u76ae\u5c42\u5de5\u4f5c\u65b9\u5f0f\u7684\u6a21\u4eff\u3002\u5b83\u4e0d\u518d\u5c06\u56fe\u50cf\u89c6\u4e3a\u4e00\u4e2a\u65e0\u5e8f\u7684\u50cf\u7d20\u96c6\u5408&#xff0c;\u800c\u662f\u901a\u8fc7\u4e00\u79cd\u5de7\u5999\u7684\u65b9\u5f0f&#xff0c;\u9010\u5c42\u5730\u3001\u6709\u7ed3\u6784\u5730\u4ece\u4e2d\u63d0\u53d6\u4fe1\u606f\u2014\u2014\u4ece\u6700\u5e95\u5c42\u7684\u8fb9\u7f18\u3001\u89d2\u70b9&#xff0c;\u5230\u4e2d\u95f4\u5c42\u7684\u773c\u775b\u3001\u9f3b\u5b50\u7b49\u90e8\u4ef6&#xff0c;\u518d\u5230\u6700\u9ad8\u5c42\u7684\u7269\u4f53\u4e43\u81f3\u573a\u666f\u3002<\/p>\n<p>\u5728\u672c\u7ae0\u7684\u63a2\u7d22\u4e2d&#xff0c;\u6211\u4eec\u5c06\u9996\u5148\u6df1\u5165CNN\u8d56\u4ee5\u6210\u529f\u7684\u4e09\u5927\u57fa\u77f3\u601d\u60f3&#xff1a;\u5c40\u90e8\u8fde\u63a5\u3001\u6743\u503c\u5171\u4eab\u4e0e\u6c60\u5316\u3002\u968f\u540e&#xff0c;\u6211\u4eec\u5c06\u50cf\u89e3\u5256\u5b66\u5bb6\u4e00\u6837&#xff0c;\u9010\u4e00\u5256\u6790\u6784\u6210CNN\u7684\u6838\u5fc3\u7ec4\u4ef6\u3002\u63a5\u7740&#xff0c;\u6211\u4eec\u5c06\u7a7f\u8d8a\u65f6\u7a7a&#xff0c;\u56de\u987e\u4eceLeNet-5\u5230ResNet\u7684\u7ecf\u5178CNN\u67b6\u6784\u6ce2\u6f9c\u58ee\u9614\u7684\u6f14\u8fdb\u53f2\u8bd7&#xff0c;\u89c1\u8bc1\u4e00\u4ee3\u4ee3AI\u5148\u9a71\u4eec\u5982\u4f55\u5c06\u7f51\u7edc\u63a8\u5411\u65b0\u7684\u6df1\u5ea6\u4e0e\u9ad8\u5ea6\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u5b66\u4f1a\u5982\u4f55\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u5229\u7528\u8fc1\u79fb\u5b66\u4e60\u6765\u89e3\u51b3\u6211\u4eec\u81ea\u5df1\u7684\u95ee\u9898&#xff0c;\u5e76\u4e00\u7aa5CNN\u5728\u56fe\u50cf\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b\u3001\u56fe\u50cf\u5206\u5272\u7b49\u66f4\u5e7f\u9614\u4e16\u754c\u4e2d\u7684\u5f3a\u5927\u5e94\u7528\u3002<\/p>\n<p>\u51c6\u5907\u597d\u5f00\u542f\u673a\u5668\u7684\u201c\u6167\u773c\u201d\u4e86\u5417&#xff1f;\u8ba9\u6211\u4eec\u4e00\u540c\u8fdb\u5165CNN\u7684\u5947\u5999\u4e16\u754c\u3002<\/p>\n<h4>7.1 CNN\u7684\u6838\u5fc3\u601d\u60f3&#xff1a;\u6e90\u4e8e\u89c6\u89c9\u7684\u667a\u6167<\/h4>\n<p>CNN\u4e4b\u6240\u4ee5\u80fd\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u53d6\u5f97\u5982\u6b64\u5de8\u5927\u7684\u6210\u529f&#xff0c;\u5e76\u975e\u5076\u7136\u3002\u5b83\u7684\u8bbe\u8ba1\u54f2\u5b66&#xff0c;\u4e0e\u6211\u4eec\u751f\u7269\u89c6\u89c9\u7cfb\u7edf\u5904\u7406\u4fe1\u606f\u7684\u65b9\u5f0f\u60ca\u4eba\u5730\u76f8\u4f3c\u3002\u8fd9\u4e09\u5927\u6838\u5fc3\u601d\u60f3\u2014\u2014\u5c40\u90e8\u8fde\u63a5\u3001\u6743\u503c\u5171\u4eab\u548c\u6c60\u5316\u2014\u2014\u5171\u540c\u6784\u6210\u4e86CNN\u9ad8\u6548\u800c\u5f3a\u5927\u7684\u57fa\u7840\u3002<\/p>\n<h5>7.1.1 \u5c40\u90e8\u8fde\u63a5&#xff08;Local Connectivity&#xff09;&#xff1a;\u4e13\u6ce8\u201c\u4e00\u9685\u201d<\/h5>\n<p>\u751f\u7269\u89c6\u89c9\u7684\u542f\u53d1 \u8ba9\u6211\u4eec\u5148\u601d\u8003\u4e00\u4e0b\u6211\u4eec\u81ea\u5df1\u7684\u773c\u775b\u662f\u5982\u4f55\u5de5\u4f5c\u7684\u3002\u8bfa\u8d1d\u5c14\u5956\u5f97\u4e3bHubel\u548cWiesel\u7684\u7814\u7a76\u53d1\u73b0&#xff0c;\u732b\u7684\u89c6\u89c9\u76ae\u5c42\u4e2d&#xff0c;\u6bcf\u4e2a\u795e\u7ecf\u5143\u90fd\u53ea\u5bf9\u89c6\u91ce\u4e2d\u7684\u4e00\u4e2a\u975e\u5e38\u5c0f\u7684\u5c40\u90e8\u533a\u57df\u6709\u53cd\u5e94\u3002\u8fd9\u4e2a\u533a\u57df&#xff0c;\u88ab\u79f0\u4e3a\u795e\u7ecf\u5143\u7684\u611f\u53d7\u91ce&#xff08;Receptive Field&#xff09;\u3002\u6211\u4eec\u770b\u4e16\u754c&#xff0c;\u5e76\u4e0d\u662f\u4e00\u773c\u5c31\u611f\u77e5\u5230\u6240\u6709\u50cf\u7d20&#xff0c;\u800c\u662f\u7531\u65e0\u6570\u4e2a\u8fd9\u6837\u7684\u795e\u7ecf\u5143&#xff0c;\u5404\u81ea\u8d1f\u8d23\u81ea\u5df1\u7684\u4e00\u5c0f\u7247\u201c\u8d23\u4efb\u7530\u201d&#xff0c;\u7136\u540e\u5927\u8111\u518d\u5c06\u8fd9\u4e9b\u5c40\u90e8\u4fe1\u606f\u6574\u5408\u8d77\u6765&#xff0c;\u5f62\u6210\u5b8c\u6574\u7684\u89c6\u89c9\u611f\u77e5\u3002<\/p>\n<p>CNN\u7684\u5b9e\u73b0 CNN\u5b8c\u7f8e\u5730\u501f\u9274\u4e86\u8fd9\u4e00\u673a\u5236\u3002\u5728\u4e00\u4e2a\u5377\u79ef\u5c42\u4e2d&#xff0c;\u5176\u795e\u7ecf\u5143&#xff08;\u6216\u8005\u8bf4&#xff0c;\u8f93\u51fa\u7279\u5f81\u56fe\u4e0a\u7684\u4e00\u4e2a\u5355\u5143&#xff09;\u4e0d\u518d\u4e0e\u524d\u4e00\u5c42\u7684\u6240\u6709\u795e\u7ecf\u5143\u76f8\u8fde\u63a5&#xff08;\u8fd9\u6b63\u662fMLP\u7684\u505a\u6cd5&#xff09;&#xff0c;\u800c\u4ec5\u4ec5\u8fde\u63a5\u5230\u8f93\u5165\u56fe\u50cf&#xff08;\u6216\u524d\u4e00\u5c42\u7279\u5f81\u56fe&#xff09;\u7684\u4e00\u4e2a\u5f88\u5c0f\u7684\u5c40\u90e8\u533a\u57df\u3002 \u8fd9\u4e2a\u8bbe\u8ba1\u80cc\u540e&#xff0c;\u8574\u542b\u7740\u4e00\u4e2a\u975e\u5e38\u7b26\u5408\u76f4\u89c9\u7684\u5047\u8bbe&#xff1a;\u7269\u4f53\u7684\u5c40\u90e8\u7279\u5f81&#xff0c;\u662f\u7531\u5176\u5c40\u90e8\u50cf\u7d20\u51b3\u5b9a\u7684\u3002\u4f8b\u5982&#xff0c;\u8981\u5224\u65ad\u4e00\u4e2a\u5730\u65b9\u662f\u5426\u5b58\u5728\u4e00\u4e2a\u201c\u89d2\u70b9\u201d&#xff0c;\u6211\u4eec\u53ea\u9700\u8981\u89c2\u5bdf\u8fd9\u4e2a\u70b9\u5468\u56f4\u4e00\u5c0f\u5708\u50cf\u7d20\u7684\u6a21\u5f0f\u5373\u53ef&#xff0c;\u800c\u65e0\u9700\u5173\u5fc3\u56fe\u50cf\u8fdc\u5904\u7684\u5185\u5bb9\u3002 \u5c40\u90e8\u8fde\u63a5\u8fd9\u4e00\u601d\u60f3&#xff0c;\u4f7f\u5f97CNN\u7684\u8fde\u63a5\u6570\u548c\u53c2\u6570\u6570\u91cf&#xff0c;\u76f8\u8f83\u4e8e\u5168\u8fde\u63a5\u7684MLP&#xff0c;\u5f97\u5230\u4e86\u6307\u6570\u7ea7\u7684\u51cf\u5c11&#xff0c;\u8fd9\u662fCNN\u80fd\u591f\u5904\u7406\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u7684\u7b2c\u4e00\u4e2a\u5173\u952e\u3002<\/p>\n<h5>7.1.2 \u6743\u503c\u5171\u4eab&#xff08;Weight Sharing&#xff09;&#xff1a;\u4e00\u628a\u201c\u6807\u5c3a\u201d\u91cf\u5929\u4e0b<\/h5>\n<p>\u5c40\u90e8\u8fde\u63a5\u89e3\u51b3\u4e86\u53c2\u6570\u8fc7\u591a\u7684\u95ee\u9898&#xff0c;\u4f46\u5982\u679c\u6bcf\u4e2a\u5c40\u90e8\u533a\u57df\u90fd\u9700\u8981\u4e00\u5957\u72ec\u7acb\u7684\u6743\u91cd\u53bb\u5b66\u4e60&#xff0c;\u53c2\u6570\u91cf\u4f9d\u7136\u5e9e\u5927\u3002\u6743\u503c\u5171\u4eab&#xff0c;\u662fCNN\u7684\u7b2c\u4e8c\u4e2a\u3001\u4e5f\u662f\u66f4\u5177\u9769\u547d\u6027\u7684\u601d\u60f3\u3002<\/p>\n<p>\u56fe\u50cf\u7684\u5e73\u79fb\u4e0d\u53d8\u6027 \u8ba9\u6211\u4eec\u518d\u601d\u8003\u4e00\u4e2a\u89c6\u89c9\u5e38\u8bc6&#xff1a;\u4e00\u4e2a\u7279\u5b9a\u7684\u6a21\u5f0f&#xff0c;\u6bd4\u5982\u4e00\u53ea\u9e1f\u7684\u773c\u775b&#xff0c;\u65e0\u8bba\u5b83\u51fa\u73b0\u5728\u56fe\u50cf\u7684\u5de6\u4e0a\u89d2&#xff0c;\u8fd8\u662f\u53f3\u4e0b\u89d2&#xff0c;\u5b83\u4f5c\u4e3a\u201c\u773c\u775b\u201d\u7684\u8fd9\u4e2a\u672c\u8d28\u7279\u5f81\u662f\u4e0d\u4f1a\u6539\u53d8\u7684\u3002\u6211\u4eec\u5927\u8111\u4e2d\u7684\u201c\u773c\u775b\u63a2\u6d4b\u5668\u201d&#xff0c;\u4e0d\u4f1a\u56e0\u4e3a\u76ee\u6807\u4f4d\u7f6e\u7684\u6539\u53d8\u800c\u9700\u8981\u91cd\u65b0\u5b66\u4e60\u3002\u8fd9\u79cd\u7279\u6027&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u5e73\u79fb\u4e0d\u53d8\u6027&#xff08;Translation Invariance&#xff09;\u3002<\/p>\n<p>\u5377\u79ef\u6838&#xff08;Kernel\/Filter&#xff09; \u4e3a\u4e86\u5728\u795e\u7ecf\u7f51\u7edc\u4e2d\u5b9e\u73b0\u8fd9\u4e00\u70b9&#xff0c;CNN\u5f15\u5165\u4e86\u5377\u79ef\u6838&#xff08;Kernel&#xff09;&#xff0c;\u4e5f\u5e38\u88ab\u79f0\u4e3a\u6ee4\u6ce2\u5668&#xff08;Filter&#xff09;\u3002\u4e00\u4e2a\u5377\u79ef\u6838&#xff0c;\u672c\u8d28\u4e0a\u5c31\u662f\u4e00\u4e2a\u5c0f\u578b\u7684\u6743\u91cd\u77e9\u9635&#xff08;\u4f8b\u59823&#215;3\u62165&#215;5&#xff09;&#xff0c;\u5b83\u88ab\u8bbe\u8ba1\u7528\u6765\u68c0\u6d4b\u4e00\u79cd\u7279\u5b9a\u7684\u5c40\u90e8\u7279\u5f81&#xff08;\u5982\u4e00\u4e2a\u7279\u5b9a\u7684\u8fb9\u7f18\u3001\u4e00\u79cd\u7279\u5b9a\u7684\u989c\u8272\u7ec4\u5408\u6216\u4e00\u4e2a\u89d2\u70b9&#xff09;\u3002 \u6743\u503c\u5171\u4eab\u7684\u6838\u5fc3\u5728\u4e8e&#xff0c;\u8fd9\u4e2a\u5377\u79ef\u6838\u4f1a\u50cf\u4e00\u4e2a\u6ed1\u52a8\u7a97\u53e3\u4e00\u6837&#xff0c;\u7cfb\u7edf\u6027\u5730\u626b\u63cf\u6574\u5f20\u8f93\u5165\u56fe\u50cf\u3002\u5728\u6bcf\u4e00\u4e2a\u4f4d\u7f6e&#xff0c;\u5b83\u90fd\u4e0e\u56fe\u50cf\u7684\u5bf9\u5e94\u5c40\u90e8\u533a\u57df\u8fdb\u884c\u8ba1\u7b97&#xff08;\u70b9\u79ef\u8fd0\u7b97&#xff09;&#xff0c;\u5e76\u5c06\u7ed3\u679c\u8bb0\u5f55\u5728\u8f93\u51fa\u7279\u5f81\u56fe\u7684\u76f8\u5e94\u4f4d\u7f6e\u3002\u91cd\u8981\u7684\u662f&#xff0c;\u5728\u6574\u4e2a\u6ed1\u52a8\u7684\u8fc7\u7a0b\u4e2d&#xff0c;\u8fd9\u4e2a\u5377\u79ef\u6838\u7684\u6743\u91cd\u662f\u56fa\u5b9a\u4e0d\u53d8\u7684\u3002<\/p>\n<p>\u5de8\u5927\u7684\u6548\u7387\u63d0\u5347 \u6743\u503c\u5171\u4eab\u7684\u610f\u4e49\u662f\u5de8\u5927\u7684\u3002\u5b83\u610f\u5473\u7740&#xff0c;\u6211\u4eec\u53ea\u9700\u8981\u5b66\u4e60\u4e00\u4e2a\u201c\u773c\u775b\u63a2\u6d4b\u5668\u201d&#xff08;\u5373\u4e00\u4e2a\u5377\u79ef\u6838&#xff09;&#xff0c;\u5c31\u53ef\u4ee5\u7528\u5b83\u6765\u68c0\u6d4b\u51fa\u56fe\u50cf\u4e2d\u6240\u6709\u4f4d\u7f6e\u7684\u773c\u775b\u3002\u6211\u4eec\u4e0d\u518d\u9700\u8981\u4e3a\u56fe\u50cf\u7684\u6bcf\u4e00\u4e2a\u50cf\u7d20\u4f4d\u7f6e&#xff0c;\u90fd\u5355\u72ec\u53bb\u5b66\u4e60\u4e00\u4e2a\u201c\u773c\u775b\u63a2\u6d4b\u5668\u201d\u3002 \u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u4e00\u4e2a\u5377\u79ef\u5c42\u9700\u8981\u5b66\u4e60\u7684\u53c2\u6570\u6570\u91cf&#xff0c;\u4ec5\u4ec5\u662f\u5176\u5305\u542b\u7684\u5377\u79ef\u6838\u7684\u6743\u91cd\u6570\u91cf&#xff0c;\u8fd9\u4e0e\u8f93\u5165\u56fe\u50cf\u7684\u5c3a\u5bf8\u5b8c\u5168\u65e0\u5173\u3002\u8fd9\u4f7f\u5f97CNN\u5728\u53c2\u6570\u6548\u7387\u4e0a\u8fbe\u5230\u4e86\u6781\u81f4&#xff0c;\u4e5f\u662f\u5176\u80fd\u591f\u4ece\u6709\u9650\u6570\u636e\u4e2d\u5b66\u5230\u6cdb\u5316\u80fd\u529b\u6781\u5f3a\u7684\u7279\u5f81\u7684\u6839\u672c\u539f\u56e0\u3002<\/p>\n<h5>7.1.3 \u6c60\u5316&#xff08;Pooling&#xff09;&#xff1a;\u53bb\u7c97\u53d6\u7cbe&#xff0c;\u964d\u7ef4\u51cf\u53c2<\/h5>\n<p>\u5728\u901a\u8fc7\u5377\u79ef\u5c42\u63d0\u53d6\u5230\u4e00\u7cfb\u5217\u7279\u5f81\u4e4b\u540e&#xff08;\u4f8b\u5982&#xff0c;\u6211\u4eec\u5f97\u5230\u4e86\u4e00\u5f20\u56fe&#xff0c;\u4e0a\u9762\u6807\u8bb0\u4e86\u6240\u6709\u201c\u773c\u775b\u201d\u7279\u5f81\u88ab\u6fc0\u6d3b\u7684\u4f4d\u7f6e&#xff09;&#xff0c;CNN\u901a\u5e38\u4f1a\u8fdb\u884c\u4e00\u6b21\u6c60\u5316\u64cd\u4f5c\u3002<\/p>\n<p>\u76ee\u7684\u4e0e\u601d\u60f3 \u6c60\u5316\u7684\u6838\u5fc3\u601d\u60f3\u662f\u964d\u4f4e\u7279\u5f81\u56fe\u7684\u7a7a\u95f4\u5206\u8fa8\u7387&#xff0c;\u5373\u8fdb\u884c\u4e0b\u91c7\u6837&#xff08;Downsampling&#xff09;\u3002\u8fd9\u57fa\u4e8e\u4e00\u4e2a\u6d1e\u89c1&#xff1a;\u5f53\u6211\u4eec\u68c0\u6d4b\u5230\u4e00\u4e2a\u7279\u5f81\u540e&#xff0c;\u5176\u7cbe\u786e\u5230\u50cf\u7d20\u7ea7\u522b\u7684\u4f4d\u7f6e&#xff0c;\u5f80\u5f80\u4e0d\u5982\u5b83\u4e0e\u5176\u4ed6\u7279\u5f81\u7684\u76f8\u5bf9\u4f4d\u7f6e\u5173\u7cfb\u6765\u5f97\u91cd\u8981\u3002\u4f8b\u5982&#xff0c;\u77e5\u9053\u4e00\u53ea\u773c\u775b\u7684\u5de6\u8fb9\u6709\u4e00\u4e2a\u9f3b\u5b50&#xff0c;\u6bd4\u77e5\u9053\u8fd9\u53ea\u773c\u775b\u5728\u56fe\u50cf\u7684(105, 210)\u50cf\u7d20\u4f4d\u7f6e\u8981\u6709\u7528\u5f97\u591a\u3002 \u6c60\u5316\u64cd\u4f5c&#xff0c;\u5c31\u662f\u5728\u4e00\u4e2a\u5c40\u90e8\u90bb\u57df\u5185&#xff08;\u4f8b\u5982\u4e00\u4e2a2&#215;2\u7684\u7a97\u53e3&#xff09;&#xff0c;\u7528\u4e00\u4e2a\u5355\u4e00\u7684\u503c\u6765\u6982\u62ec\u8fd9\u4e2a\u533a\u57df\u7684\u7279\u5f81\u4fe1\u606f&#xff0c;\u4ece\u800c\u5b9e\u73b0\u4fe1\u606f\u7684\u6d53\u7f29\u548c\u964d\u7ef4\u3002<\/p>\n<p>\u6548\u679c \u6c60\u5316\u5c42\u901a\u5e38\u4e0d\u5305\u542b\u4efb\u4f55\u9700\u8981\u5b66\u4e60\u7684\u53c2\u6570&#xff0c;\u4f46\u5b83\u5e26\u6765\u4e86\u4e09\u5927\u597d\u5904&#xff1a;<\/p>\n<li>\u964d\u4f4e\u8ba1\u7b97\u91cf&#xff1a;\u5b83\u663e\u8457\u51cf\u5c0f\u4e86\u7279\u5f81\u56fe\u7684\u5c3a\u5bf8&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a2&#215;2\u7684\u6c60\u5316\u4f1a\u5c06\u7279\u5f81\u56fe\u7684\u5bbd\u9ad8\u5404\u51cf\u534a&#xff0c;\u5c3a\u5bf8\u53d8\u4e3a\u539f\u6765\u76841\/4&#xff09;&#xff0c;\u4ece\u800c\u5927\u5e45\u964d\u4f4e\u4e86\u540e\u7eed\u7f51\u7edc\u5c42\u7684\u53c2\u6570\u6570\u91cf\u548c\u8ba1\u7b97\u8d1f\u62c5\u3002<\/li>\n<li>\u589e\u52a0\u611f\u53d7\u91ce&#xff1a;\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u91cd\u8981\u7684\u95f4\u63a5\u6548\u679c\u3002\u7531\u4e8e\u6c60\u5316\u5c42\u7f29\u5c0f\u4e86\u7279\u5f81\u56fe&#xff0c;\u4f7f\u5f97\u540e\u7eed\u7684\u5377\u79ef\u5c42\u867d\u7136\u5176\u5377\u79ef\u6838\u5c3a\u5bf8\u4e0d\u53d8&#xff0c;\u4f46\u5176\u6bcf\u4e00\u4e2a\u5355\u5143\u6240\u80fd\u201c\u770b\u5230\u201d\u7684**\u539f\u59cb\u8f93\u5165\u56fe\u50cf\u7684\u533a\u57df&#xff08;\u5373\u611f\u53d7\u91ce&#xff09;**\u5374\u53d8\u5927\u4e86\u3002\u8fd9\u4f7f\u5f97\u7f51\u7edc\u80fd\u591f\u5b66\u4e60\u5230\u66f4\u5927\u5c3a\u5ea6\u3001\u66f4\u62bd\u8c61\u7684\u7279\u5f81\u3002<\/li>\n<li>\u63d0\u4f9b\u5e73\u79fb\u4e0d\u53d8\u6027&#xff1a;\u6c60\u5316\u64cd\u4f5c\u4f7f\u5f97\u6a21\u578b\u5bf9\u4e8e\u7279\u5f81\u5728\u56fe\u50cf\u4e2d\u7684\u5fae\u5c0f\u4f4d\u79fb&#xff0c;\u5177\u6709\u4e86\u4e00\u5b9a\u7684\u5bb9\u5fcd\u5ea6\u3002\u4f8b\u5982&#xff0c;\u5373\u4f7f\u76ee\u6807\u57282&#215;2\u7684\u7a97\u53e3\u5185\u79fb\u52a8\u4e86\u4e00\u4e2a\u50cf\u7d20&#xff0c;\u53ea\u8981\u5b83\u4ecd\u7136\u662f\u8fd9\u4e2a\u7a97\u53e3\u5185\u7684\u6700\u5f3a\u54cd\u5e94&#xff0c;\u6700\u5927\u6c60\u5316\u7684\u8f93\u51fa\u7ed3\u679c\u5c31\u4e0d\u4f1a\u6539\u53d8\u3002\u8fd9\u589e\u5f3a\u4e86\u6a21\u578b\u7684\u9c81\u68d2\u6027\u3002<\/li>\n<p>\u8fd9\u4e09\u5927\u601d\u60f3\u2014\u2014\u5c40\u90e8\u8fde\u63a5\u3001\u6743\u503c\u5171\u4eab\u3001\u6c60\u5316\u2014\u2014\u5982\u4e09\u6839\u64ce\u5929\u4e4b\u67f1&#xff0c;\u5171\u540c\u652f\u6491\u8d77\u4e86CNN\u8fd9\u5ea7\u5b8f\u4f1f\u7684\u5927\u53a6\u3002\u5b83\u4eec\u4f7f\u5f97CNN\u80fd\u591f\u4ee5\u4e00\u79cd\u6781\u5176\u9ad8\u6548\u3001\u4e14\u7b26\u5408\u89c6\u89c9\u89c4\u5f8b\u7684\u65b9\u5f0f&#xff0c;\u4ece\u539f\u59cb\u50cf\u7d20\u4e2d\u9010\u5c42\u63d0\u53d6\u51fa\u8d8a\u6765\u8d8a\u62bd\u8c61\u3001\u8d8a\u6765\u8d8a\u6709\u610f\u4e49\u7684\u7279\u5f81\u8868\u793a\u3002<\/p>\n<hr \/>\n<h4>7.2 \u6838\u5fc3\u7ec4\u4ef6\u8be6\u89e3&#xff1a;\u6784\u5efaCNN\u7684\u201c\u4e50\u9ad8\u79ef\u6728\u201d<\/h4>\n<p>\u7406\u89e3\u4e86CNN\u7684\u6838\u5fc3\u601d\u60f3\u540e&#xff0c;\u6211\u4eec\u5c31\u53ef\u4ee5\u6765\u8be6\u7ec6\u5ba1\u89c6\u6784\u6210\u4e00\u4e2a\u5178\u578bCNN\u6a21\u578b\u7684\u201c\u4e50\u9ad8\u79ef\u6728\u201d\u4e86\u2014\u2014\u5377\u79ef\u5c42\u3001\u6c60\u5316\u5c42\u548c\u5168\u8fde\u63a5\u5c42\u3002<\/p>\n<h5>7.2.1 \u5377\u79ef\u5c42&#xff08;Convolutional Layer&#xff09;&#xff1a;\u7279\u5f81\u63d0\u53d6\u7684\u5f15\u64ce<\/h5>\n<p>\u5377\u79ef\u5c42\u662fCNN\u7684\u5fc3\u810f\u548c\u7075\u9b42&#xff0c;\u5b83\u8d1f\u8d23\u6267\u884c\u6700\u4e3b\u8981\u7684\u7279\u5f81\u63d0\u53d6\u5de5\u4f5c\u3002<\/p>\n<p>\u5173\u952e\u5143\u7d20 \u4e00\u6b21\u5377\u79ef\u64cd\u4f5c&#xff0c;\u6d89\u53ca\u4ee5\u4e0b\u51e0\u4e2a\u5173\u952e\u5143\u7d20&#xff1a;<\/p>\n<ul>\n<li>\u8f93\u5165\u7279\u5f81\u56fe&#xff08;Input Feature Map&#xff09;&#xff1a;\u5b83\u53ef\u4ee5\u662f\u539f\u59cb\u7684\u56fe\u50cf&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a224x224x3\u7684\u5f69\u8272\u56fe\u50cf&#xff0c;3\u4ee3\u8868RGB\u4e09\u4e2a\u989c\u8272\u901a\u9053&#xff09;&#xff0c;\u4e5f\u53ef\u4ee5\u662f\u524d\u4e00\u4e2a\u5377\u79ef\u5c42\u8f93\u51fa\u7684\u7279\u5f81\u56fe\u3002<\/li>\n<li>\u5377\u79ef\u6838&#xff08;Kernel \/ Filter&#xff09;&#xff1a;\u4e00\u4e2a\u5c0f\u7684\u6743\u91cd\u77e9\u9635\u3002\u9700\u8981\u6ce8\u610f\u7684\u662f&#xff0c;\u5377\u79ef\u6838\u7684\u6df1\u5ea6\u5fc5\u987b\u4e0e\u8f93\u5165\u7279\u5f81\u56fe\u7684\u6df1\u5ea6\u76f8\u5339\u914d\u3002\u4f8b\u5982&#xff0c;\u5982\u679c\u8981\u5728\u4e00\u4e2a224x224x3\u7684\u56fe\u50cf\u4e0a\u505a\u5377\u79ef&#xff0c;\u90a3\u4e48\u5377\u79ef\u6838\u7684\u5c3a\u5bf8\u5c31\u5fc5\u987b\u662fkxk_x3&#xff08;\u59823x3x3&#xff09;\u3002\u5b83\u4f1a\u540c\u65f6\u5728\u4e09\u4e2a\u901a\u9053\u4e0a\u8fdb\u884c\u8ba1\u7b97&#xff0c;\u5e76\u5c06\u7ed3\u679c\u76f8\u52a0&#xff0c;\u5f97\u5230\u4e00\u4e2a\u5355\u4e00\u7684\u8f93\u51fa\u503c\u3002<\/li>\n<li>\u8f93\u51fa\u7279\u5f81\u56fe&#xff08;Output Feature Map&#xff09;&#xff1a;\u4e5f\u79f0\u4e3a\u6fc0\u6d3b\u56fe&#xff08;Activation Map&#xff09;\u3002\u5b83\u662f\u5377\u79ef\u6838\u5728\u8f93\u5165\u7279\u5f81\u56fe\u4e0a\u6ed1\u52a8\u5e76\u8ba1\u7b97\u540e\u751f\u6210\u7684\u7ed3\u679c\u3002\u8f93\u51fa\u7279\u5f81\u56fe\u4e0a\u7684\u6bcf\u4e00\u4e2a\u50cf\u7d20\u503c&#xff0c;\u90fd\u4ee3\u8868\u4e86\u5377\u79ef\u6838\u6240\u5bf9\u5e94\u7684\u90a3\u4e2a\u7279\u5b9a\u7279\u5f81&#xff0c;\u5728\u8f93\u5165\u56fe\u50cf\u8be5\u4f4d\u7f6e\u7684\u6fc0\u6d3b\u5f3a\u5ea6\u3002<\/li>\n<\/ul>\n<p>\u8d85\u53c2\u6570\u8be6\u89e3 \u5728\u5b9a\u4e49\u4e00\u4e2a\u5377\u79ef\u5c42\u65f6&#xff0c;\u6211\u4eec\u9700\u8981\u8bbe\u5b9a\u51e0\u4e2a\u5173\u952e\u7684\u8d85\u53c2\u6570&#xff0c;\u5b83\u4eec\u5171\u540c\u51b3\u5b9a\u4e86\u8be5\u5c42\u7684\u884c\u4e3a&#xff1a;<\/p>\n<ul>\n<li>\u6ee4\u6ce2\u5668\u6570\u91cf&#xff08;Number of Filters&#xff09;&#xff1a;\u8fd9\u662f\u6700\u91cd\u8981\u7684\u8d85\u53c2\u6570\u4e4b\u4e00\u3002\u5b83\u51b3\u5b9a\u4e86\u4e00\u4e2a\u5377\u79ef\u5c42\u8981\u5b66\u4e60\u591a\u5c11\u79cd\u4e0d\u540c\u7c7b\u578b\u7684\u7279\u5f81\u3002\u5982\u679c\u6211\u4eec\u8bbe\u5b9a\u6ee4\u6ce2\u5668\u6570\u91cf\u4e3a64&#xff0c;\u90a3\u4e48\u8fd9\u4e2a\u5377\u79ef\u5c42\u5c31\u4f1a\u62e5\u670964\u4e2a\u72ec\u7acb\u7684\u5377\u79ef\u6838&#xff0c;\u6bcf\u4e2a\u6838\u90fd\u53bb\u5b66\u4e60\u4e00\u79cd\u4e0d\u540c\u7684\u6a21\u5f0f\u3002\u6700\u7ec8&#xff0c;\u8be5\u5c42\u4f1a\u8f93\u51fa\u4e00\u4e2a\u6df1\u5ea6\u4e3a64\u7684\u7279\u5f81\u56fe\u3002<\/li>\n<li>\u6ee4\u6ce2\u5668\u5c3a\u5bf8&#xff08;Filter Size&#xff09;&#xff1a;\u5373\u5377\u79ef\u6838\u7684\u5bbd\u5ea6\u548c\u9ad8\u5ea6&#xff0c;\u901a\u5e38\u9009\u62e9\u8f83\u5c0f\u7684\u5c3a\u5bf8&#xff0c;\u59823&#215;3\u62165&#215;5\u3002\u5c0f\u7684\u6ee4\u6ce2\u5668\u5c3a\u5bf8\u610f\u5473\u7740\u66f4\u5c11\u7684\u53c2\u6570\u548c\u66f4\u7cbe\u7ec6\u7684\u7279\u5f81\u3002<\/li>\n<li>\u6b65\u957f&#xff08;Stride&#xff09;&#xff1a;\u6307\u5377\u79ef\u6838\u5728\u8f93\u5165\u7279\u5f81\u56fe\u4e0a\u6bcf\u6b21\u6ed1\u52a8\u7684\u50cf\u7d20\u8ddd\u79bb\u3002\u6b65\u957f\u4e3a1\u8868\u793a\u9010\u50cf\u7d20\u6ed1\u52a8&#xff0c;\u6b65\u957f\u4e3a2\u5219\u8868\u793a\u6bcf\u6b21\u8df3\u8fc7\u4e00\u4e2a\u50cf\u7d20\u3002\u8f83\u5927\u7684\u6b65\u957f\u4f1a\u4ea7\u751f\u66f4\u5c0f\u7684\u8f93\u51fa\u7279\u5f81\u56fe&#xff0c;\u6709\u7c7b\u4f3c\u6c60\u5316\u7684\u964d\u91c7\u6837\u6548\u679c\u3002<\/li>\n<li>\u586b\u5145&#xff08;Padding&#xff09;&#xff1a;\u5982\u679c\u5728\u5377\u79ef\u64cd\u4f5c\u524d&#xff0c;\u4e0d\u5148\u5728\u8f93\u5165\u7279\u5f81\u56fe\u7684\u8fb9\u7f18\u8fdb\u884c\u586b\u5145&#xff0c;\u90a3\u4e48\u8f93\u51fa\u7279\u5f81\u56fe\u7684\u5c3a\u5bf8\u5c06\u4f1a\u6bd4\u8f93\u5165\u5c0f\u3002\u586b\u5145\u901a\u5e38\u662f\u5728\u8f93\u5165\u7684\u8fb9\u7f18\u5468\u56f4\u8865\u4e0a\u4e00\u5708\u6216\u51e0\u5708\u76840\u3002\u6700\u5e38\u7528\u7684padding&#061;&#039;same&#039;\u8bbe\u7f6e&#xff0c;\u53ef\u4ee5\u786e\u4fdd\u5728\u6b65\u957f\u4e3a1\u7684\u60c5\u51b5\u4e0b&#xff0c;\u8f93\u51fa\u7279\u5f81\u56fe\u7684\u7a7a\u95f4\u5c3a\u5bf8\u4e0e\u8f93\u5165\u7279\u5f81\u56fe\u5b8c\u5168\u76f8\u540c&#xff0c;\u8fd9\u5728\u6784\u5efa\u6df1\u5c42\u7f51\u7edc\u65f6\u975e\u5e38\u65b9\u4fbf\u3002<\/li>\n<\/ul>\n<h5>7.2.2 \u6c60\u5316\u5c42&#xff08;Pooling Layer&#xff09;&#xff1a;\u4fe1\u606f\u7684\u9ad8\u5ea6\u6d53\u7f29<\/h5>\n<p>\u6c60\u5316\u5c42\u901a\u5e38\u7d27\u8ddf\u5728\u5377\u79ef\u5c42\u4e4b\u540e&#xff0c;\u7528\u4e8e\u5bf9\u63d0\u53d6\u51fa\u7684\u7279\u5f81\u8fdb\u884c\u964d\u7ef4\u548c\u6982\u62ec\u3002<\/p>\n<p>\u6700\u5927\u6c60\u5316&#xff08;Max Pooling&#xff09; \u5de5\u4f5c\u539f\u7406&#xff1a;\u8fd9\u662f\u6700\u5e38\u7528\u3001\u6700\u6709\u6548\u7684\u6c60\u5316\u65b9\u5f0f\u3002\u5b83\u5c06\u7279\u5f81\u56fe\u5212\u5206\u4e3a\u82e5\u5e72\u4e2a\u4e0d\u91cd\u53e0\u7684\u77e9\u5f62\u533a\u57df&#xff08;\u6c60\u5316\u7a97\u53e3&#xff0c;\u59822&#215;2&#xff09;&#xff0c;\u7136\u540e&#xff0c;\u5728\u6bcf\u4e2a\u533a\u57df\u5185&#xff0c;\u53ea\u53d6\u6700\u5927\u503c\u4f5c\u4e3a\u552f\u4e00\u7684\u8f93\u51fa\u3002 \u76f4\u89c2\u7406\u89e3&#xff1a;\u6700\u5927\u6c60\u5316\u4f20\u9012\u4e86\u4e00\u4e2a\u975e\u5e38\u660e\u786e\u7684\u4fe1\u606f&#xff1a;\u201c\u6211\u53ea\u5173\u5fc3\u8fd9\u4e2a\u533a\u57df\u5185\u662f\u5426\u5b58\u5728\u6211\u60f3\u8981\u7684\u90a3\u4e2a\u7279\u5f81&#xff0c;\u4ee5\u53ca\u8fd9\u4e2a\u7279\u5f81\u7684\u6700\u5f3a\u54cd\u5e94\u662f\u4ec0\u4e48\u201d\u3002\u5b83\u5bf9\u7279\u5f81\u7684\u4f4d\u7f6e\u4e0d\u654f\u611f&#xff0c;\u53ea\u5bf9\u7279\u5f81\u7684\u6709\u65e0\u548c\u5f3a\u5ea6\u654f\u611f&#xff0c;\u8fd9\u662f\u4e00\u79cd\u975e\u5e38\u6709\u6548\u7684\u975e\u7ebf\u6027\u4e0b\u91c7\u6837\u3002<\/p>\n<p>\u5e73\u5747\u6c60\u5316&#xff08;Average Pooling&#xff09; \u5de5\u4f5c\u539f\u7406&#xff1a;\u4e0e\u6700\u5927\u6c60\u5316\u7c7b\u4f3c&#xff0c;\u4f46\u5b83\u8ba1\u7b97\u7684\u662f\u6c60\u5316\u7a97\u53e3\u5185\u6240\u6709\u50cf\u7d20\u503c\u7684\u5e73\u5747\u503c\u4f5c\u4e3a\u8f93\u51fa\u3002 \u76f4\u89c2\u7406\u89e3&#xff1a;\u5e73\u5747\u6c60\u5316\u4fdd\u7559\u4e86\u6bcf\u4e2a\u533a\u57df\u7279\u5f81\u7684\u201c\u6574\u4f53\u80cc\u666f\u4fe1\u606f\u201d&#xff0c;\u5176\u8f93\u51fa\u66f4\u4e3a\u5e73\u6ed1\u3002\u5728\u5386\u53f2\u4e0a\u5b83\u66fe\u88ab\u4f7f\u7528&#xff0c;\u4f46\u5728\u73b0\u4ee3CNN\u4e2d&#xff0c;\u9664\u4e86\u5728\u7f51\u7edc\u672b\u7aef\u7528\u4e8e\u5168\u5c40\u5e73\u5747\u6c60\u5316&#xff08;Global Average Pooling&#xff09;\u4e4b\u5916&#xff0c;\u5728\u4e2d\u95f4\u5c42&#xff0c;\u5176\u6548\u679c\u901a\u5e38\u4e0d\u5982\u6700\u5927\u6c60\u5316\u3002<\/p>\n<h5>7.2.3 \u5168\u8fde\u63a5\u5c42&#xff08;Fully Connected Layer&#xff09;&#xff1a;\u7279\u5f81\u7684\u201c\u7ffb\u8bd1\u5b98\u201d<\/h5>\n<p>\u5728\u7ecf\u8fc7\u4e86\u591a\u8f6e\u201c\u5377\u79ef-\u6fc0\u6d3b-\u6c60\u5316\u201d\u7684\u7279\u5f81\u63d0\u53d6\u4e4b\u540e&#xff0c;\u6211\u4eec\u5f97\u5230\u4e86\u4e00\u7ec4\u9ad8\u5ea6\u62bd\u8c61\u3001\u4f46\u7a7a\u95f4\u7ef4\u5ea6\u8f83\u5c0f\u7684\u7279\u5f81\u56fe\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u9700\u8981\u6839\u636e\u8fd9\u4e9b\u7279\u5f81\u6765\u505a\u51fa\u6700\u7ec8\u7684\u51b3\u7b56\u3002\u8fd9\u4e2a\u4efb\u52a1&#xff0c;\u5c31\u4ea4\u7ed9\u4e86\u4f4d\u4e8eCNN\u67b6\u6784\u672b\u7aef\u7684\u5168\u8fde\u63a5\u5c42\u3002<\/p>\n<p>\u89d2\u8272\u4e0e\u4f4d\u7f6e \u5168\u8fde\u63a5\u5c42\u901a\u5e38\u662fCNN\u7684\u6700\u540e\u51e0\u4e2a\u5c42\u3002\u5728\u5b83\u4e4b\u524d&#xff0c;\u6700\u540e\u4e00\u5c42\u5377\u79ef\u6216\u6c60\u5316\u5c42\u8f93\u51fa\u7684\u7acb\u4f53\u7279\u5f81\u56fe&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a7x7x512\u7684\u7279\u5f81\u56fe&#xff09;\u4f1a\u88ab**\u5c55\u5e73&#xff08;Flatten&#xff09;**\u6210\u4e00\u4e2a\u957f\u957f\u7684\u4e00\u7ef4\u5411\u91cf&#xff08;7*7*512 &#061; 25088\u7ef4&#xff09;\u3002<\/p>\n<p>\u5de5\u4f5c\u539f\u7406\u4e0e\u529f\u80fd \u8fd9\u4e2a\u4e00\u7ef4\u5411\u91cf&#xff0c;\u968f\u540e\u4f1a\u88ab\u9001\u5165\u4e00\u4e2a\u6216\u591a\u4e2a\u6211\u4eec\u65e9\u5df2\u719f\u6089\u7684\u5168\u8fde\u63a5\u5c42&#xff08;\u5373MLP&#xff09;\u3002 \u5168\u8fde\u63a5\u5c42\u7684\u4f5c\u7528&#xff0c;\u5c31\u50cf\u4e00\u4e2a\u201c\u7ffb\u8bd1\u5b98\u201d\u3002\u5b83\u63a5\u6536\u524d\u9762\u6240\u6709\u5377\u79ef\u5c42\u8f9b\u8f9b\u82e6\u82e6\u63d0\u53d6\u51fa\u7684\u3001\u9ad8\u5ea6\u62bd\u8c61\u7684\u3001\u5206\u5e03\u5f0f\u7684\u7279\u5f81\u8868\u793a&#xff08;\u4f8b\u5982&#xff0c;\u201c\u6709\u6bdb\u8338\u8338\u7684\u8033\u6735\u201d\u3001\u201c\u6709\u80e1\u987b\u201d\u3001\u201c\u6709\u732b\u4e00\u6837\u7684\u773c\u775b\u201d\u7b49\u7279\u5f81\u7684\u6fc0\u6d3b\u503c&#xff09;&#xff0c;\u7136\u540e\u5bf9\u8fd9\u4e9b\u9ad8\u7ea7\u7279\u5f81\u8fdb\u884c\u52a0\u6743\u7ec4\u5408\u4e0e\u975e\u7ebf\u6027\u53d8\u6362&#xff0c;\u6700\u7ec8\u5c06\u5b83\u4eec\u201c\u7ffb\u8bd1\u201d\u6210\u6211\u4eec\u4efb\u52a1\u6240\u9700\u8981\u7684\u6700\u7ec8\u8f93\u51fa\u3002 \u4f8b\u5982&#xff0c;\u5728\u4e00\u4e2a1000\u7c7b\u7684\u56fe\u50cf\u5206\u7c7b\u4efb\u52a1\u4e2d&#xff0c;\u6700\u540e\u4e00\u4e2a\u5168\u8fde\u63a5\u5c42\u7684\u8f93\u51fa\u7ef4\u5ea6\u5c31\u662f1000&#xff0c;\u5e76\u901a\u8fc7Softmax\u6fc0\u6d3b\u51fd\u6570&#xff0c;\u5c06\u5176\u8f6c\u6362\u4e3a\u5bf9\u8fd91000\u4e2a\u7c7b\u522b\u7684\u9884\u6d4b\u6982\u7387\u3002<\/p>\n<p>\u901a\u8fc7\u8fd9\u4e09\u79cd\u6838\u5fc3\u7ec4\u4ef6\u7684\u6709\u673a\u7ec4\u5408\u4e0e\u5806\u53e0&#xff0c;CNN\u6784\u5efa\u8d77\u4e86\u4e00\u4e2a\u4ece\u5c40\u90e8\u5230\u5168\u5c40\u3001\u4ece\u5177\u4f53\u5230\u62bd\u8c61\u7684\u3001\u5f3a\u5927\u7684\u5206\u5c42\u7279\u5f81\u5b66\u4e60\u4f53\u7cfb\u3002<\/p>\n<h4>7.3 \u7ecf\u5178CNN\u67b6\u6784\u6f14\u8fdb&#xff1a;\u4e00\u90e8\u6d53\u7f29\u7684\u89c6\u89c9AI\u53f2<\/h4>\n<p>\u4ece\u7b2c\u4e00\u4e2a\u80fd\u591f\u5b9e\u9645\u5e94\u7528\u7684CNN&#xff0c;\u5230\u5982\u4eca\u52a8\u8f84\u4e0a\u767e\u5c42\u7684\u5e9e\u7136\u5927\u7269&#xff0c;\u8fd9\u6761\u6f14\u8fdb\u4e4b\u8def\u5145\u6ee1\u4e86\u667a\u6167\u7684\u95ea\u5149\u4e0e\u601d\u60f3\u7684\u78b0\u649e\u3002\u7406\u89e3\u8fd9\u4e9b\u7ecf\u5178\u67b6\u6784&#xff0c;\u5c31\u662f\u7406\u89e3CNN\u53d1\u5c55\u7684\u8109\u7edc\u4e0e\u7cbe\u9ad3\u3002<\/p>\n<h5>7.3.1 LeNet-5 (1998)&#xff1a;\u5f00\u5c71\u9f3b\u7956<\/h5>\n<p>\u5386\u53f2\u5730\u4f4d \u5728\u6df1\u5ea6\u5b66\u4e60\u8fd8\u672a\u6210\u4e3a\u6f6e\u6d41\u768420\u4e16\u7eaa90\u5e74\u4ee3&#xff0c;Yann LeCun\u6559\u6388\u4fbf\u8bbe\u8ba1\u51fa\u4e86LeNet-5&#xff0c;\u5e76\u6210\u529f\u5730\u5c06\u5176\u5e94\u7528\u4e8e\u7f8e\u56fd\u90ae\u653f\u7cfb\u7edf\u7684\u624b\u5199\u6570\u5b57\u8bc6\u522b\u4efb\u52a1\u4e2d\u3002\u5b83\u88ab\u516c\u8ba4\u4e3a\u7b2c\u4e00\u4e2a\u88ab\u6210\u529f\u5927\u89c4\u6a21\u5546\u4e1a\u5e94\u7528\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff0c;\u662f\u5f53\u4e4b\u65e0\u6127\u7684\u5f00\u5c71\u9f3b\u7956\u3002<\/p>\n<p>\u6838\u5fc3\u8d21\u732e LeNet-5\u7684\u67b6\u6784\u867d\u7136\u5728\u4eca\u5929\u770b\u6765\u975e\u5e38\u5c0f\u5de7&#xff0c;\u4f46\u5b83\u5374\u5960\u5b9a\u4e86\u73b0\u4ee3CNN\u7684\u57fa\u672c\u84dd\u56fe\u3002\u5b83\u9996\u6b21\u5b8c\u6574\u5730\u5c55\u793a\u4e86\u201c\u8f93\u5165 -&gt; \u5377\u79ef -&gt; \u6c60\u5316 -&gt; \u5377\u79ef -&gt; \u6c60\u5316 -&gt; \u5168\u8fde\u63a5 -&gt; \u5168\u8fde\u63a5 -&gt; \u8f93\u51fa\u201d\u8fd9\u4e00\u7ecf\u5178\u8303\u5f0f\u3002\u5b83\u4f7f\u7528\u7684\u6fc0\u6d3b\u51fd\u6570\u662fSigmoid\u548cTanh&#xff0c;\u6c60\u5316\u65b9\u5f0f\u662f\u5e73\u5747\u6c60\u5316\u3002LeNet-5\u7684\u6210\u529f&#xff0c;\u8bc1\u660e\u4e86\u8fd9\u79cd\u5206\u5c42\u7279\u5f81\u63d0\u53d6\u7684\u67b6\u6784\u662f\u884c\u4e4b\u6709\u6548\u7684\u3002<\/p>\n<h5>7.3.2 AlexNet (2012)&#xff1a;\u738b\u8005\u5f52\u6765<\/h5>\n<p>\u5386\u53f2\u5730\u4f4d \u5728LeNet\u4e4b\u540e&#xff0c;\u7531\u4e8e\u8ba1\u7b97\u80fd\u529b\u7684\u9650\u5236\u548cSVM\u7b49\u4f20\u7edf\u65b9\u6cd5\u7684\u5f3a\u52bf&#xff0c;\u795e\u7ecf\u7f51\u7edc\u7ecf\u5386\u4e86\u4e00\u6bb5\u6f2b\u957f\u7684\u201c\u5bd2\u51ac\u201d\u3002\u76f4\u52302012\u5e74&#xff0c;Alex Krizhevsky\u3001Ilya Sutskever\u548cGeoff Hinton\u5e26\u7740AlexNet&#xff0c;\u5728\u5f53\u5e74\u7684ImageNet\u5927\u89c4\u6a21\u89c6\u89c9\u8bc6\u522b\u6311\u6218\u8d5b&#xff08;ILSVRC&#xff09;\u4e2d&#xff0c;\u4ee5\u78be\u538b\u6027\u7684\u4f18\u52bf&#xff08;Top-5\u9519\u8bef\u738715.3%&#xff0c;\u8fdc\u4f4e\u4e8e\u7b2c\u4e8c\u540d\u768426.2%&#xff09;\u4e00\u4e3e\u593a\u51a0\u3002\u8fd9\u4e00\u4e8b\u4ef6&#xff0c;\u5982\u540c\u4e00\u58f0\u60ca\u96f7&#xff0c;\u5ba3\u544a\u4e86\u6df1\u5ea6\u5b66\u4e60\u65f6\u4ee3\u7684\u738b\u8005\u5f52\u6765&#xff0c;\u5f15\u7206\u4e86\u81f3\u4eca\u4ecd\u5728\u6301\u7eed\u7684\u4eba\u5de5\u667a\u80fd\u9769\u547d\u3002<\/p>\n<p>\u6838\u5fc3\u8d21\u732e AlexNet\u7684\u6210\u529f\u5e76\u975e\u5076\u7136&#xff0c;\u5b83\u5efa\u7acb\u5728LeNet\u7684\u57fa\u7840\u4e0a&#xff0c;\u5e76\u5f15\u5165\u4e86\u51e0\u4e2a\u5173\u952e\u7684\u3001\u5177\u6709\u5212\u65f6\u4ee3\u610f\u4e49\u7684\u521b\u65b0&#xff1a;<\/p>\n<ul>\n<li>\u66f4\u6df1\u66f4\u5bbd\u7684\u7f51\u7edc&#xff1a;\u5b83\u62e5\u67095\u4e2a\u5377\u79ef\u5c42\u548c3\u4e2a\u5168\u8fde\u63a5\u5c42&#xff0c;\u6bd4LeNet-5\u5927\u5f97\u591a&#xff0c;\u80fd\u591f\u5b66\u4e60\u66f4\u590d\u6742\u7684\u7279\u5f81\u3002<\/li>\n<li>ReLU\u6fc0\u6d3b\u51fd\u6570&#xff1a;\u5b83\u9996\u6b21\u5728\u5927\u578bCNN\u4e2d\u7528ReLU\u66ff\u6362\u4e86\u4f20\u7edf\u7684Sigmoid\/Tanh\u3002ReLU\u7684\u975e\u9971\u548c\u6027\u6781\u5927\u5730\u7f13\u89e3\u4e86\u68af\u5ea6\u6d88\u5931\u95ee\u9898&#xff0c;\u4f7f\u5f97\u8bad\u7ec3\u6df1\u5c42\u7f51\u7edc\u6210\u4e3a\u53ef\u80fd&#xff0c;\u6536\u655b\u901f\u5ea6\u4e5f\u5feb\u5f97\u591a\u3002<\/li>\n<li>Dropout&#xff1a;\u5728\u6700\u540e\u7684\u5168\u8fde\u63a5\u5c42\u4e2d\u4f7f\u7528\u4e86Dropout\u6280\u672f&#xff0c;\u6709\u6548\u5730\u6291\u5236\u4e86\u7531\u4e8e\u6a21\u578b\u53c2\u6570\u8fc7\u591a\u800c\u5bfc\u81f4\u7684\u8fc7\u62df\u5408\u95ee\u9898\u3002<\/li>\n<li>GPU\u8bad\u7ec3&#xff1a;\u8fd9\u662f\u5176\u6210\u529f\u7684\u5173\u952e\u79d8\u8bc0\u3002\u4f5c\u8005\u521b\u9020\u6027\u5730\u4f7f\u7528\u4e86\u4e24\u5757NVIDIA GTX 580 GPU\u8fdb\u884c\u5e76\u884c\u8bad\u7ec3&#xff0c;\u6781\u5927\u5730\u63d0\u5347\u4e86\u8ba1\u7b97\u6548\u7387&#xff0c;\u4f7f\u5f97\u8bad\u7ec3\u5982\u6b64\u5e9e\u5927\u7684\u7f51\u7edc\u6210\u4e3a\u73b0\u5b9e\u3002<\/li>\n<\/ul>\n<h5>7.3.3 VGGNet (2014)&#xff1a;\u6781\u81f4\u7684\u7b80\u7ea6\u4e0e\u6df1\u5ea6<\/h5>\n<p>\u8bbe\u8ba1\u54f2\u5b66 AlexNet\u4e4b\u540e&#xff0c;\u725b\u6d25\u5927\u5b66\u7684\u89c6\u89c9\u51e0\u4f55\u7ec4&#xff08;Visual Geometry Group, VGG&#xff09;\u63d0\u51fa\u4e86VGGNet\u3002\u5b83\u7684\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66\u662f&#xff1a;\u5927\u9053\u81f3\u7b80\u3002VGGNet\u6452\u5f03\u4e86\u82b1\u54e8\u7684\u7ed3\u6784&#xff0c;\u65e8\u5728\u63a2\u7d22\u4e00\u4e2a\u7eaf\u7cb9\u7684\u95ee\u9898\u2014\u2014\u7f51\u7edc\u6df1\u5ea6\u5bf9\u6027\u80fd\u7a76\u7adf\u6709\u591a\u5927\u5f71\u54cd&#xff1f;\u4e3a\u6b64&#xff0c;\u5b83\u5c06\u6574\u4e2a\u7f51\u7edc\u90fd\u7528\u4e00\u79cd\u6781\u5176\u89c4\u6574\u3001\u7edf\u4e00\u7684\u7ec4\u4ef6\u6765\u642d\u5efa\u3002<\/p>\n<p>\u6838\u5fc3\u8d21\u732e VGGNet\u6700\u6838\u5fc3\u7684\u8d21\u732e&#xff0c;\u662f\u8bc1\u660e\u4e86\u4f7f\u7528\u591a\u4e2a\u5806\u53e0\u7684\u3001\u975e\u5e38\u5c0f\u7684&#xff08;3&#215;3&#xff09;\u5377\u79ef\u6838&#xff0c;\u5176\u6548\u679c\u4f18\u4e8e\u4f7f\u7528\u4e00\u4e2a\u5927\u7684\u5377\u79ef\u6838\u3002\u4f8b\u5982&#xff0c;\u4e24\u4e2a3&#215;3\u7684\u5377\u79ef\u5c42\u5806\u53e0&#xff0c;\u5176\u611f\u53d7\u91ce\u7b49\u6548\u4e8e\u4e00\u4e2a5&#215;5\u7684\u5377\u79ef\u5c42&#xff0c;\u4f46\u53c2\u6570\u91cf\u66f4\u5c11&#xff0c;\u4e14\u80fd\u5f15\u5165\u66f4\u591a\u7684\u975e\u7ebf\u6027\u53d8\u6362&#xff0c;\u8868\u8fbe\u80fd\u529b\u66f4\u5f3a\u3002VGGNet\u51ed\u501f\u5176\u6781\u81f4\u7b80\u7ea6\u3001\u89c4\u6574\u7684\u201c\u79ef\u6728\u5757\u201d\u5f0f\u7ed3\u6784&#xff0c;\u5c06\u7f51\u7edc\u6df1\u5ea6\u63a8\u5411\u4e8616-19\u5c42&#xff0c;\u53d6\u5f97\u4e86\u4f18\u5f02\u7684\u6027\u80fd\u3002\u7531\u4e8e\u5176\u7ed3\u6784\u7b80\u5355\u6e05\u6670&#xff0c;VGGNet\u81f3\u4eca\u4ecd\u662f\u8bb8\u591a\u7814\u7a76\u548c\u5e94\u7528\u4e2d\u975e\u5e38\u53d7\u6b22\u8fce\u7684\u57fa\u7ebf\u6a21\u578b\u3002<\/p>\n<h5>7.3.4 GoogLeNet (Inception) (2014)&#xff1a;\u5bbd\u5ea6\u7684\u63a2\u7d22<\/h5>\n<p>\u8bbe\u8ba1\u54f2\u5b66 \u4e0eVGGNet\u540c\u5e74&#xff0c;Google\u56e2\u961f\u63d0\u51fa\u4e86GoogLeNet&#xff0c;\u5e76\u8d62\u5f97\u4e862014\u5e74\u7684ImageNet\u6311\u6218\u8d5b\u3002\u5b83\u63d0\u51fa\u4e86\u4e00\u4e2a\u4e0d\u540c\u7684\u95ee\u9898&#xff1a;\u5728\u7f51\u7edc\u4e0d\u65ad\u53d8\u6df1\u7684\u540c\u65f6&#xff0c;\u6211\u4eec\u80fd\u5426\u8ba9\u7f51\u7edc\u4e5f\u53d8\u5f97\u66f4\u201c\u5bbd\u201d&#xff0c;\u5e76\u4e14\u66f4\u9ad8\u6548&#xff1f;GoogLeNet\u7684\u6838\u5fc3&#xff0c;\u662f\u5176\u7cbe\u5de7\u8bbe\u8ba1\u7684Inception\u6a21\u5757\u3002<\/p>\n<p>\u6838\u5fc3\u8d21\u732e Inception\u6a21\u5757\u7684\u601d\u60f3\u662f&#xff0c;\u5bf9\u4e8e\u4e00\u4e2a\u8f93\u5165\u7279\u5f81\u56fe&#xff0c;\u6211\u4eec\u5f88\u96be\u9884\u77e5\u7528\u591a\u5927\u5c3a\u5bf8\u7684\u5377\u79ef\u6838\u53bb\u5904\u7406\u624d\u662f\u6700\u4f18\u7684\u3002\u90a3\u4e48&#xff0c;\u4f55\u4e0d\u5c06\u4e0d\u540c\u5c3a\u5bf8\u7684\u64cd\u4f5c\u5e76\u884c\u5730\u6267\u884c\u4e00\u904d&#xff0c;\u7136\u540e\u8ba9\u7f51\u7edc\u81ea\u5df1\u53bb\u5b66\u4e60\u5982\u4f55\u7ec4\u5408\u5b83\u4eec&#xff1f; \u4e00\u4e2aInception\u6a21\u5757\u4f1a\u5e76\u884c\u5730\u5bf9\u8f93\u5165\u4f7f\u75281&#215;1\u5377\u79ef\u30013&#215;3\u5377\u79ef\u30015&#215;5\u5377\u79ef\u4ee5\u53ca3&#215;3\u6700\u5927\u6c60\u5316&#xff0c;\u7136\u540e\u5c06\u6240\u6709\u8fd9\u4e9b\u64cd\u4f5c\u7684\u8f93\u51fa\u7279\u5f81\u56fe\u5728\u6df1\u5ea6\u8fd9\u4e2a\u7ef4\u5ea6\u4e0a**\u62fc\u63a5&#xff08;Concatenate&#xff09;**\u8d77\u6765\u3002\u8fd9\u79cd\u201c\u7f51\u7edc\u4e2d\u7684\u7f51\u7edc\u201d&#xff08;Network-in-Network&#xff09;\u7ed3\u6784&#xff0c;\u6781\u5927\u5730\u63d0\u5347\u4e86\u7f51\u7edc\u7684\u5bbd\u5ea6\u548c\u5bf9\u4e0d\u540c\u5c3a\u5ea6\u7279\u5f81\u7684\u9002\u5e94\u80fd\u529b\u3002\u540c\u65f6&#xff0c;\u5b83\u8fd8\u5de7\u5999\u5730\u4f7f\u75281&#215;1\u7684\u5377\u79ef\u6838\u6765\u8fdb\u884c\u964d\u7ef4&#xff0c;\u6781\u5927\u5730\u51cf\u5c11\u4e86\u8ba1\u7b97\u91cf\u3002<\/p>\n<h5>7.3.5 ResNet (Residual Network) (2015)&#xff1a;\u8de8\u8d8a\u6df1\u5ea6\u7684\u5929\u5811<\/h5>\n<p>\u9762\u4e34\u7684\u95ee\u9898 \u968f\u7740\u7f51\u7edc\u8d8a\u6765\u8d8a\u6df1&#xff0c;\u4eba\u4eec\u53d1\u73b0\u4e86\u4e00\u4e2a\u4ee4\u4eba\u56f0\u60d1\u7684\u73b0\u8c61&#xff1a;\u7f51\u7edc\u9000\u5316&#xff08;Degradation&#xff09;\u3002\u6309\u7406\u8bf4&#xff0c;\u4e00\u4e2a\u66f4\u6df1\u7684\u6a21\u578b&#xff0c;\u5176\u89e3\u7a7a\u95f4\u5305\u542b\u4e86\u90a3\u4e2a\u66f4\u6d45\u7684\u6a21\u578b&#xff0c;\u6027\u80fd\u81f3\u5c11\u4e0d\u5e94\u8be5\u66f4\u5dee\u3002\u4f46\u5b9e\u9a8c\u8868\u660e&#xff0c;\u5f53\u7f51\u7edc\u5806\u53e0\u5230\u4e00\u5b9a\u6df1\u5ea6\u540e&#xff08;\u4f8b\u598256\u5c42&#xff09;&#xff0c;\u5176\u8bad\u7ec3\u8bef\u5dee\u548c\u6d4b\u8bd5\u8bef\u5dee\u53cd\u800c\u4f1a\u6bd4\u4e00\u4e2a\u8f83\u6d45\u7684\u7f51\u7edc&#xff08;\u4f8b\u598220\u5c42&#xff09;\u66f4\u9ad8\u3002\u8fd9\u8bf4\u660e&#xff0c;\u8ba9\u4e00\u4e2a\u6df1\u5ea6\u7f51\u7edc\u53bb\u5b66\u4e60\u4e00\u4e2a\u6052\u7b49\u6620\u5c04&#xff08;\u5373\u4ec0\u4e48\u90fd\u4e0d\u505a&#xff0c;\u76f4\u63a5\u8f93\u51fa\u8f93\u5165&#xff09;\u90fd\u662f\u975e\u5e38\u56f0\u96be\u7684\u3002<\/p>\n<p>\u6838\u5fc3\u8d21\u732e 2015\u5e74&#xff0c;\u4f55\u607a\u660e\u7b49\u51e0\u4f4d\u534e\u4eba\u7814\u7a76\u5458\u63d0\u51fa\u7684\u6b8b\u5dee\u7f51\u7edc&#xff08;ResNet&#xff09;&#xff0c;\u5929\u624d\u822c\u5730\u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898\u3002\u5176\u6838\u5fc3\u662f\u5f15\u5165\u4e86\u6b8b\u5dee\u8fde\u63a5&#xff08;Residual Connection&#xff09;&#xff0c;\u4e5f\u5e38\u88ab\u79f0\u4e3a\u5feb\u6377\u8fde\u63a5&#xff08;Shortcut Connection&#xff09;\u3002 \u8fd9\u4e2a\u8fde\u63a5\u5141\u8bb8\u8f93\u5165\u4fe1\u53f7 x \u53ef\u4ee5\u201c\u6284\u8fd1\u9053\u201d&#xff0c;\u76f4\u63a5\u8df3\u8fc7\u4e00\u4e2a\u6216\u591a\u4e2a\u5377\u79ef\u5c42&#xff0c;\u5728\u8fd9\u4e9b\u5c42\u7684\u8f93\u51fa F(x) \u4e4b\u540e&#xff0c;\u76f4\u63a5\u4e0e\u4e4b\u76f8\u52a0&#xff0c;\u5f97\u5230\u6700\u7ec8\u7684\u8f93\u51fa H(x) &#061; F(x) &#043; x\u3002 \u8fd9\u4e2a\u7b80\u5355\u7684\u6539\u52a8&#xff0c;\u5f7b\u5e95\u6539\u53d8\u4e86\u7f51\u7edc\u7684\u5b66\u4e60\u76ee\u6807\u3002\u7f51\u7edc\u4e0d\u518d\u9700\u8981\u53bb\u62df\u5408\u4e00\u4e2a\u671f\u671b\u7684\u5b8c\u6574\u6620\u5c04 H(x)&#xff0c;\u800c\u4ec5\u4ec5\u9700\u8981\u53bb\u5b66\u4e60\u8f93\u5165\u4e0e\u8f93\u51fa\u4e4b\u95f4\u7684\u6b8b\u5dee&#xff08;Residual&#xff09;F(x)\u3002\u5982\u679c\u67d0\u4e2a\u5c42\u5bf9\u4e8e\u5f53\u524d\u4efb\u52a1\u662f\u5197\u4f59\u7684&#xff0c;\u7f51\u7edc\u53ea\u9700\u8981\u5c06\u8fd9\u4e00\u5c42\u7684\u6743\u91cd F(x) \u5b66\u62100\u5373\u53ef&#xff0c;\u6b64\u65f6 H(x) &#061; x&#xff0c;\u4e00\u4e2a\u6052\u7b49\u6620\u5c04\u88ab\u8f7b\u677e\u5b9e\u73b0\u3002<\/p>\n<p>\u5386\u53f2\u5730\u4f4d ResNet\u7684\u51fa\u73b0&#xff0c;\u5982\u540c\u4e00\u5ea7\u6865\u6881&#xff0c;\u8de8\u8d8a\u4e86\u963b\u788d\u7f51\u7edc\u8d70\u5411\u66f4\u6df1\u7684\u5929\u5811\u3002\u5b83\u4f7f\u5f97\u8bad\u7ec3\u6570\u767e\u5c42\u751a\u81f3\u4e0a\u5343\u5c42\u7684\u8d85\u6df1\u795e\u7ecf\u7f51\u7edc\u6210\u4e3a\u53ef\u80fd&#xff0c;\u6781\u5927\u5730\u63d0\u5347\u4e86\u6a21\u578b\u7684\u6027\u80fd&#xff0c;\u5e76\u518d\u6b21\u5237\u65b0\u4e86ImageNet\u7684\u8bb0\u5f55\u3002\u6b8b\u5dee\u8fde\u63a5\u7684\u601d\u60f3&#xff0c;\u6210\u4e3a\u4e86\u540e\u7eed\u51e0\u4e4e\u6240\u6709\u5148\u8fdbCNN\u67b6\u6784\u7684\u6807\u914d\u3002<\/p>\n<hr \/>\n<h4>7.4 \u8fc1\u79fb\u5b66\u4e60\u4e0e\u5fae\u8c03&#xff1a;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a<\/h4>\n<p>\u6211\u4eec\u521a\u521a\u56de\u987e\u7684\u8fd9\u4e9b\u7ecf\u5178\u67b6\u6784&#xff0c;\u5c24\u5176\u662fResNet\u7b49&#xff0c;\u5b83\u4eec\u7684\u5f3a\u5927\u6027\u80fd\u80cc\u540e&#xff0c;\u662f\u5728ImageNet\u8fd9\u6837\u62e5\u6709\u4e0a\u767e\u4e07\u5f20\u56fe\u7247\u3001\u4e0a\u5343\u4e2a\u7c7b\u522b\u7684\u6570\u636e\u96c6\u4e0a&#xff0c;\u4f7f\u7528\u6d77\u91cfGPU\u8d44\u6e90\u8bad\u7ec3\u6570\u5468\u7684\u7ed3\u679c\u3002\u8fd9\u5bf9\u4e8e\u7edd\u5927\u591a\u6570\u4e2a\u4eba\u548c\u4e2d\u5c0f\u578b\u7ec4\u7ec7\u6765\u8bf4&#xff0c;\u662f\u5b8c\u5168\u65e0\u6cd5\u590d\u73b0\u7684\u3002\u90a3\u4e48&#xff0c;\u6211\u4eec\u662f\u5426\u5c31\u65e0\u6cd5\u4eab\u53d7\u5230\u8fd9\u4e9b\u5f3a\u5927\u6a21\u578b\u7684\u7ea2\u5229\u4e86\u5462&#xff1f;\u7b54\u6848\u662f&#xff1a;\u6211\u4eec\u53ef\u4ee5&#xff0c;\u901a\u8fc7\u8fc1\u79fb\u5b66\u4e60\u3002<\/p>\n<p>7.4.1 \u8fc1\u79fb\u5b66\u4e60\u7684\u6838\u5fc3\u601d\u60f3<\/p>\n<p>\u8fc1\u79fb\u5b66\u4e60\u7684\u57fa\u77f3\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u6d1e\u89c1&#xff1a;\u77e5\u8bc6\u662f\u53ef\u4ee5\u8fc1\u79fb\u7684\u3002\u4e00\u4e2a\u5728ImageNet\u4e0a\u9884\u8bad\u7ec3\u597d\u7684CNN\u6a21\u578b&#xff0c;\u5b83\u5728\u9760\u8fd1\u8f93\u5165\u7684\u6d45\u5c42\u7f51\u7edc\u4e2d&#xff0c;\u5b66\u5230\u7684\u662f\u975e\u5e38\u901a\u7528\u7684\u89c6\u89c9\u7279\u5f81&#xff0c;\u6bd4\u5982\u8fb9\u7f18\u3001\u89d2\u70b9\u3001\u7eb9\u7406\u3001\u989c\u8272\u5757\u7b49\u3002\u8fd9\u4e9b\u57fa\u7840\u7279\u5f81&#xff0c;\u5bf9\u4e8e\u89e3\u51b3\u51e0\u4e4e\u6240\u6709\u7684\u8ba1\u7b97\u673a\u89c6\u89c9\u4efb\u52a1\u90fd\u662f\u6709\u7528\u7684&#xff0c;\u65e0\u8bba\u662f\u8bc6\u522b\u732b\u72d7&#xff0c;\u8fd8\u662f\u8bca\u65ad\u533b\u7597\u5f71\u50cf\u3002<\/p>\n<p>7.4.2 \u8fc1\u79fb\u5b66\u4e60\u5b9e\u8df5\u7b56\u7565<\/p>\n<p>\u6211\u4eec\u53ef\u4ee5\u201c\u501f\u7528\u201d\u8fd9\u4e9b\u5728\u5927\u89c4\u6a21\u6570\u636e\u96c6\u4e0a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff08;Pre-trained Model&#xff09;&#xff0c;\u5c06\u5176\u4f5c\u4e3a\u6211\u4eec\u81ea\u5df1\u4efb\u52a1\u7684\u4e00\u4e2a\u5f3a\u5927\u7684\u3001\u73b0\u6210\u7684\u7279\u5f81\u63d0\u53d6\u5668\u3002\u5177\u4f53\u64cd\u4f5c\u4e0a&#xff0c;\u901a\u5e38\u6709\u4e24\u79cd\u7b56\u7565&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u4f5c\u4e3a\u56fa\u5b9a\u7684\u7279\u5f81\u63d0\u53d6\u5668 \u8fd9\u662f\u6700\u7b80\u5355\u7684\u65b9\u6cd5\u3002\u6211\u4eec\u52a0\u8f7d\u4e00\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b&#xff08;\u5982ResNet-50&#xff09;&#xff0c;\u7136\u540e\u201c\u51bb\u7ed3\u201d\u5176\u6240\u6709\u5377\u79ef\u5c42&#xff08;\u5373\u5728\u8bad\u7ec3\u4e2d\u4fdd\u6301\u5b83\u4eec\u7684\u6743\u91cd\u4e0d\u53d8&#xff09;\u3002\u6211\u4eec\u53ea\u79fb\u9664\u5176\u539f\u59cb\u7684\u5206\u7c7b\u5934&#xff08;\u5373\u6700\u540e\u7684\u5168\u8fde\u63a5\u5c42&#xff09;&#xff0c;\u6362\u4e0a\u6211\u4eec\u81ea\u5df1\u4e3a\u65b0\u4efb\u52a1\u8bbe\u8ba1\u7684\u3001\u65b0\u7684\u3001\u672a\u8bad\u7ec3\u7684\u5206\u7c7b\u5934\u3002\u5728\u8bad\u7ec3\u65f6&#xff0c;\u6211\u4eec\u53ea\u66f4\u65b0\u8fd9\u4e2a\u65b0\u5206\u7c7b\u5934\u7684\u53c2\u6570\u3002\u8fd9\u79cd\u65b9\u6cd5\u975e\u5e38\u9002\u5408\u5f53\u6211\u4eec\u7684\u65b0\u4efb\u52a1\u6570\u636e\u96c6\u5f88\u5c0f\u7684\u65f6\u5019\u3002<\/p>\n<\/li>\n<li>\n<p>\u5fae\u8c03&#xff08;Fine-tuning&#xff09; \u8fd9\u662f\u4e00\u79cd\u66f4\u5f3a\u5927\u3001\u4e5f\u66f4\u5e38\u7528\u7684\u7b56\u7565\u3002\u6211\u4eec\u540c\u6837\u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\u5e76\u66ff\u6362\u5206\u7c7b\u5934\u3002\u4f46\u8fd9\u4e00\u6b21&#xff0c;\u6211\u4eec\u4e0d\u5b8c\u5168\u51bb\u7ed3\u5377\u79ef\u5c42\u3002\u6211\u4eec\u7528\u4e00\u4e2a\u975e\u5e38\u5c0f\u7684\u5b66\u4e60\u7387&#xff0c;\u89e3\u51bb\u5e76\u7ee7\u7eed\u8bad\u7ec3\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u90e8\u5206\u6216\u5168\u90e8\u5377\u79ef\u5c42&#xff08;\u901a\u5e38\u662f\u66f4\u9760\u8fd1\u8f93\u51fa\u7684\u3001\u66f4\u62bd\u8c61\u7684\u9ad8\u5c42\u7279\u5f81\u5c42&#xff09;\u3002 \u8fd9\u6837\u505a&#xff0c;\u53ef\u4ee5\u8ba9\u90a3\u4e9b\u9884\u8bad\u7ec3\u597d\u7684\u3001\u901a\u7528\u7684\u7279\u5f81&#xff0c;\u6839\u636e\u6211\u4eec\u65b0\u4efb\u52a1\u7684\u6570\u636e\u5206\u5e03&#xff0c;\u8fdb\u884c\u4e00\u4e9b\u7ec6\u5fae\u7684\u8c03\u6574&#xff0c;\u4f7f\u5176\u66f4\u80fd\u201c\u4e13\u7cbe\u201d\u4e8e\u6211\u4eec\u7684\u7279\u5b9a\u4efb\u52a1\u3002\u5fae\u8c03\u901a\u5e38\u80fd\u5e26\u6765\u6bd4\u56fa\u5b9a\u7279\u5f81\u63d0\u53d6\u5668\u66f4\u597d\u7684\u6027\u80fd&#xff0c;\u5c24\u5176\u662f\u5728\u6211\u4eec\u7684\u65b0\u4efb\u52a1\u6570\u636e\u91cf\u4e0d\u662f\u7279\u522b\u5c0f\u7684\u60c5\u51b5\u4e0b\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u8fc1\u79fb\u5b66\u4e60&#xff0c;\u8ba9\u6211\u4eec\u5f97\u4ee5\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u7528\u6709\u9650\u7684\u6570\u636e\u548c\u8ba1\u7b97\u8d44\u6e90&#xff0c;\u5feb\u901f\u5730\u6784\u5efa\u51fa\u6027\u80fd\u4f18\u5f02\u7684\u89c6\u89c9\u6a21\u578b\u3002<\/p>\n<hr \/>\n<h4>7.5 CNN\u7684\u5e94\u7528&#xff1a;\u4ece\u5206\u7c7b\u5230\u66f4\u5e7f\u9614\u7684\u4e16\u754c<\/h4>\n<p>\u867d\u7136\u6211\u4eec\u672c\u7ae0\u7684\u8ba8\u8bba\u5927\u591a\u56f4\u7ed5\u56fe\u50cf\u5206\u7c7b\u5c55\u5f00&#xff0c;\u4f46\u8fd9\u4ec5\u4ec5\u662fCNN\u80fd\u529b\u7684\u51b0\u5c71\u4e00\u89d2\u3002\u638c\u63e1\u4e86\u5206\u5c42\u7279\u5f81\u63d0\u53d6\u8fd9\u4e00\u6838\u5fc3\u80fd\u529b\u540e&#xff0c;CNN\u88ab\u5e7f\u6cdb\u5730\u5e94\u7528\u4e8e\u5404\u79cd\u66f4\u590d\u6742\u7684\u89c6\u89c9\u4efb\u52a1\u4e2d\u3002<\/p>\n<p>7.5.1 \u56fe\u50cf\u5206\u7c7b&#xff08;Image Classification&#xff09;<\/p>\n<p>\u8fd9\u662fCNN\u6700\u57fa\u7840\u3001\u6700\u6838\u5fc3\u7684\u5e94\u7528&#xff0c;\u5373\u56de\u7b54\u201c\u8fd9\u5f20\u56fe\u7247\u91cc\u6709\u4ec0\u4e48&#xff1f;\u201d\u7684\u95ee\u9898\u3002\u5b83\u662f\u6240\u6709\u5176\u4ed6\u89c6\u89c9\u4efb\u52a1\u7684\u57fa\u7840\u3002<\/p>\n<p>7.5.2 \u76ee\u6807\u68c0\u6d4b&#xff08;Object Detection&#xff09;<\/p>\n<p>\u76ee\u6807\u68c0\u6d4b\u7684\u4efb\u52a1\u66f4\u8fdb\u4e00\u6b65&#xff0c;\u5b83\u9700\u8981\u56de\u7b54\u201c\u56fe\u7247\u91cc\u7684\u7269\u4f53\u5206\u522b\u662f\u4ec0\u4e48&#xff1f;\u5b83\u4eec\u5728\u54ea\u91cc&#xff1f;\u201d\u3002\u5b83\u4e0d\u4ec5\u8981\u8bc6\u522b\u51fa\u56fe\u50cf\u4e2d\u6240\u6709\u7269\u4f53\u7684\u7c7b\u522b&#xff0c;\u8fd8\u8981\u7528\u4e00\u4e2a\u7d27\u5bc6\u7684**\u8fb9\u754c\u6846&#xff08;Bounding Box&#xff09;**\u6765\u6807\u51fa\u6bcf\u4e2a\u7269\u4f53\u7684\u4f4d\u7f6e\u3002<\/p>\n<ul>\n<li>\u7ecf\u5178\u65b9\u6cd5&#xff1a;\n<ul>\n<li>\u4e24\u9636\u6bb5\u65b9\u6cd5&#xff1a;\u4ee5\u00a0Faster R-CNN\u00a0\u4e3a\u4ee3\u8868\u3002\u5b83\u9996\u5148\u901a\u8fc7\u4e00\u4e2a\u201c\u533a\u57df\u63d0\u8bae\u7f51\u7edc\u201d&#xff08;RPN&#xff09;\u6765\u751f\u6210\u4e0a\u5343\u4e2a\u53ef\u80fd\u5305\u542b\u7269\u4f53\u7684\u201c\u5019\u9009\u533a\u57df\u201d&#xff0c;\u7136\u540e&#xff0c;\u518d\u5bf9\u8fd9\u4e9b\u5019\u9009\u533a\u57df\u9010\u4e00\u8fdb\u884c\u5206\u7c7b\u548c\u8fb9\u754c\u6846\u7684\u7cbe\u786e\u56de\u5f52\u3002\u8fd9\u7c7b\u65b9\u6cd5\u901a\u5e38\u7cbe\u5ea6\u66f4\u9ad8\u3002<\/li>\n<li>\u5355\u9636\u6bb5\u65b9\u6cd5&#xff1a;\u4ee5\u00a0YOLO (You Only Look Once)\u00a0\u7cfb\u5217\u4e3a\u4ee3\u8868\u3002\u5b83\u6452\u5f03\u4e86\u5019\u9009\u533a\u57df\u7684\u6b65\u9aa4&#xff0c;\u5c06\u76ee\u6807\u68c0\u6d4b\u89c6\u4e3a\u4e00\u4e2a\u5355\u4e00\u7684\u56de\u5f52\u95ee\u9898&#xff0c;\u76f4\u63a5\u5728\u6574\u4e2a\u56fe\u50cf\u4e0a&#xff0c;\u4e00\u6b21\u6027\u5730\u9884\u6d4b\u51fa\u6240\u6709\u7269\u4f53\u7684\u7c7b\u522b\u548c\u8fb9\u754c\u6846\u5750\u6807\u3002\u8fd9\u7c7b\u65b9\u6cd5\u901a\u5e38\u901f\u5ea6\u6781\u5feb&#xff0c;\u975e\u5e38\u9002\u5408\u9700\u8981\u5b9e\u65f6\u5904\u7406\u7684\u5e94\u7528\u573a\u666f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>7.5.3 \u56fe\u50cf\u5206\u5272&#xff08;Image Segmentation&#xff09;<\/p>\n<p>\u56fe\u50cf\u5206\u5272\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u6700\u4e3a\u7cbe\u7ec6\u7684\u4efb\u52a1&#xff0c;\u5b83\u8ffd\u6c42\u5bf9\u56fe\u50cf\u7684\u50cf\u7d20\u7ea7\u522b\u7684\u7406\u89e3\u3002<\/p>\n<ul>\n<li>\u4e3b\u8981\u7c7b\u578b&#xff1a;\n<ul>\n<li>\u8bed\u4e49\u5206\u5272&#xff08;Semantic Segmentation&#xff09;&#xff1a;\u5b83\u7684\u76ee\u6807\u662f&#xff0c;\u5c06\u56fe\u50cf\u4e2d\u7684\u6bcf\u4e00\u4e2a\u50cf\u7d20\u90fd\u5206\u914d\u5230\u4e00\u4e2a\u7c7b\u522b\u3002\u4f8b\u5982&#xff0c;\u5c06\u56fe\u4e2d\u6240\u6709\u7684\u201c\u725b\u201d\u90fd\u6807\u8bb0\u4e3a\u84dd\u8272&#xff0c;\u6240\u6709\u7684\u201c\u8349\u5730\u201d\u90fd\u6807\u8bb0\u4e3a\u7eff\u8272&#xff0c;\u6240\u6709\u7684\u201c\u5929\u7a7a\u201d\u90fd\u6807\u8bb0\u4e3a\u9752\u8272\u3002\u5b83\u5173\u5fc3\u7684\u662f\u7c7b\u522b&#xff0c;\u4e0d\u533a\u5206\u540c\u7c7b\u4e2d\u7684\u4e2a\u4f53\u3002<\/li>\n<li>\u5b9e\u4f8b\u5206\u5272&#xff08;Instance Segmentation&#xff09;&#xff1a;\u8fd9\u662f\u6700\u5177\u6311\u6218\u6027\u7684\u4efb\u52a1\u3002\u5b83\u5728\u8bed\u4e49\u5206\u5272\u7684\u57fa\u7840\u4e0a&#xff0c;\u8fd8\u9700\u8981\u533a\u5206\u51fa\u540c\u7c7b\u522b\u7684\u4e0d\u540c\u4e2a\u4f53\u3002\u4f8b\u5982&#xff0c;\u5c06\u7b2c\u4e00\u5934\u725b\u6807\u4e3a\u84dd\u8272&#xff0c;\u5c06\u65c1\u8fb9\u7684\u7b2c\u4e8c\u5934\u725b\u6807\u4e3a\u7eff\u8272\u3002\u5b83\u9700\u8981\u540c\u65f6\u5b8c\u6210\u76ee\u6807\u68c0\u6d4b\u548c\u8bed\u4e49\u5206\u5272\u4e24\u9879\u5de5\u4f5c\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u4e00\u540c\u8e0f\u4e0a\u4e86\u4e00\u6bb5\u6d1e\u6089\u56fe\u50cf\u5965\u79d8\u7684\u7cbe\u5f69\u65c5\u7a0b\u3002\u6211\u4eec\u4ece\u8d4b\u4e88CNN\u72ec\u7279\u201c\u89c6\u89c9\u667a\u6167\u201d\u7684\u4e09\u5927\u57fa\u77f3\u601d\u60f3\u2014\u2014\u5c40\u90e8\u8fde\u63a5\u3001\u6743\u503c\u5171\u4eab\u4e0e\u6c60\u5316\u2014\u2014\u51fa\u53d1&#xff0c;\u6df1\u5165\u7406\u89e3\u4e86\u5176\u9ad8\u6548\u7279\u5f81\u63d0\u53d6\u7684\u672c\u8d28\u3002\u6211\u4eec\u8be6\u7ec6\u89e3\u6784\u4e86\u6784\u6210CNN\u7684\u6838\u5fc3\u7ec4\u4ef6&#xff0c;\u5e76\u89c1\u8bc1\u4e86\u4eceLeNet-5\u5230ResNet\u7684\u6ce2\u6f9c\u58ee\u9614\u7684\u67b6\u6784\u6f14\u8fdb\u53f2\u8bd7&#xff0c;\u9886\u7565\u4e86\u6df1\u5ea6\u3001\u5bbd\u5ea6\u4e0e\u8fde\u63a5\u7684\u827a\u672f\u3002<\/p>\n<p>\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u6211\u4eec\u5b66\u4f1a\u4e86\u5982\u4f55\u901a\u8fc7\u8fc1\u79fb\u5b66\u4e60\u4e0e\u5fae\u8c03&#xff0c;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u5c06\u8fd9\u4e9b\u5f3a\u5927\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u5e94\u7528\u4e8e\u6211\u4eec\u81ea\u5df1\u7684\u4efb\u52a1\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u8fd8\u5c06\u89c6\u91ce\u62d3\u5c55\u5230\u4e86\u76ee\u6807\u68c0\u6d4b\u548c\u56fe\u50cf\u5206\u5272\u7b49\u66f4\u5e7f\u9614\u7684\u5e94\u7528\u9886\u57df&#xff0c;\u4e00\u7aa5CNN\u6539\u53d8\u4e16\u754c\u7684\u5de8\u5927\u6f5c\u529b\u3002<\/p>\n<p>\u638c\u63e1\u4e86CNN&#xff0c;\u8bfb\u8005\u4fbf\u62e5\u6709\u4e86\u5f00\u542f\u8ba1\u7b97\u673a\u89c6\u89c9\u4e16\u754c\u5927\u95e8\u7684\u94a5\u5319\u3002\u5728\u4e0b\u4e00\u7ae0&#xff0c;\u6211\u4eec\u5c06\u8f6c\u5411\u53e6\u4e00\u4e2a\u540c\u6837\u8ff7\u4eba\u7684\u9886\u57df\u2014\u2014\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff0c;\u53bb\u63a2\u7d22\u53e6\u4e00\u79cd\u5f3a\u5927\u7684\u795e\u7ecf\u7f51\u7edc\u67b6\u6784&#xff0c;\u770b\u5b83\u5982\u4f55\u7406\u89e3\u4eba\u7c7b\u8bed\u8a00\u7684\u5e8f\u5217\u4e4b\u7f8e\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u516b\u7ae0&#xff1a;\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09; \u2014\u2014 \u7406\u89e3\u5e8f\u5217\u7684\u667a\u6167<\/h3>\n<p>\u5f53\u795e\u7ecf\u7f51\u7edc\u62e5\u6709\u201c\u8bb0\u5fc6\u201d<\/p>\n<p>\u5728\u524d\u4e00\u7ae0&#xff0c;\u6211\u4eec\u63a2\u7d22\u4e86\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u5982\u4f55\u901a\u8fc7\u5176\u72ec\u7279\u7684\u7a7a\u95f4\u611f\u77e5\u80fd\u529b&#xff0c;\u6d1e\u6089\u4e86\u9759\u6001\u56fe\u50cf\u7684\u5965\u79d8\u3002\u7136\u800c&#xff0c;\u6211\u4eec\u6240\u5904\u7684\u4e16\u754c&#xff0c;\u66f4\u591a\u7684\u662f\u4ee5\u4e00\u79cd\u52a8\u6001\u7684\u3001\u6d41\u52a8\u7684\u5f62\u5f0f\u5b58\u5728\u7684\u2014\u2014\u8bed\u8a00\u7684\u5c55\u5f00\u3001\u97f3\u4e50\u7684\u6d41\u6dcc\u3001\u65f6\u95f4\u7684\u63a8\u79fb\u3002\u8fd9\u4e9b\u5e8f\u5217\u6570\u636e\u7684\u6838\u5fc3\u9b45\u529b&#xff0c;\u5728\u4e8e\u5176\u5185\u5728\u7684\u65f6\u95f4&#xff08;\u6216\u987a\u5e8f&#xff09;\u4f9d\u8d56\u6027&#xff1a;\u4e00\u4e2a\u8bcd\u7684\u610f\u4e49&#xff0c;\u5f80\u5f80\u53d6\u51b3\u4e8e\u5b83\u524d\u9762\u51fa\u73b0\u8fc7\u7684\u8bcd\u8bed&#xff1b;\u4eca\u5929\u7684\u80a1\u4ef7&#xff0c;\u4e0e\u8fc7\u53bb\u6570\u5929\u7684\u5e02\u573a\u8868\u73b0\u606f\u606f\u76f8\u5173\u3002<\/p>\n<p>\u6211\u4eec\u4e4b\u524d\u5b66\u4e60\u7684MLP\u548cCNN&#xff0c;\u672c\u8d28\u4e0a\u90fd\u662f\u524d\u9988\u795e\u7ecf\u7f51\u7edc&#xff08;Feed-forward Neural Networks&#xff09;\u3002\u5b83\u4eec\u662f\u201c\u5931\u5fc6\u201d\u7684\u3002\u5bf9\u4e8e\u4e00\u4e2a\u8f93\u5165\u5e8f\u5217&#xff0c;\u5b83\u4eec\u5728\u5904\u7406\u7b2c\u4e09\u4e2a\u5143\u7d20\u65f6&#xff0c;\u5df2\u7ecf\u5b8c\u5168\u5fd8\u8bb0\u4e86\u7b2c\u4e00\u4e2a\u548c\u7b2c\u4e8c\u4e2a\u5143\u7d20\u662f\u4ec0\u4e48\u3002\u8fd9\u79cd\u7ed3\u6784&#xff0c;\u4f7f\u5b83\u4eec\u65e0\u6cd5\u6355\u6349\u5e8f\u5217\u4e2d\u81f3\u5173\u91cd\u8981\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002<\/p>\n<p>\u4e3a\u4e86\u8ba9\u795e\u7ecf\u7f51\u7edc\u62e5\u6709\u201c\u8bb0\u5fc6\u201d&#xff0c;\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;Recurrent Neural Network, RNN&#xff09;\u5e94\u8fd0\u800c\u751f\u3002RNN\u7684\u8bbe\u8ba1\u4e2d&#xff0c;\u8574\u542b\u7740\u4e00\u79cd\u4f18\u96c5\u800c\u5f3a\u5927\u7684\u601d\u60f3&#xff1a;\u5b83\u5f15\u5165\u4e86\u4e00\u4e2a\u5faa\u73af&#xff08;Recurrent&#xff09;\u7684\u8fde\u63a5&#xff0c;\u5141\u8bb8\u4fe1\u606f\u5728\u7f51\u7edc\u5904\u7406\u5e8f\u5217\u7684\u4e0d\u540c\u65f6\u95f4\u6b65\u4e4b\u95f4&#xff0c;\u5f97\u4ee5\u6301\u7eed\u5b58\u5728&#xff08;Persist&#xff09;\u3002\u8fd9\u4e2a\u5faa\u73af&#xff0c;\u5c31\u50cf\u662f\u4e3a\u7f51\u7edc\u690d\u5165\u4e86\u4e00\u9897\u201c\u8bb0\u5fc6\u6838\u5fc3\u201d&#xff0c;\u4f7f\u5176\u80fd\u591f\u5c06\u8fc7\u53bb\u7684\u4fe1\u606f&#xff0c;\u5982\u6d93\u6d93\u7ec6\u6d41\u822c&#xff0c;\u4e0d\u65ad\u6c47\u5165\u5bf9\u5f53\u524d\u8f93\u5165\u7684\u7406\u89e3\u4e4b\u4e2d\u3002<\/p>\n<p>\u5728\u672c\u7ae0\u7684\u65c5\u7a0b\u4e2d&#xff0c;\u6211\u4eec\u5c06\u9996\u5148\u89e3\u6784RNN\u6700\u57fa\u7840\u7684\u5faa\u73af\u673a\u5236&#xff0c;\u5e76\u76f4\u9762\u5176\u4e0e\u751f\u4ff1\u6765\u7684\u201c\u963f\u5580\u7409\u65af\u4e4b\u8e35\u201d\u2014\u2014\u957f\u671f\u4f9d\u8d56\u95ee\u9898\u3002\u968f\u540e&#xff0c;\u6211\u4eec\u5c06\u6df1\u5165\u63a2\u7d22\u5176\u4e24\u4e2a\u5f3a\u5927\u7684\u73b0\u4ee3\u53d8\u4f53\u2014\u2014\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;\u548c\u95e8\u63a7\u5faa\u73af\u5355\u5143&#xff08;GRU&#xff09;&#xff0c;\u770b\u5b83\u4eec\u662f\u5982\u4f55\u901a\u8fc7\u7cbe\u5de7\u7edd\u4f26\u7684\u201c\u95e8\u63a7\u201d\u8bbe\u8ba1&#xff0c;\u5b9e\u73b0\u4e86\u5bf9\u9065\u8fdc\u8bb0\u5fc6\u7684\u6709\u6548\u6355\u6349\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u5b66\u4e60\u5982\u4f55\u901a\u8fc7\u53cc\u5411\u548c\u5806\u53e0\u7684\u65b9\u5f0f&#xff0c;\u6784\u5efa\u51fa\u66f4\u5f3a\u5927\u7684\u5e8f\u5217\u6a21\u578b&#xff0c;\u5e76\u6700\u7ec8\u5c06\u5b83\u4eec\u5e94\u7528\u4e8e\u81ea\u7136\u8bed\u8a00\u5904\u7406\u548c\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u7b49\u771f\u5b9e\u4e16\u754c\u7684\u4efb\u52a1\u4e2d\u3002<\/p>\n<p>\u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u4e00\u540c\u5f00\u542f\u8fd9\u6bb5\u65c5\u7a0b&#xff0c;\u53bb\u63a2\u7d22\u795e\u7ecf\u7f51\u7edc\u662f\u5982\u4f55\u5b66\u4f1a\u7406\u89e3\u65f6\u95f4&#xff0c;\u5e76\u62e5\u6709\u90a3\u4efd\u7406\u89e3\u5e8f\u5217\u7684\u72ec\u7279\u667a\u6167\u7684\u3002<\/p>\n<h4>8.1 RNN\u7684\u7ed3\u6784\u4e0e\u6311\u6218&#xff1a;\u4f18\u96c5\u7684\u5faa\u73af\u4e0e\u8106\u5f31\u7684\u8bb0\u5fc6<\/h4>\n<h5>8.1.1 \u5faa\u73af\u7684\u6838\u5fc3&#xff1a;\u9690\u85cf\u72b6\u6001&#xff08;Hidden State&#xff09;<\/h5>\n<ul>\n<li>\n<p>\u7ed3\u6784\u5256\u6790 \u4e4d\u4e00\u770b&#xff0c;\u4e00\u4e2aRNN\u5355\u5143\u7684\u7ed3\u6784\u56fe\u4e0e\u6211\u4eec\u719f\u6089\u7684\u666e\u901a\u795e\u7ecf\u7f51\u7edc\u5c42\u4f3c\u4e4e\u5f88\u76f8\u4f3c\u3002\u4f46\u5176\u4e2d\u6709\u4e00\u4e2a\u81f3\u5173\u91cd\u8981\u7684\u533a\u522b&#xff1a;\u4e00\u4e2a\u6307\u5411\u5176\u81ea\u8eab\u7684\u5faa\u73af\u7bad\u5934\u3002\u8fd9\u4e2a\u770b\u4f3c\u7b80\u5355\u7684\u5faa\u73af&#xff0c;\u6b63\u662fRNN\u6240\u6709\u9b54\u529b\u7684\u6765\u6e90\u3002\u5b83\u8868\u660e&#xff0c;\u8be5\u5c42\u7684\u8f93\u51fa&#xff0c;\u4e0d\u4ec5\u4f1a\u4f20\u9012\u7ed9\u4e0b\u4e00\u5c42&#xff0c;\u8fd8\u4f1a\u4f5c\u4e3a\u4e0b\u4e00\u6b21\u8ba1\u7b97\u7684\u8f93\u5165&#xff0c;\u518d\u6b21\u56de\u5230\u81ea\u8eab\u3002<\/p>\n<\/li>\n<li>\n<p>\u9690\u85cf\u72b6\u6001 h_t \u8fd9\u4e2a\u5728\u65f6\u95f4\u4e2d\u5faa\u73af\u4f20\u9012\u7684\u4fe1\u606f&#xff0c;\u88ab\u79f0\u4e3a\u9690\u85cf\u72b6\u6001&#xff08;Hidden State&#xff09;&#xff0c;\u6211\u4eec\u901a\u5e38\u7528 h_t \u6765\u8868\u793a\u5728\u65f6\u95f4\u6b65 t \u7684\u9690\u85cf\u72b6\u6001\u3002\u6211\u4eec\u53ef\u4ee5\u5c06 h_t \u5f62\u8c61\u5730\u6bd4\u4f5c\u662fRNN\u5728\u5904\u7406\u5b8c\u5e8f\u5217\u4e2d\u7b2c t \u4e2a\u5143\u7d20\u540e&#xff0c;\u6240\u5f62\u6210\u7684**\u201c\u77ac\u65f6\u8bb0\u5fc6\u201d**\u3002<\/p>\n<p>\u5728\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65 t&#xff0c;RNN\u5355\u5143\u4f1a\u63a5\u6536\u4e24\u4e2a\u8f93\u5165&#xff1a;<\/p>\n<li>\u5f53\u524d\u5e8f\u5217\u7684\u8f93\u5165\u00a0x_t&#xff08;\u4f8b\u5982&#xff0c;\u53e5\u5b50\u4e2d\u7684\u7b2c\u00a0t\u00a0\u4e2a\u8bcd\u7684\u8bcd\u5411\u91cf&#xff09;\u3002<\/li>\n<li>\u4e0a\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001\u00a0h_{t-1}&#xff08;\u5373\u7f51\u7edc\u5728\u5904\u7406\u5b8c\u7b2c\u00a0t-1\u00a0\u4e2a\u5143\u7d20\u540e\u7559\u4e0b\u7684\u201c\u8bb0\u5fc6\u201d&#xff09;\u3002<\/li>\n<p>RNN\u5355\u5143\u5185\u90e8\u7684\u64cd\u4f5c\u975e\u5e38\u7b80\u5355&#xff0c;\u5b83\u5c06\u8fd9\u4e24\u4e2a\u8f93\u5165\u8fdb\u884c\u7ebf\u6027\u53d8\u6362&#xff0c;\u7136\u540e\u901a\u8fc7\u4e00\u4e2a\u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570&#xff08;\u901a\u5e38\u662fTanh&#xff09;&#xff0c;\u6765\u751f\u6210\u65b0\u7684\u9690\u85cf\u72b6\u6001 h_t&#xff1a; h_t &#061; tanh(W_hh * h_{t-1} &#043; W_xh * x_t &#043; b_h) \u5176\u4e2d&#xff0c;W_hh \u662f\u4f5c\u7528\u4e8e\u4e0a\u4e00\u4e2a\u9690\u85cf\u72b6\u6001\u7684\u6743\u91cd\u77e9\u9635&#xff0c;W_xh \u662f\u4f5c\u7528\u4e8e\u5f53\u524d\u8f93\u5165\u7684\u6743\u91cd\u77e9\u9635&#xff0c;b_h \u662f\u504f\u7f6e\u9879\u3002\u540c\u65f6&#xff0c;RNN\u4e5f\u53ef\u4ee5\u6839\u636e\u65b0\u7684\u9690\u85cf\u72b6\u6001 h_t \u751f\u6210\u5f53\u524d\u65f6\u95f4\u6b65\u7684\u8f93\u51fa y_t&#xff1a; y_t &#061; W_hy * h_t &#043; b_y<\/p>\n<p>\u6700\u5173\u952e\u7684\u4e00\u70b9\u662f&#xff0c;\u5728\u5904\u7406\u6574\u4e2a\u5e8f\u5217\u7684\u8fc7\u7a0b\u4e2d&#xff0c;\u6743\u91cd\u77e9\u9635 W_hh, W_xh, W_hy \u548c\u504f\u7f6e b_h, b_y \u662f\u5171\u4eab\u7684&#xff0c;\u8fd9\u4e0eCNN\u4e2d\u7684\u6743\u503c\u5171\u4eab\u601d\u60f3\u5f02\u66f2\u540c\u5de5&#xff0c;\u6781\u5927\u5730\u51cf\u5c11\u4e86\u6a21\u578b\u7684\u53c2\u6570\u91cf\u3002<\/p>\n<\/li>\n<li>\n<p>\u6309\u65f6\u95f4\u5c55\u5f00&#xff08;Unfolding in Time&#xff09; \u4e3a\u4e86\u66f4\u6e05\u6670\u5730\u7406\u89e3\u4fe1\u606f\u662f\u5982\u4f55\u5728RNN\u4e2d\u6d41\u52a8\u7684&#xff0c;\u4e5f\u4e3a\u4e86\u80fd\u591f\u5728\u8ba1\u7b97\u673a\u4e0a\u8fdb\u884c\u5b9e\u9645\u7684\u68af\u5ea6\u8ba1\u7b97&#xff0c;\u6211\u4eec\u901a\u5e38\u4f1a\u5c06RNN\u7684\u5faa\u73af\u7ed3\u6784&#xff0c;\u6cbf\u7740\u65f6\u95f4\u5e8f\u5217\u7684\u957f\u5ea6\u5c55\u5f00\u6210\u4e00\u4e2a\u6ca1\u6709\u5faa\u73af\u7684\u3001\u7ebf\u6027\u7684\u7f51\u7edc\u3002<\/p>\n<p>\u60f3\u8c61\u4e00\u4e0b&#xff0c;\u6211\u4eec\u6709\u4e00\u4e2a\u957f\u5ea6\u4e3a3\u7684\u5e8f\u5217 (x_1, x_2, x_3)\u3002\u5c55\u5f00\u540e\u7684RNN\u770b\u8d77\u6765\u5c31\u50cf\u4e00\u4e2a\u4e09\u5c42\u7684\u795e\u7ecf\u7f51\u7edc&#xff1a;<\/p>\n<li>\u5728\u65f6\u95f4\u6b651&#xff0c;RNN\u63a5\u6536\u521d\u59cb\u9690\u85cf\u72b6\u6001\u00a0h_0&#xff08;\u901a\u5e38\u662f\u96f6\u5411\u91cf&#xff09;\u548c\u8f93\u5165\u00a0x_1&#xff0c;\u8ba1\u7b97\u51fa\u00a0h_1\u3002<\/li>\n<li>\u5728\u65f6\u95f4\u6b652&#xff0c;RNN\u63a5\u6536\u00a0h_1\u00a0\u548c\u8f93\u5165\u00a0x_2&#xff0c;\u8ba1\u7b97\u51fa\u00a0h_2\u3002<\/li>\n<li>\u5728\u65f6\u95f4\u6b653&#xff0c;RNN\u63a5\u6536\u00a0h_2\u00a0\u548c\u8f93\u5165\u00a0x_3&#xff0c;\u8ba1\u7b97\u51fa\u00a0h_3\u3002<\/li>\n<p>\u8fd9\u4e2a\u5c55\u5f00\u540e\u7684\u7f51\u7edc\u6e05\u6670\u5730\u63ed\u793a\u4e86RNN\u7684\u672c\u8d28&#xff1a;\u5b83\u662f\u4e00\u4e2a\u53c2\u6570\u5171\u4eab\u7684\u3001\u975e\u5e38\u6df1\u7684\u524d\u9988\u795e\u7ecf\u7f51\u7edc&#xff0c;\u5176\u6df1\u5ea6\u7b49\u4e8e\u5e8f\u5217\u7684\u957f\u5ea6\u3002\u4fe1\u606f&#xff08;\u9690\u85cf\u72b6\u6001&#xff09;\u5c31\u50cf\u4e00\u6761\u4f20\u9001\u5e26&#xff0c;\u5c06\u8fc7\u53bb\u7684\u4fe1\u606f\u4e00\u6b65\u6b65\u5730\u4f20\u9012\u5230\u672a\u6765\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>8.1.2 RNN\u7684\u201c\u963f\u5580\u7409\u65af\u4e4b\u8e35\u201d&#xff1a;\u957f\u671f\u4f9d\u8d56\u95ee\u9898<\/h5>\n<p>RNN\u7684\u5faa\u73af\u7ed3\u6784\u8d4b\u4e88\u4e86\u5b83\u8bb0\u5fc6\u7684\u80fd\u529b&#xff0c;\u4f46\u8fd9\u79cd\u8bb0\u5fc6\u5374\u662f\u8106\u5f31\u548c\u77ed\u6682\u7684\u3002<\/p>\n<ul>\n<li>\n<p>\u77ed\u671f\u8bb0\u5fc6\u7684\u56f0\u5883 \u8ba9\u6211\u4eec\u6765\u770b\u4e00\u4e2a\u957f\u53e5\u5b50&#xff1a;\u201cThe clouds are in the sky.\u201d&#xff08;\u4e91\u5728\u5929\u7a7a\u4e2d&#xff09;\u3002\u8981\u9884\u6d4b\u6700\u540e\u4e00\u4e2a\u8bcd\u662f\u201csky\u201d&#xff0c;\u6211\u4eec\u53ea\u9700\u8981\u770b\u5b83\u524d\u9762\u7d27\u90bb\u7684\u51e0\u4e2a\u8bcd\u5373\u53ef&#xff0c;RNN\u53ef\u4ee5\u5f88\u597d\u5730\u5904\u7406\u8fd9\u79cd\u77ed\u671f\u4f9d\u8d56\u3002 \u4f46\u518d\u770b\u53e6\u4e00\u4e2a\u53e5\u5b50&#xff1a;\u201cI grew up in France&#8230; therefore, I speak fluent French.\u201d&#xff08;\u6211\u5728\u6cd5\u56fd\u957f\u5927\u2026\u2026\u56e0\u6b64&#xff0c;\u6211\u80fd\u8bf4\u6d41\u5229\u7684\u6cd5\u8bed&#xff09;\u3002\u8981\u9884\u6d4b\u6700\u540e\u7684\u201cFrench\u201d&#xff0c;\u7f51\u7edc\u5fc5\u987b\u80fd\u591f\u56de\u5fc6\u8d77\u53e5\u5b50\u5f00\u5934\u63d0\u5230\u7684\u201cFrance\u201d\u8fd9\u4e2a\u5173\u952e\u4fe1\u606f\u3002\u5f53\u8fd9\u4e24\u4e2a\u8bcd\u8bed\u5728\u5e8f\u5217\u4e2d\u76f8\u8ddd\u5f88\u8fdc\u65f6&#xff0c;\u7b80\u5355\u7684RNN\u5c31\u5f88\u96be\u5c06\u8fd9\u4e2a\u65e9\u671f\u7684\u5173\u952e\u4fe1\u606f&#xff0c;\u6709\u6548\u5730\u4f20\u9012\u5230\u9700\u8981\u5b83\u7684\u9065\u8fdc\u672a\u6765\u3002\u8fd9\u5c31\u662f\u8457\u540d\u7684\u957f\u671f\u4f9d\u8d56\u95ee\u9898&#xff08;Long-Term Dependencies Problem&#xff09;\u3002<\/p>\n<\/li>\n<li>\n<p>\u68af\u5ea6\u6d88\u5931&#xff08;Vanishing Gradients&#xff09; \u8fd9\u4e2a\u95ee\u9898\u7684\u6839\u6e90&#xff0c;\u5728\u4e8e\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u8fc7\u7a0b\u4e2d\u7684\u8870\u51cf\u3002\u6211\u4eec\u5df2\u7ecf\u77e5\u9053&#xff0c;\u5c55\u5f00\u540e\u7684RNN\u662f\u4e00\u4e2a\u975e\u5e38\u6df1\u7684\u7f51\u7edc\u3002\u5f53\u6211\u4eec\u8ba1\u7b97\u635f\u5931\u51fd\u6570\u5173\u4e8e\u65e9\u671f\u65f6\u95f4\u6b65\u53c2\u6570\u7684\u68af\u5ea6\u65f6&#xff0c;\u8fd9\u4e2a\u68af\u5ea6\u9700\u8981\u4ece\u540e\u5f80\u524d&#xff0c;\u7a7f\u8fc7\u8bb8\u591a\u4e2a\u65f6\u95f4\u6b65\u7684RNN\u5355\u5143\u3002 \u6839\u636e\u94fe\u5f0f\u6cd5\u5219&#xff0c;\u8fd9\u4e2a\u68af\u5ea6\u4f1a\u5305\u542b\u4e00\u957f\u4e32\u7684\u8fde\u4e58\u9879&#xff0c;\u5176\u4e2d\u6700\u5173\u952e\u7684\u662f\u591a\u4e2a W_hh \u77e9\u9635\u548cTanh\u6fc0\u6d3b\u51fd\u6570\u5bfc\u6570\u7684\u8fde\u4e58\u3002\u7531\u4e8eTanh\u51fd\u6570\u7684\u5bfc\u6570\u503c\u57df\u5728 (0, 1] \u4e4b\u95f4&#xff0c;\u8fdc\u5c0f\u4e8e1&#xff0c;\u8fd9\u4e00\u957f\u4e32\u7684\u8fde\u4e58\u4f1a\u5bfc\u81f4\u68af\u5ea6\u503c\u4ee5\u6307\u6570\u7ea7\u7684\u901f\u5ea6\u8870\u51cf\u3002\u5f53\u68af\u5ea6\u4f20\u64ad\u5230\u65e9\u671f\u7684\u51e0\u4e2a\u65f6\u95f4\u6b65\u65f6&#xff0c;\u5b83\u51e0\u4e4e\u5df2\u7ecf\u8870\u51cf\u4e3a\u96f6\u3002 \u68af\u5ea6\u6d88\u5931\u610f\u5473\u7740&#xff0c;\u6a21\u578b\u51e0\u4e4e\u65e0\u6cd5\u4ece\u957f\u8ddd\u79bb\u7684\u4f9d\u8d56\u5173\u7cfb\u4e2d\u5b66\u4e60\u5230\u4efb\u4f55\u4e1c\u897f\u3002\u7f51\u7edc\u4f1a\u53d8\u5f97\u201c\u76ee\u5149\u77ed\u6d45\u201d&#xff0c;\u53ea\u80fd\u5b66\u4e60\u5230\u77ed\u671f\u6a21\u5f0f\u3002<\/p>\n<\/li>\n<li>\n<p>\u68af\u5ea6\u7206\u70b8&#xff08;Exploding Gradients&#xff09; \u4e0e\u68af\u5ea6\u6d88\u5931\u76f8\u5bf9&#xff0c;\u867d\u7136\u5728\u5b9e\u8df5\u4e2d\u4e0d\u90a3\u4e48\u5e38\u89c1&#xff0c;\u4f46\u5982\u679c\u6743\u91cd\u77e9\u9635 W_hh \u7684\u67d0\u4e9b\u503c\u8f83\u5927&#xff0c;\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u7684\u8fde\u4e58\u8fc7\u7a0b\u4e2d&#xff0c;\u4e5f\u53ef\u80fd\u6307\u6570\u7ea7\u5730\u589e\u957f&#xff0c;\u6700\u7ec8\u53d8\u6210\u4e00\u4e2a\u5de8\u5927\u7684\u6570\u503c&#xff0c;\u5bfc\u81f4\u6a21\u578b\u7684\u6743\u91cd\u66f4\u65b0\u8fc7\u5927&#xff0c;\u8bad\u7ec3\u8fc7\u7a0b\u53d8\u5f97\u4e0d\u7a33\u5b9a\u751a\u81f3\u53d1\u6563\u3002 \u5e78\u8fd0\u7684\u662f&#xff0c;\u68af\u5ea6\u7206\u70b8\u95ee\u9898\u76f8\u5bf9\u5bb9\u6613\u88ab\u53d1\u73b0\u548c\u5904\u7406\u3002\u4e00\u4e2a\u7b80\u5355\u800c\u6709\u6548\u7684\u89e3\u51b3\u65b9\u6848\u662f\u68af\u5ea6\u88c1\u526a&#xff08;Gradient Clipping&#xff09;&#xff1a;\u5728\u6bcf\u6b21\u6743\u91cd\u66f4\u65b0\u524d&#xff0c;\u68c0\u67e5\u68af\u5ea6\u7684\u8303\u6570\u3002\u5982\u679c\u8303\u6570\u8d85\u8fc7\u4e86\u4e00\u4e2a\u9884\u8bbe\u7684\u9608\u503c&#xff0c;\u5c31\u6309\u6bd4\u4f8b\u7f29\u5c0f\u68af\u5ea6&#xff0c;\u4f7f\u5176\u8303\u6570\u56de\u5230\u9608\u503c\u4e4b\u5185\u3002<\/p>\n<\/li>\n<\/ul>\n<p>\u68af\u5ea6\u6d88\u5931&#xff0c;\u662f\u6807\u51c6RNN\u96be\u4ee5\u903e\u8d8a\u7684\u9e3f\u6c9f&#xff0c;\u4e5f\u662f\u9650\u5236\u5176\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u53d1\u6325\u4f5c\u7528\u7684\u6700\u5927\u969c\u788d\u3002\u4e3a\u4e86\u514b\u670d\u8fd9\u4e00\u6311\u6218&#xff0c;\u7814\u7a76\u8005\u4eec\u8bbe\u8ba1\u51fa\u4e86\u66f4\u590d\u6742\u7684\u3001\u5e26\u6709\u7cbe\u5de7\u201c\u95e8\u63a7\u201d\u673a\u5236\u7684\u5faa\u73af\u5355\u5143\u3002<\/p>\n<hr \/>\n<h4>8.2 \u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;&#xff1a;\u8bb0\u5fc6\u201c\u95e8\u201d\u7684\u8bbe\u8ba1\u54f2\u5b66<\/h4>\n<p>\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;\u662fHochreiter\u548cSchmidhuber\u57281997\u5e74\u63d0\u51fa\u7684\u4e00\u79cd\u7279\u6b8a\u7684RNN&#xff0c;\u5b83\u88ab\u7cbe\u5fc3\u8bbe\u8ba1\u7528\u6765\u89e3\u51b3\u957f\u671f\u4f9d\u8d56\u95ee\u9898\u3002\u81f3\u4eca&#xff0c;\u5b83\u53ca\u5176\u53d8\u4f53\u4f9d\u7136\u662f\u5e8f\u5217\u5efa\u6a21\u4efb\u52a1\u4e2d\u6700\u4e3b\u6d41\u3001\u6700\u5f3a\u5927\u7684\u5de5\u5177\u3002<\/p>\n<h5>8.2.1 \u6838\u5fc3\u601d\u60f3&#xff1a;\u5f15\u5165\u201c\u7ec6\u80de\u72b6\u6001\u201d\u4f5c\u4e3a\u8bb0\u5fc6\u4e3b\u5e72<\/h5>\n<ul>\n<li>\n<p>\u7ec6\u80de\u72b6\u6001&#xff08;Cell State&#xff09;C_t LSTM\u7684\u9769\u547d\u6027\u521b\u65b0&#xff0c;\u662f\u5728RNN\u7684\u9690\u85cf\u72b6\u6001 h_t \u4e4b\u5916&#xff0c;\u989d\u5916\u5f15\u5165\u4e86\u4e00\u4e2a\u7ec6\u80de\u72b6\u6001&#xff08;Cell State&#xff09;C_t\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u8fd9\u4e2a\u7ec6\u80de\u72b6\u6001 C_t \u60f3\u8c61\u6210\u4e00\u6761\u8d2f\u7a7f\u6574\u4e2a\u65f6\u95f4\u94fe\u7684\u8bb0\u5fc6\u201c\u9ad8\u901f\u516c\u8def\u201d\u3002 \u4fe1\u606f\u5728\u8fd9\u6761\u9ad8\u901f\u516c\u8def\u4e0a\u53ef\u4ee5\u975e\u5e38\u987a\u7545\u5730\u3001\u51e0\u4e4e\u4e0d\u7ecf\u6539\u53d8\u5730\u5411\u524d\u6d41\u52a8\u3002\u5b83\u53ea\u5728\u5c11\u6570\u51e0\u4e2a\u5730\u65b9&#xff0c;\u53d7\u5230\u4e00\u4e9b\u88ab\u79f0\u4e3a**\u201c\u95e8&#xff08;Gates&#xff09;\u201d**\u7684\u7ed3\u6784\u8fdb\u884c\u7cbe\u7ec6\u7684\u3001\u53ef\u63a7\u7684\u8c03\u8282\u3002\u8fd9\u79cd\u8bbe\u8ba1&#xff0c;\u4f7f\u5f97\u68af\u5ea6\u5728\u53cd\u5411\u4f20\u64ad\u65f6&#xff0c;\u4e5f\u80fd\u591f\u6cbf\u7740\u8fd9\u6761\u201c\u9ad8\u901f\u516c\u8def\u201d\u987a\u7545\u5730\u56de\u4f20&#xff0c;\u4ece\u800c\u4ece\u6839\u672c\u4e0a\u89e3\u51b3\u4e86\u68af\u5ea6\u6d88\u5931\u7684\u95ee\u9898\u3002<\/p>\n<\/li>\n<li>\n<p>\u201c\u95e8\u201d\u7684\u9690\u55bb LSTM\u4e2d\u7684\u201c\u95e8\u201d&#xff0c;\u662f\u4e00\u79cd\u8ba9\u4fe1\u606f\u9009\u62e9\u6027\u901a\u8fc7\u7684\u7ed3\u6784\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u5176\u6bd4\u4f5c\u662f\u4fe1\u606f\u6d41\u7ba1\u9053\u4e0a\u7684**\u201c\u9600\u95e8\u201d\u6216\u201c\u5f00\u5173\u201d\u3002\u5728\u6280\u672f\u4e0a&#xff0c;\u4e00\u4e2a\u95e8\u5c31\u662f\u4e00\u4e2aSigmoid\u6fc0\u6d3b\u51fd\u6570\u5c42**&#xff0c;\u540e\u9762\u901a\u5e38\u8ddf\u7740\u4e00\u4e2a\u6309\u5143\u7d20\u76f8\u4e58\u7684\u64cd\u4f5c\u3002 Sigmoid\u5c42\u7684\u8f93\u51fa\u57280\u52301\u4e4b\u95f4&#xff0c;\u8fd9\u4e2a\u503c\u5c31\u4ee3\u8868\u4e86\u9600\u95e8\u7684\u201c\u5f00\u5408\u7a0b\u5ea6\u201d&#xff1a;<\/p>\n<ul>\n<li>\u8f93\u51fa\u4e3a0&#xff0c;\u8868\u793a\u201c\u9600\u95e8\u5173\u95ed\u201d&#xff0c;\u4e0d\u5141\u8bb8\u4efb\u4f55\u4fe1\u606f\u901a\u8fc7\u3002<\/li>\n<li>\u8f93\u51fa\u4e3a1&#xff0c;\u8868\u793a\u201c\u9600\u95e8\u5b8c\u5168\u6253\u5f00\u201d&#xff0c;\u5141\u8bb8\u6240\u6709\u4fe1\u606f\u901a\u8fc7\u3002<\/li>\n<li>\u8f93\u51fa\u57280\u548c1\u4e4b\u95f4&#xff0c;\u8868\u793a\u201c\u9600\u95e8\u534a\u5f00\u201d&#xff0c;\u5141\u8bb8\u4e00\u90e8\u5206\u4fe1\u606f\u6309\u6bd4\u4f8b\u901a\u8fc7\u3002 LSTM\u6b63\u662f\u901a\u8fc7\u8fd9\u4e9b\u53ef\u5b66\u4e60\u7684\u95e8&#xff0c;\u6765\u667a\u80fd\u5730\u51b3\u5b9a\u4f55\u65f6\u8bfb\u53d6\u3001\u5199\u5165\u548c\u9057\u5fd8\u4fe1\u606f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>8.2.2 \u4e09\u5927\u95e8\u63a7\u673a\u5236&#xff1a;\u7cbe\u5de7\u7684\u8bb0\u5fc6\u7ba1\u7406\u8005<\/h5>\n<p>\u4e00\u4e2a\u6807\u51c6\u7684LSTM\u5355\u5143&#xff0c;\u7531\u4e09\u6247\u8fd9\u6837\u7684\u95e8\u6765\u5171\u540c\u7ba1\u7406\u548c\u4fdd\u62a4\u7ec6\u80de\u72b6\u6001\u3002<\/p>\n<ul>\n<li>\n<p>\u9057\u5fd8\u95e8&#xff08;Forget Gate&#xff09; \u4f5c\u7528&#xff1a;\u8fd9\u6247\u95e8\u662fLSTM\u7684\u201c\u8bb0\u5fc6\u6e05\u7406\u5de5\u201d\u3002\u5b83\u7684\u804c\u8d23\u662f\u5ba1\u89c6\u4e0a\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001 h_{t-1} \u548c\u5f53\u524d\u8f93\u5165 x_t&#xff0c;\u7136\u540e\u51b3\u5b9a\u5e94\u8be5\u4ece\u4e0a\u4e00\u4e2a\u7ec6\u80de\u72b6\u6001 C_{t-1} \u4e2d\u4e22\u5f03\u6216\u9057\u5fd8\u54ea\u4e9b\u65e7\u7684\u4fe1\u606f\u3002 \u4f8b\u5982&#xff0c;\u5f53\u5f00\u59cb\u5904\u7406\u4e00\u4e2a\u65b0\u7684\u53e5\u5b50\u65f6&#xff0c;\u9057\u5fd8\u95e8\u53ef\u80fd\u4f1a\u51b3\u5b9a\u5fd8\u8bb0\u4e0a\u4e00\u4e2a\u53e5\u5b50\u7684\u4e3b\u8bed\u4fe1\u606f\u3002 f_t &#061; \u03c3(W_f * [h_{t-1}, x_t] &#043; b_f) \u8fd9\u91cc\u7684 f_t \u5c31\u662f\u4e00\u4e2a\u4ecb\u4e8e0\u548c1\u4e4b\u95f4\u7684\u5411\u91cf&#xff0c;\u5b83\u5c06\u4e0e C_{t-1} \u6309\u5143\u7d20\u76f8\u4e58&#xff0c;\u51b3\u5b9a\u65e7\u8bb0\u5fc6\u7684\u4fdd\u7559\u7a0b\u5ea6\u3002<\/p>\n<\/li>\n<li>\n<p>\u8f93\u5165\u95e8&#xff08;Input Gate&#xff09; \u4f5c\u7528&#xff1a;\u8fd9\u6247\u95e8\u662f\u201c\u65b0\u77e5\u8bc6\u8bb0\u5f55\u5458\u201d\u3002\u5b83\u7684\u804c\u8d23\u662f\u51b3\u5b9a\u54ea\u4e9b\u65b0\u7684\u4fe1\u606f\u5c06\u88ab\u5438\u7eb3&#xff0c;\u5e76\u5b58\u653e\u5230\u5f53\u524d\u7684\u7ec6\u80de\u72b6\u6001 C_t \u4e2d\u3002 \u8fd9\u4e2a\u8fc7\u7a0b\u5206\u4e3a\u4e24\u6b65&#xff1a;<\/p>\n<li>\u9996\u5148&#xff0c;\u4e00\u4e2aSigmoid\u5c42&#xff08;\u5373\u8f93\u5165\u95e8&#xff09;\u51b3\u5b9a\u6211\u4eec\u9700\u8981\u66f4\u65b0\u54ea\u4e9b\u503c\u3002\u00a0i_t &#061; \u03c3(W_i * [h_{t-1}, x_t] &#043; b_i)<\/li>\n<li>\u7136\u540e&#xff0c;\u4e00\u4e2aTanh\u5c42\u521b\u5efa\u4e00\u4e2a\u5019\u9009\u7684\u65b0\u4fe1\u606f\u5411\u91cf\u00a0C\u0303_t&#xff0c;\u5305\u542b\u4e86\u6240\u6709\u53ef\u80fd\u88ab\u6dfb\u52a0\u7684\u65b0\u77e5\u8bc6\u3002\u00a0C\u0303_t &#061; tanh(W_C * [h_{t-1}, x_t] &#043; b_C)\u00a0\u6700\u540e&#xff0c;\u5c06\u8fd9\u4e24\u90e8\u5206\u7ed3\u5408\u8d77\u6765&#xff0c;\u5f97\u5230\u6700\u7ec8\u8981\u6dfb\u52a0\u7684\u65b0\u8bb0\u5fc6&#xff1a;i_t * C\u0303_t\u3002<\/li>\n<\/li>\n<li>\n<p>\u66f4\u65b0\u7ec6\u80de\u72b6\u6001 \u73b0\u5728&#xff0c;\u6211\u4eec\u53ef\u4ee5\u5c06\u65e7\u7684\u7ec6\u80de\u72b6\u6001 C_{t-1} \u548c\u65b0\u7684\u5019\u9009\u8bb0\u5fc6\u7ed3\u5408\u8d77\u6765&#xff0c;\u5f97\u5230\u5f53\u524d\u7684\u7ec6\u80de\u72b6\u6001 C_t \u4e86&#xff1a; C_t &#061; f_t * C_{t-1} &#043; i_t * C\u0303_t \u8fd9\u4e2a\u516c\u5f0f\u975e\u5e38\u4f18\u7f8e&#xff1a;\u6211\u4eec\u9996\u5148\u7528\u9057\u5fd8\u95e8 f_t \u4e58\u4ee5\u65e7\u72b6\u6001 C_{t-1}&#xff0c;\u4e22\u5f03\u6389\u51b3\u5b9a\u8981\u5fd8\u8bb0\u7684\u90e8\u5206&#xff1b;\u7136\u540e&#xff0c;\u52a0\u4e0a\u7531\u8f93\u5165\u95e8 i_t \u7b5b\u9009\u8fc7\u7684\u65b0\u4fe1\u606f C\u0303_t\u3002<\/p>\n<\/li>\n<li>\n<p>\u8f93\u51fa\u95e8&#xff08;Output Gate&#xff09; \u4f5c\u7528&#xff1a;\u8fd9\u6247\u95e8\u662f\u201c\u8bb0\u5fc6\u8868\u8fbe\u8005\u201d\u3002\u7ec6\u80de\u72b6\u6001 C_t \u4e2d\u5b58\u50a8\u4e86\u4e30\u5bcc\u7684\u957f\u671f\u548c\u77ed\u671f\u8bb0\u5fc6&#xff0c;\u4f46\u6211\u4eec\u5e76\u4e0d\u4e00\u5b9a\u9700\u8981\u5c06\u6240\u6709\u8fd9\u4e9b\u8bb0\u5fc6\u90fd\u4f5c\u4e3a\u5f53\u524d\u65f6\u95f4\u6b65\u7684\u8f93\u51fa\u3002\u8f93\u51fa\u95e8\u7684\u804c\u8d23&#xff0c;\u5c31\u662f\u51b3\u5b9a\u7ec6\u80de\u72b6\u6001\u4e2d\u7684\u54ea\u4e9b\u90e8\u5206&#xff0c;\u5c06\u88ab\u63d0\u70bc\u5e76\u7528\u4f5c\u5f53\u524d\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001 h_t&#xff08;h_t \u65e2\u662f\u7ed9\u4e0b\u4e00\u5c42\u7684\u8f93\u51fa&#xff0c;\u4e5f\u662f\u7ed9\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u5faa\u73af\u8f93\u5165&#xff09;\u3002 \u8fd9\u4e2a\u8fc7\u7a0b\u540c\u6837\u5206\u4e3a\u4e24\u6b65&#xff1a;<\/p>\n<li>\u9996\u5148&#xff0c;\u4e00\u4e2aSigmoid\u5c42&#xff08;\u5373\u8f93\u51fa\u95e8&#xff09;\u51b3\u5b9a\u7ec6\u80de\u72b6\u6001\u7684\u54ea\u4e9b\u90e8\u5206\u53ef\u4ee5\u88ab\u8f93\u51fa\u3002\u00a0o_t &#061; \u03c3(W_o * [h_{t-1}, x_t] &#043; b_o)<\/li>\n<li>\u7136\u540e&#xff0c;\u6211\u4eec\u5c06\u66f4\u65b0\u540e\u7684\u7ec6\u80de\u72b6\u6001\u00a0C_t\u00a0\u901a\u8fc7\u4e00\u4e2aTanh\u5c42&#xff08;\u5c06\u5176\u503c\u538b\u7f29\u5230-1\u52301\u4e4b\u95f4&#xff09;&#xff0c;\u518d\u4e0e\u8f93\u51fa\u95e8\u7684\u8f93\u51fa\u00a0o_t\u00a0\u76f8\u4e58&#xff0c;\u5f97\u5230\u6700\u7ec8\u7684\u9690\u85cf\u72b6\u6001\u00a0h_t\u3002\u00a0h_t &#061; o_t * tanh(C_t)<\/li>\n<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u8fd9\u5957\u9057\u5fd8\u3001\u8f93\u5165\u3001\u8f93\u51fa\u7684\u7cbe\u5de7\u95e8\u63a7\u673a\u5236&#xff0c;LSTM\u5355\u5143\u5f97\u4ee5\u50cf\u4e00\u4e2a\u9ad8\u6548\u7684\u8bb0\u5fc6\u7ba1\u7406\u8005\u4e00\u6837&#xff0c;\u667a\u80fd\u5730\u7ef4\u62a4\u7740\u90a3\u6761\u8bb0\u5fc6\u7684\u201c\u9ad8\u901f\u516c\u8def\u201d&#xff0c;\u4ece\u800c\u6210\u529f\u5730\u6355\u6349\u5230\u4e86\u5e8f\u5217\u6570\u636e\u4e2d\u9065\u8fdc\u7684\u4f9d\u8d56\u5173\u7cfb\u3002<\/p>\n<h4>8.3 \u95e8\u63a7\u5faa\u73af\u5355\u5143&#xff08;GRU&#xff09;&#xff1a;LSTM\u7684\u4f18\u96c5\u7b80\u5316<\/h4>\n<p>\u95e8\u63a7\u5faa\u73af\u5355\u5143&#xff08;Gated Recurrent Unit, GRU&#xff09;\u7531Cho\u7b49\u4eba\u57282014\u5e74\u63d0\u51fa&#xff0c;\u53ef\u4ee5\u770b\u4f5c\u662fLSTM\u7684\u4e00\u4e2a\u620f\u5267\u6027\u7684\u3001\u4f46\u975e\u5e38\u6d41\u884c\u7684\u53d8\u4f53\u3002\u5b83\u5728\u4fdd\u6301LSTM\u5f3a\u5927\u6027\u80fd\u7684\u540c\u65f6&#xff0c;\u5bf9\u5185\u90e8\u7ed3\u6784\u8fdb\u884c\u4e86\u7b80\u5316&#xff0c;\u4f7f\u5f97\u6a21\u578b\u53c2\u6570\u66f4\u5c11&#xff0c;\u8ba1\u7b97\u6548\u7387\u66f4\u9ad8\u3002<\/p>\n<h5>8.3.1 \u7ed3\u6784\u4e0a\u7684\u6539\u53d8<\/h5>\n<p>GRU\u5bf9LSTM\u7684\u6838\u5fc3\u67b6\u6784\u505a\u4e86\u4e24\u4e2a\u4e3b\u8981\u7684\u6539\u53d8&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u878d\u5408\u7ec6\u80de\u72b6\u6001\u4e0e\u9690\u85cf\u72b6\u6001 GRU\u6700\u5927\u80c6\u7684\u6539\u52a8&#xff0c;\u662f\u53d6\u6d88\u4e86\u72ec\u7acb\u7684\u7ec6\u80de\u72b6\u6001 C_t\u3002\u5b83\u5c06LSTM\u4e2d\u7684\u7ec6\u80de\u72b6\u6001\u548c\u9690\u85cf\u72b6\u6001\u7684\u529f\u80fd&#xff0c;\u5408\u5e76\u5230\u4e86\u4e00\u4e2a\u5355\u4e00\u7684\u9690\u85cf\u72b6\u6001 h_t \u4e4b\u4e2d\u3002\u8fd9\u6761\u66fe\u7ecf\u7684\u8bb0\u5fc6\u201c\u9ad8\u901f\u516c\u8def\u201d\u88ab\u5e76\u5165\u4e86\u5e38\u89c4\u8f66\u9053&#xff0c;\u4f46\u901a\u8fc7\u66f4\u7cbe\u5de7\u7684\u4ea4\u901a\u89c4\u5219&#xff08;\u95e8\u63a7&#xff09;\u6765\u907f\u514d\u62e5\u5835\u3002<\/p>\n<\/li>\n<li>\n<p>\u4e24\u6247\u95e8 \u76f8\u5e94\u5730&#xff0c;GRU\u5c06LSTM\u7684\u4e09\u6247\u95e8&#xff08;\u9057\u5fd8\u95e8\u3001\u8f93\u5165\u95e8\u3001\u8f93\u51fa\u95e8&#xff09;\u7b80\u5316\u4e3a\u4e86\u4e24\u6247\u95e8&#xff1a;\u66f4\u65b0\u95e8&#xff08;Update Gate&#xff09;\u548c\u91cd\u7f6e\u95e8&#xff08;Reset Gate&#xff09;\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>8.3.2 GRU\u7684\u5de5\u4f5c\u673a\u5236<\/h5>\n<ul>\n<li>\n<p>\u66f4\u65b0\u95e8 z_t \u8fd9\u6247\u95e8\u7684\u4f5c\u7528&#xff0c;\u975e\u5e38\u7c7b\u4f3c\u4e8eLSTM\u4e2d\u9057\u5fd8\u95e8\u548c\u8f93\u5165\u95e8\u7684\u7ed3\u5408\u4f53\u3002\u5b83\u51b3\u5b9a\u4e86\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a&#xff0c;\u65b0\u7684\u9690\u85cf\u72b6\u6001 h_t \u5e94\u8be5\u76f4\u63a5\u7ee7\u627f\u4e0a\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001 h_{t-1}&#xff0c;\u4ee5\u53ca\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a&#xff0c;\u5b83\u5e94\u8be5\u63a5\u6536\u65b0\u8ba1\u7b97\u51fa\u7684\u5019\u9009\u9690\u85cf\u72b6\u6001 h\u0303_t\u3002 z_t &#061; \u03c3(W_z * [h_{t-1}, x_t]) h_t &#061; (1 &#8211; z_t) * h_{t-1} &#043; z_t * h\u0303_t \u5f53 z_t \u7684\u503c\u63a5\u8fd10\u65f6&#xff0c;h_t \u51e0\u4e4e\u5b8c\u5168\u7531\u65e7\u72b6\u6001 h_{t-1} \u6784\u6210&#xff0c;\u4fe1\u606f\u88ab\u76f4\u63a5\u4fdd\u7559&#xff1b;\u5f53 z_t \u63a5\u8fd11\u65f6&#xff0c;h_t \u5219\u4e3b\u8981\u7531\u65b0\u7684\u5019\u9009\u72b6\u6001 h\u0303_t \u6784\u6210\u3002<\/p>\n<\/li>\n<li>\n<p>\u91cd\u7f6e\u95e8 r_t \u8fd9\u6247\u95e8\u7684\u4f5c\u7528&#xff0c;\u662f\u63a7\u5236\u5728\u8ba1\u7b97\u65b0\u7684\u5019\u9009\u9690\u85cf\u72b6\u6001 h\u0303_t \u65f6&#xff0c;\u5e94\u8be5\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a\u5ffd\u7565\u6389\u8fc7\u53bb\u7684\u9690\u85cf\u72b6\u6001 h_{t-1}\u3002 r_t &#061; \u03c3(W_r * [h_{t-1}, x_t]) h\u0303_t &#061; tanh(W_h * [r_t * h_{t-1}, x_t]) \u5f53\u91cd\u7f6e\u95e8 r_t \u7684\u503c\u63a5\u8fd10\u65f6&#xff0c;r_t * h_{t-1} \u8fd9\u4e00\u9879\u5c31\u8d8b\u8fd1\u4e8e\u96f6&#xff0c;\u8fd9\u610f\u5473\u7740\u5728\u8ba1\u7b97\u65b0\u7684\u5019\u9009\u8bb0\u5fc6\u65f6&#xff0c;\u5c06\u5b8c\u5168\u5ffd\u7565\u6389\u8fc7\u53bb\u7684\u8bb0\u5fc6&#xff0c;\u53ea\u4f9d\u8d56\u4e8e\u5f53\u524d\u7684\u8f93\u5165 x_t\u3002\u8fd9\u4f7f\u5f97GRU\u80fd\u591f\u6709\u6548\u5730\u629b\u5f03\u4e0e\u672a\u6765\u65e0\u5173\u7684\u65e7\u4fe1\u606f\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>8.3.3 LSTM vs. GRU&#xff1a;\u5982\u4f55\u9009\u62e9&#xff1f;<\/h5>\n<p>\u8fd9\u662f\u4e00\u4e2a\u5728\u5b9e\u8df5\u4e2d\u7ecf\u5e38\u88ab\u95ee\u5230\u7684\u95ee\u9898\u3002<\/p>\n<ul>\n<li>\u6027\u80fd&#xff1a;\u5927\u91cf\u7684\u5b9e\u9a8c\u8868\u660e&#xff0c;\u5728\u7edd\u5927\u591a\u6570\u4efb\u52a1\u4e0a&#xff0c;\u7cbe\u5fc3\u8c03\u53c2\u7684LSTM\u548cGRU\u7684\u6027\u80fd\u90fd\u975e\u5e38\u76f8\u4f3c&#xff0c;\u6ca1\u6709\u4e00\u4e2a\u6a21\u578b\u88ab\u8bc1\u660e\u5728\u6240\u6709\u60c5\u51b5\u4e0b\u90fd\u7edd\u5bf9\u4f18\u4e8e\u53e6\u4e00\u4e2a\u3002<\/li>\n<li>\u6548\u7387&#xff1a;\u7531\u4e8eGRU\u7684\u5185\u90e8\u7ed3\u6784\u66f4\u7b80\u5355&#xff0c;\u95e8\u66f4\u5c11&#xff0c;\u56e0\u6b64\u5b83\u7684\u53c2\u6570\u6570\u91cf\u4e5f\u66f4\u5c11\u3002\u8fd9\u610f\u5473\u7740GRU\u7684\u8ba1\u7b97\u6548\u7387\u901a\u5e38\u66f4\u9ad8&#xff0c;\u8bad\u7ec3\u901f\u5ea6\u66f4\u5feb&#xff0c;\u5728\u6570\u636e\u96c6\u8f83\u5c0f\u7684\u60c5\u51b5\u4e0b&#xff0c;\u4e5f\u53ef\u80fd\u56e0\u4e3a\u53c2\u6570\u5c11\u800c\u5177\u6709\u66f4\u597d\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<li>\u5b9e\u8df5\u5efa\u8bae&#xff1a;\n<ul>\n<li>\u5982\u679c\u521a\u5f00\u59cb\u4e00\u4e2a\u65b0\u9879\u76ee&#xff0c;LSTM\u662f\u4e00\u4e2a\u975e\u5e38\u7a33\u5065\u548c\u5f3a\u5927\u7684\u8d77\u70b9&#xff0c;\u56e0\u4e3a\u5b83\u7684\u4e09\u95e8\u7ed3\u6784\u63d0\u4f9b\u4e86\u66f4\u7cbe\u7ec6\u7684\u63a7\u5236&#xff0c;\u8868\u8fbe\u80fd\u529b\u5728\u7406\u8bba\u4e0a\u53ef\u80fd\u66f4\u5f3a\u3002<\/li>\n<li>\u5982\u679c\u975e\u5e38\u5173\u5fc3\u8ba1\u7b97\u6548\u7387&#xff0c;\u6216\u8005\u5e0c\u671b\u6a21\u578b\u66f4\u8f7b\u91cf\u7ea7&#xff0c;GRU\u662f\u4e00\u4e2a\u7edd\u4f73\u7684\u66ff\u4ee3\u65b9\u6848\u3002<\/li>\n<li>\u6700\u597d\u7684\u65b9\u6cd5&#xff0c;\u5f80\u5f80\u662f\u6839\u636e\u5177\u4f53\u4efb\u52a1\u7684\u9a8c\u8bc1\u96c6\u6027\u80fd&#xff0c;\u6765\u5b9e\u9a8c\u6027\u5730\u51b3\u5b9a\u54ea\u4e00\u4e2a\u66f4\u9002\u5408\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>8.4 \u6269\u5c55RNN&#xff1a;\u6784\u5efa\u66f4\u5f3a\u5927\u7684\u5e8f\u5217\u6a21\u578b<\/h4>\n<p>\u638c\u63e1\u4e86LSTM\u548cGRU\u8fd9\u6837\u7684\u5f3a\u5927\u5faa\u73af\u5355\u5143\u540e&#xff0c;\u6211\u4eec\u8fd8\u53ef\u4ee5\u901a\u8fc7\u6539\u53d8\u7f51\u7edc\u7684\u8fde\u63a5\u65b9\u5f0f&#xff0c;\u6765\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u578b\u7684\u8868\u8fbe\u80fd\u529b\u3002<\/p>\n<h5>8.4.1 \u53cc\u5411RNN&#xff08;Bidirectional RNN&#xff09;<\/h5>\n<ul>\n<li>\n<p>\u5355\u5411\u7684\u5c40\u9650 \u6211\u4eec\u76ee\u524d\u8ba8\u8bba\u7684\u6240\u6709RNN&#xff0c;\u90fd\u662f\u5355\u5411\u7684\u3002\u5b83\u4eec\u5728\u5904\u7406\u4e00\u4e2a\u5e8f\u5217\u65f6&#xff0c;\u53ea\u80fd\u4ece\u5de6\u5230\u53f3&#xff0c;\u6309\u90e8\u5c31\u73ed\u5730\u8fdb\u884c\u3002\u8fd9\u610f\u5473\u7740&#xff0c;\u5728\u65f6\u95f4\u6b65 t \u505a\u51b3\u7b56\u65f6&#xff0c;\u6a21\u578b\u53ea\u80fd\u5229\u7528\u8fc7\u53bb&#xff08;t-1\u53ca\u4e4b\u524d&#xff09;\u7684\u4fe1\u606f\u3002 \u4f46\u5728\u8bb8\u591a\u4efb\u52a1\u4e2d&#xff0c;\u5c24\u5176\u662f\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e2d&#xff0c;\u7406\u89e3\u4e00\u4e2a\u8bcd\u7684\u51c6\u786e\u542b\u4e49&#xff0c;\u5f80\u5f80\u9700\u8981\u540c\u65f6\u8003\u8651\u5176\u5de6\u4fa7\u548c\u53f3\u4fa7\u7684\u4e0a\u4e0b\u6587\u3002\u4f8b\u5982&#xff0c;\u5728\u53e5\u5b50\u201cHe banked the plane to the left.\u201d\u4e2d&#xff0c;\u8981\u7406\u89e3\u201cbanked\u201d&#xff08;\u503e\u659c&#xff09;\u7684\u542b\u4e49&#xff0c;\u53f3\u4fa7\u7684\u201cplane\u201d&#xff08;\u98de\u673a&#xff09;\u4e00\u8bcd\u81f3\u5173\u91cd\u8981\u3002<\/p>\n<\/li>\n<li>\n<p>\u5de5\u4f5c\u539f\u7406 \u53cc\u5411RNN&#xff08;BRNN&#xff09;\u6b63\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u800c\u8bbe\u8ba1\u7684\u3002\u5b83\u7684\u7ed3\u6784\u975e\u5e38\u76f4\u89c2&#xff1a;\u5b83\u7531\u4e24\u4e2a\u72ec\u7acb\u7684RNN&#xff08;\u53ef\u4ee5\u662fSimple RNN, LSTM\u6216GRU&#xff09;\u7ec4\u6210&#xff0c;\u5e76\u6392\u8fd0\u884c\u3002<\/p>\n<li>\u4e00\u4e2a\u6b63\u5411RNN&#xff0c;\u6309\u6b63\u5e38\u987a\u5e8f&#xff08;\u4ece\u00a0t&#061;1\u00a0\u5230\u00a0t&#061;T&#xff09;\u5904\u7406\u8f93\u5165\u5e8f\u5217&#xff0c;\u751f\u6210\u4e00\u7cfb\u5217\u6b63\u5411\u7684\u9690\u85cf\u72b6\u6001\u00a0(h\u20d7_1, h\u20d7_2, &#8230;, h\u20d7_T)\u3002<\/li>\n<li>\u4e00\u4e2a\u53cd\u5411RNN&#xff0c;\u6309\u76f8\u53cd\u987a\u5e8f&#xff08;\u4ece\u00a0t&#061;T\u00a0\u5230\u00a0t&#061;1&#xff09;\u5904\u7406\u8f93\u5165\u5e8f\u5217&#xff0c;\u751f\u6210\u4e00\u7cfb\u5217\u53cd\u5411\u7684\u9690\u85cf\u72b6\u6001\u00a0(h\u20d6_1, h\u20d6_2, &#8230;, h\u20d6_T)\u3002 \u5728\u4efb\u4f55\u4e00\u4e2a\u65f6\u95f4\u6b65\u00a0t&#xff0c;\u8be5\u65f6\u95f4\u6b65\u7684\u6700\u7ec8\u9690\u85cf\u72b6\u6001\u8868\u793a&#xff0c;\u5c31\u662f\u5c06\u6b63\u5411\u9690\u85cf\u72b6\u6001\u00a0h\u20d7_t\u00a0\u548c\u53cd\u5411\u9690\u85cf\u72b6\u6001\u00a0h\u20d6_t\u00a0\u8fdb\u884c\u62fc\u63a5&#xff08;Concatenate&#xff09;&#xff1a;h_t &#061; [h\u20d7_t ; h\u20d6_t]\u3002<\/li>\n<\/li>\n<li>\n<p>\u4f18\u52bf \u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;h_t \u4e2d\u5c31\u540c\u65f6\u5305\u542b\u4e86\u6765\u81ea\u8fc7\u53bb\u548c\u672a\u6765\u7684\u4e0a\u4e0b\u6587\u4fe1\u606f&#xff0c;\u6781\u5927\u5730\u589e\u5f3a\u4e86\u6a21\u578b\u5728\u6bcf\u4e2a\u65f6\u95f4\u6b65\u4e0a\u7684\u7279\u5f81\u8868\u793a\u80fd\u529b\u3002\u53cc\u5411\u7ed3\u6784\u5bf9\u4e8eNLP\u4e2d\u7684\u8bb8\u591a\u4efb\u52a1&#xff0c;\u5982\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u3001\u60c5\u611f\u5206\u6790\u3001\u95ee\u7b54\u7cfb\u7edf\u7b49&#xff0c;\u51e0\u4e4e\u662f\u6807\u914d\u3002\u5176\u552f\u4e00\u7684\u7f3a\u70b9\u662f&#xff0c;\u9700\u8981\u5b8c\u6574\u7684\u8f93\u5165\u5e8f\u5217\u624d\u80fd\u5f00\u59cb\u8ba1\u7b97&#xff0c;\u56e0\u6b64\u4e0d\u9002\u7528\u4e8e\u9700\u8981\u5b9e\u65f6\u9884\u6d4b\u7684\u5728\u7ebf\u4efb\u52a1\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>8.4.2 \u6df1\u5ea6&#xff08;\u5806\u53e0&#xff09;RNN&#xff08;Deep\/Stacked RNN&#xff09;<\/h5>\n<ul>\n<li>\n<p>\u589e\u52a0\u201c\u6df1\u5ea6\u201d \u6b63\u5982\u6211\u4eec\u901a\u8fc7\u5806\u53e0\u5377\u79ef\u5c42\u6765\u6784\u5efa\u6df1\u5ea6CNN&#xff0c;\u4ee5\u5b66\u4e60\u5230\u4ece\u4f4e\u7ea7\u5230\u9ad8\u7ea7\u7684\u7a7a\u95f4\u7279\u5f81\u4e00\u6837&#xff0c;\u6211\u4eec\u4e5f\u53ef\u4ee5\u5806\u53e0\u591a\u4e2aRNN\u5c42\u6765\u6784\u5efa\u6df1\u5ea6RNN\u3002<\/p>\n<\/li>\n<li>\n<p>\u5de5\u4f5c\u539f\u7406 \u5176\u5de5\u4f5c\u539f\u7406\u975e\u5e38\u7b80\u5355&#xff1a;<\/p>\n<li>\u7b2c\u4e00\u5c42RNN&#xff08;\u6700\u5e95\u5c42&#xff09;\u63a5\u6536\u539f\u59cb\u7684\u8f93\u5165\u5e8f\u5217\u00a0(x_1, x_2, &#8230;, x_T)&#xff0c;\u5e76\u8f93\u51fa\u4e00\u4e2a\u9690\u85cf\u72b6\u6001\u5e8f\u5217\u00a0(h\u00b9_1, h\u00b9_2, &#8230;, h\u00b9_T)\u3002<\/li>\n<li>\u7b2c\u4e8c\u5c42RNN\u5c06\u7b2c\u4e00\u5c42\u8f93\u51fa\u7684\u9690\u85cf\u72b6\u6001\u5e8f\u5217\u00a0(h\u00b9_1, h\u00b9_2, &#8230;, h\u00b9_T)\u00a0\u4f5c\u4e3a\u81ea\u5df1\u7684\u8f93\u5165\u5e8f\u5217&#xff0c;\u5e76\u8ba1\u7b97\u51fa\u7b2c\u4e8c\u5c42\u7684\u9690\u85cf\u72b6\u6001\u5e8f\u5217\u00a0(h\u00b2_1, h\u00b2_2, &#8230;, h\u00b2_T)\u3002<\/li>\n<li>\u8fd9\u4e2a\u8fc7\u7a0b\u53ef\u4ee5\u4e00\u76f4\u91cd\u590d\u4e0b\u53bb\u3002<\/li>\n<\/li>\n<li>\n<p>\u4f18\u52bf \u5806\u53e0RNN\u5141\u8bb8\u6a21\u578b\u5728\u4e0d\u540c\u7684\u5c42\u6b21\u4e0a\u5b66\u4e60\u7279\u5f81\u3002\u5e95\u5c42\u7684RNN\u53ef\u80fd\u5b66\u4e60\u5230\u4e00\u4e9b\u5c40\u90e8\u7684\u3001\u4f4e\u7ea7\u7684\u5e8f\u5217\u6a21\u5f0f&#xff0c;\u800c\u66f4\u9ad8\u5c42\u7684RNN\u5219\u53ef\u4ee5\u5728\u6b64\u57fa\u7840\u4e0a&#xff0c;\u5b66\u4e60\u5230\u66f4\u957f\u671f\u7684\u3001\u66f4\u62bd\u8c61\u7684\u65f6\u95f4\u4f9d\u8d56\u5173\u7cfb\u3002\u5728\u5b9e\u8df5\u4e2d&#xff0c;\u4e00\u4e2a2\u52304\u5c42\u7684\u6df1\u5ea6RNN&#xff0c;\u901a\u5e38\u80fd\u6bd4\u5355\u5c42RNN\u5e26\u6765\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\u3002<\/p>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>8.5 RNN\u7684\u5e94\u7528&#xff1a;\u4ece\u8bed\u8a00\u5230\u65f6\u95f4<\/h4>\n<p>\u638c\u63e1\u4e86LSTM\u3001GRU\u4ee5\u53ca\u53cc\u5411\u548c\u6df1\u5ea6\u6269\u5c55\u4e4b\u540e&#xff0c;\u6211\u4eec\u5c31\u62e5\u6709\u4e86\u4e00\u5957\u5f3a\u5927\u7684\u5de5\u5177\u96c6&#xff0c;\u53ef\u4ee5\u7528\u6765\u89e3\u51b3\u5404\u79cd\u4e0e\u5e8f\u5217\u76f8\u5173\u7684\u73b0\u5b9e\u95ee\u9898\u3002<\/p>\n<h5>8.5.1 \u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;<\/h5>\n<p>RNN\u662f\u73b0\u4ee3\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u57fa\u77f3\u6280\u672f\u4e4b\u4e00\u3002<\/p>\n<ul>\n<li>\n<p>\u8bcd\u5d4c\u5165&#xff08;Word Embedding&#xff09; \u8ba1\u7b97\u673a\u65e0\u6cd5\u76f4\u63a5\u7406\u89e3\u201c\u732b\u201d\u3001\u201c\u72d7\u201d\u8fd9\u6837\u7684\u8bcd\u8bed\u3002\u5728\u5c06\u6587\u672c\u5582\u7ed9RNN\u4e4b\u524d&#xff0c;\u6211\u4eec\u5fc5\u987b\u5148\u5c06\u8fd9\u4e9b\u79bb\u6563\u7684\u3001\u7b26\u53f7\u5316\u7684\u8bcd\u8bed&#xff0c;\u8f6c\u6362\u4e3a\u5bc6\u96c6\u7684\u3001\u4f4e\u7ef4\u7684\u3001\u8fde\u7eed\u7684\u6d6e\u70b9\u6570\u5411\u91cf\u3002\u8fd9\u4e2a\u5411\u91cf&#xff0c;\u5c31\u662f\u8bcd\u5d4c\u5165\u6216\u8bcd\u5411\u91cf\u3002 \u8bcd\u5d4c\u5165\u4e0d\u4ec5\u4ec5\u662f\u4e00\u4e2a\u6280\u672f\u6b65\u9aa4&#xff0c;\u5b83\u672c\u8eab\u5c31\u8574\u542b\u7740\u8bed\u4e49\u4fe1\u606f\u3002\u901a\u8fc7\u5728\u5927\u91cf\u6587\u672c\u4e0a\u8fdb\u884c\u8bad\u7ec3&#xff08;\u4f8b\u5982\u4f7f\u7528Word2Vec\u6216GloVe\u7b97\u6cd5&#xff09;&#xff0c;\u8bed\u4e49\u4e0a\u76f8\u8fd1\u7684\u8bcd\u8bed&#xff0c;\u5176\u8bcd\u5411\u91cf\u5728\u5411\u91cf\u7a7a\u95f4\u4e2d\u7684\u4f4d\u7f6e\u4e5f\u4f1a\u76f8\u4e92\u9760\u8fd1\u3002<\/p>\n<\/li>\n<li>\n<p>\u6587\u672c\u5206\u7c7b\u4e0e\u60c5\u611f\u5206\u6790 \u8fd9\u662fRNN\u6700\u76f4\u63a5\u7684\u5e94\u7528\u4e4b\u4e00\u3002\u4efb\u52a1\u662f\u5224\u65ad\u4e00\u6574\u6bb5\u6587\u672c\u5c5e\u4e8e\u54ea\u4e2a\u9884\u8bbe\u7684\u7c7b\u522b&#xff08;\u5982\u4f53\u80b2\u3001\u8d22\u7ecf\u3001\u79d1\u6280\u65b0\u95fb&#xff09;&#xff0c;\u6216\u8005\u5176\u8868\u8fbe\u7684\u60c5\u611f\u662f\u6b63\u9762\u7684\u3001\u8d1f\u9762\u7684\u8fd8\u662f\u4e2d\u6027\u7684\u3002 \u5178\u578b\u7684\u505a\u6cd5\u662f&#xff1a;\u5c06\u6587\u672c\u4e2d\u7684\u8bcd\u8bed\u4f9d\u6b21\u8f93\u5165\u4e00\u4e2aRNN&#xff08;\u901a\u5e38\u662fLSTM\u6216GRU&#xff09;\u3002\u5f53\u6574\u4e2a\u5e8f\u5217\u5904\u7406\u5b8c\u6bd5\u540e&#xff0c;\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u6700\u540e\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001 h_T&#xff0c;\u56e0\u4e3a\u5b83\u5728\u7406\u8bba\u4e0a\u5df2\u7ecf\u7f16\u7801\u4e86\u6574\u4e2a\u53e5\u5b50\u7684\u4fe1\u606f\u3002\u6216\u8005&#xff0c;\u6211\u4eec\u4e5f\u53ef\u4ee5\u5c06\u6240\u6709\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001\u8fdb\u884c\u6c60\u5316&#xff08;\u5982\u6700\u5927\u6c60\u5316\u6216\u5e73\u5747\u6c60\u5316&#xff09;&#xff0c;\u6765\u83b7\u5f97\u4e00\u4e2a\u66f4\u9c81\u68d2\u7684\u53e5\u5b50\u8868\u793a\u3002\u6700\u540e&#xff0c;\u5c06\u8fd9\u4e2a\u53e5\u5b50\u8868\u793a\u5411\u91cf\u9001\u5165\u4e00\u4e2a\u6807\u51c6\u7684\u5168\u8fde\u63a5\u5206\u7c7b\u5668&#xff0c;\u5373\u53ef\u5f97\u5230\u6700\u7ec8\u7684\u5206\u7c7b\u7ed3\u679c\u3002<\/p>\n<\/li>\n<li>\n<p>\u5e8f\u5217\u5230\u5e8f\u5217&#xff08;Seq2Seq&#xff09;\u4efb\u52a1 RNN\u5728\u66f4\u590d\u6742\u7684NLP\u4efb\u52a1\u4e2d\u4e5f\u5927\u653e\u5f02\u5f69&#xff0c;\u8fd9\u4e9b\u4efb\u52a1\u7684\u8f93\u5165\u548c\u8f93\u51fa\u90fd\u662f\u53d8\u957f\u7684\u5e8f\u5217\u3002\u4f8b\u5982&#xff1a;<\/p>\n<ul>\n<li>\u673a\u5668\u7ffb\u8bd1&#xff1a;\u8f93\u5165\u4e00\u79cd\u8bed\u8a00\u7684\u53e5\u5b50&#xff0c;\u8f93\u51fa\u53e6\u4e00\u79cd\u8bed\u8a00\u7684\u53e5\u5b50\u3002<\/li>\n<li>\u6587\u672c\u6458\u8981&#xff1a;\u8f93\u5165\u4e00\u7bc7\u957f\u6587\u7ae0&#xff0c;\u8f93\u51fa\u4e00\u4e2a\u7b80\u77ed\u7684\u6458\u8981\u3002<\/li>\n<li>\u5bf9\u8bdd\u7cfb\u7edf&#xff1a;\u8f93\u5165\u7528\u6237\u7684\u95ee\u8bdd&#xff0c;\u8f93\u51fa\u673a\u5668\u4eba\u7684\u56de\u7b54\u3002 \u8fd9\u4e9b\u4efb\u52a1\u901a\u5e38\u4f7f\u7528\u4e00\u79cd\u88ab\u79f0\u4e3a**\u201c\u7f16\u7801\u5668-\u89e3\u7801\u5668\u201d&#xff08;Encoder-Decoder&#xff09;**\u7684RNN\u67b6\u6784&#xff0c;\u5176\u4e2d\u4e00\u4e2aRNN&#xff08;\u7f16\u7801\u5668&#xff09;\u8d1f\u8d23\u5c06\u8f93\u5165\u5e8f\u5217\u538b\u7f29\u6210\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u4e0a\u4e0b\u6587\u5411\u91cf&#xff0c;\u53e6\u4e00\u4e2aRNN&#xff08;\u89e3\u7801\u5668&#xff09;\u5219\u6839\u636e\u8fd9\u4e2a\u4e0a\u4e0b\u6587\u5411\u91cf&#xff0c;\u751f\u6210\u8f93\u51fa\u5e8f\u5217\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>8.5.2 \u65f6\u95f4\u5e8f\u5217\u9884\u6d4b<\/h5>\n<p>\u9664\u4e86\u8bed\u8a00&#xff0c;\u4efb\u4f55\u4ee5\u65f6\u95f4\u4e3a\u8f74\u7684\u6570\u636e&#xff0c;\u90fd\u662fRNN\u7684\u7528\u6b66\u4e4b\u5730\u3002<\/p>\n<ul>\n<li>\n<p>\u4efb\u52a1\u5b9a\u4e49 \u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u7684\u4efb\u52a1\u662f&#xff0c;\u5229\u7528\u8fc7\u53bb\u4e00\u6bb5\u65f6\u95f4\u7684\u5386\u53f2\u89c2\u6d4b\u6570\u636e&#xff0c;\u6765\u9884\u6d4b\u672a\u6765\u4e00\u4e2a\u6216\u591a\u4e2a\u65f6\u95f4\u70b9\u7684\u503c\u3002\u5e94\u7528\u573a\u666f\u6781\u5176\u5e7f\u6cdb&#xff0c;\u5305\u62ec&#xff1a;<\/p>\n<ul>\n<li>\u91d1\u878d&#xff1a;\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u3001\u6c47\u7387\u8d70\u52bf\u3002<\/li>\n<li>\u6c14\u8c61&#xff1a;\u9884\u6d4b\u672a\u6765\u51e0\u5c0f\u65f6\u6216\u51e0\u5929\u7684\u6c14\u6e29\u3001\u964d\u96e8\u91cf\u3002<\/li>\n<li>\u5de5\u4e1a&#xff1a;\u9884\u6d4b\u7f51\u7ad9\u6d41\u91cf\u3001\u7535\u529b\u6d88\u8017\u3001\u4ea7\u54c1\u9500\u91cf\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6a21\u578b\u6784\u5efa \u6784\u5efa\u4e00\u4e2a\u57fa\u4e8eRNN\u7684\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u6a21\u578b&#xff0c;\u901a\u5e38\u9075\u5faa\u4ee5\u4e0b\u6b65\u9aa4&#xff1a;<\/p>\n<li>\u6570\u636e\u51c6\u5907&#xff1a;\u5c06\u4e00\u7ef4\u7684\u65f6\u95f4\u5e8f\u5217\u6570\u636e&#xff0c;\u8f6c\u6362\u6210\u76d1\u7763\u5b66\u4e60\u6240\u9700\u7684(\u8f93\u5165\u5e8f\u5217, \u8f93\u51fa\u76ee\u6807)\u6837\u672c\u5bf9\u3002\u4f8b\u5982&#xff0c;\u4f7f\u7528\u8fc7\u53bb30\u5929\u7684\u6570\u636e\u4f5c\u4e3a\u8f93\u5165\u5e8f\u5217&#xff0c;\u6765\u9884\u6d4b\u7b2c31\u5929\u7684\u503c\u3002<\/li>\n<li>\u6a21\u578b\u642d\u5efa&#xff1a;\u642d\u5efa\u4e00\u4e2a\u4ee5LSTM\u6216GRU\u4e3a\u6838\u5fc3\u7684RNN\u6a21\u578b\u3002\u6a21\u578b\u7684\u8f93\u5165\u662f\u5386\u53f2\u6570\u636e\u5e8f\u5217\u3002<\/li>\n<li>\u9884\u6d4b\u8f93\u51fa&#xff1a;\u6a21\u578b\u7684\u8f93\u51fa\u5c42&#xff0c;\u901a\u5e38\u662f\u4e00\u4e2a\u6216\u591a\u4e2a\u5168\u8fde\u63a5\u5c42&#xff0c;\u7528\u4e8e\u5c06RNN\u7684\u6700\u7ec8\u9690\u85cf\u72b6\u6001&#xff0c;\u56de\u5f52\u5230\u6211\u4eec\u60f3\u8981\u9884\u6d4b\u7684\u5177\u4f53\u6570\u503c\u4e0a\u3002<\/li>\n<\/li>\n<\/ul>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u6df1\u5165\u63a2\u7d22\u4e86\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09;\u90a3\u7406\u89e3\u5e8f\u5217\u7684\u72ec\u7279\u667a\u6167\u3002\u6211\u4eec\u4eceRNN\u6700\u57fa\u7840\u7684\u5faa\u73af\u7ed3\u6784\u548c\u9690\u85cf\u72b6\u6001\u51fa\u53d1&#xff0c;\u76f4\u9762\u4e86\u5176\u5728\u5904\u7406\u957f\u5e8f\u5217\u65f6\u9047\u5230\u7684\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8\u8fd9\u4e00\u6838\u5fc3\u6311\u6218\u3002<\/p>\n<p>\u4e3a\u4e86\u514b\u670d\u8fd9\u4e00\u201c\u963f\u5580\u7409\u65af\u4e4b\u8e35\u201d&#xff0c;\u6211\u4eec\u8be6\u7ec6\u89e3\u6784\u4e86\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;\u90a3\u7cbe\u5de7\u7edd\u4f26\u7684\u4e09\u5927\u95e8\u63a7\u673a\u5236&#xff0c;\u4ee5\u53ca\u5176\u4f18\u96c5\u7684\u7b80\u5316\u7248\u2014\u2014\u95e8\u63a7\u5faa\u73af\u5355\u5143&#xff08;GRU&#xff09;\u3002\u6211\u4eec\u770b\u5230&#xff0c;\u6b63\u662f\u8fd9\u4e9b\u201c\u95e8\u201d\u7684\u8bbe\u8ba1&#xff0c;\u4f7f\u5f97\u4fe1\u606f\u5f97\u4ee5\u5728\u65f6\u95f4\u7684\u201c\u9ad8\u901f\u516c\u8def\u201d\u4e0a\u9009\u62e9\u6027\u5730\u6d41\u52a8&#xff0c;\u4ece\u800c\u5b9e\u73b0\u4e86\u5bf9\u957f\u671f\u4f9d\u8d56\u7684\u6709\u6548\u6355\u6349\u3002<\/p>\n<p>\u63a5\u7740&#xff0c;\u6211\u4eec\u5b66\u4e60\u4e86\u5982\u4f55\u901a\u8fc7\u53cc\u5411\u548c\u6df1\u5ea6&#xff08;\u5806\u53e0&#xff09;\u7684\u8fde\u63a5\u65b9\u5f0f&#xff0c;\u6765\u6784\u5efa\u66f4\u5f3a\u5927\u7684\u5e8f\u5217\u6a21\u578b&#xff0c;\u4f7f\u5176\u80fd\u591f\u540c\u65f6\u611f\u77e5\u8fc7\u53bb\u4e0e\u672a\u6765&#xff0c;\u5e76\u5b66\u4e60\u5230\u66f4\u6df1\u5c42\u6b21\u7684\u62bd\u8c61\u7279\u5f81\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u8fd9\u4e9b\u5f3a\u5927\u7684\u5de5\u5177\u5e94\u7528\u4e8e\u4e24\u5927\u6838\u5fc3\u9886\u57df&#xff1a;\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;\u5982\u6587\u672c\u5206\u7c7b&#xff09;\u548c\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b&#xff0c;\u5c06\u7406\u8bba\u77e5\u8bc6\u8f6c\u5316\u4e3a\u4e86\u89e3\u51b3\u5b9e\u9645\u95ee\u9898\u7684\u80fd\u529b\u3002<\/p>\n<p>\u638c\u63e1\u4e86RNN&#xff0c;\u8bfb\u8005\u4fbf\u62e5\u6709\u4e86\u4e0e\u65f6\u95f4\u5bf9\u8bdd\u3001\u7406\u89e3\u5e8f\u5217\u4e4b\u7f8e\u7684\u94a5\u5319\u3002\u7136\u800c&#xff0c;\u5e8f\u5217\u5efa\u6a21\u7684\u63a2\u7d22\u5e76\u672a\u6b62\u6b65\u4e8e\u6b64\u3002\u5728\u4e0b\u4e00\u7ae0&#xff0c;\u6211\u4eec\u5c06\u89c1\u8bc1\u4e00\u573a\u65b0\u7684\u9769\u547d\u2014\u2014\u6ce8\u610f\u529b\u673a\u5236\u4e0eTransformer\u7684\u5d1b\u8d77&#xff0c;\u770b\u5b83\u4eec\u662f\u5982\u4f55\u98a0\u8986RNN\u7684\u5faa\u73af\u8303\u5f0f&#xff0c;\u5c06\u5e8f\u5217\u5904\u7406\u7684\u6027\u80fd\u4e0e\u6548\u7387\u63a8\u5411\u4e00\u4e2a\u5168\u65b0\u7684\u9ad8\u5cf0\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u4e5d\u7ae0&#xff1a;\u6ce8\u610f\u529b\u673a\u5236\u4e0eTransformer \u2014\u2014 \u73b0\u4ee3NLP\u7684\u57fa\u77f3<\/h3>\n<p>\u6323\u8131\u201c\u5faa\u73af\u201d\u7684\u67b7\u9501<\/p>\n<p>\u5728\u4e0a\u4e00\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u6df1\u5165\u63a2\u7d22\u4e86\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09;\u53ca\u5176\u5f3a\u5927\u7684\u53d8\u4f53LSTM\u548cGRU\u3002\u5b83\u4eec\u901a\u8fc7\u7cbe\u5999\u7684\u5faa\u73af\u7ed3\u6784\u548c\u95e8\u63a7\u673a\u5236&#xff0c;\u8d4b\u4e88\u4e86\u795e\u7ecf\u7f51\u7edc\u5b9d\u8d35\u7684\u201c\u8bb0\u5fc6\u201d\u80fd\u529b&#xff0c;\u5728\u5e8f\u5217\u5efa\u6a21\u9886\u57df\u53d6\u5f97\u4e86\u5de8\u5927\u7684\u6210\u529f\u3002\u7136\u800c&#xff0c;\u8fd9\u79cd\u6210\u529f\u7684\u80cc\u540e&#xff0c;\u4e5f\u9690\u85cf\u7740\u5176\u4e0e\u751f\u4ff1\u6765\u7684\u201c\u67b7\u9501\u201d\u3002<\/p>\n<p>\u9996\u5148&#xff0c;RNN\u987a\u5e8f\u5904\u7406\u7684\u672c\u8d28&#xff0c;\u4f7f\u5176\u5929\u751f\u96be\u4ee5\u5e76\u884c\u8ba1\u7b97\u3002\u8981\u8ba1\u7b97\u7b2c t \u4e2a\u65f6\u95f4\u6b65\u7684\u72b6\u6001&#xff0c;\u5fc5\u987b\u5148\u7b49\u5f85\u7b2c t-1 \u4e2a\u65f6\u95f4\u6b65\u8ba1\u7b97\u5b8c\u6210\u3002\u5728GPU\u5927\u884c\u5176\u9053\u7684\u4eca\u5929&#xff0c;\u8fd9\u79cd\u4e32\u884c\u4f9d\u8d56\u6781\u5927\u5730\u9650\u5236\u4e86\u6a21\u578b\u5728\u957f\u5e8f\u5217\u4e0a\u7684\u8bad\u7ec3\u6548\u7387\u3002\u5176\u6b21&#xff0c;\u65e0\u8bba\u662f\u7b80\u5355RNN\u8fd8\u662f\u590d\u6742\u7684LSTM&#xff0c;\u5b83\u4eec\u5728\u5904\u7406\u5b8c\u4e00\u4e2a\u957f\u5e8f\u5217\u540e&#xff0c;\u90fd\u8bd5\u56fe\u5c06\u5176\u5168\u90e8\u4fe1\u606f\u538b\u7f29\u6210\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u3002\u8fd9\u5c31\u50cf\u8981\u6c42\u4e00\u4f4d\u7ffb\u8bd1\u5728\u542c\u5b8c\u4e00\u6574\u6bb5\u957f\u7bc7\u6f14\u8bb2\u540e&#xff0c;\u4ec5\u51ed\u8111\u4e2d\u7684\u4e00\u4e2a\u6700\u7ec8\u5370\u8c61\u5c31\u5f00\u59cb\u7ffb\u8bd1&#xff0c;\u4e0d\u53ef\u907f\u514d\u5730\u4f1a\u9020\u6210\u4fe1\u606f\u74f6\u9888&#xff0c;\u5c24\u5176\u662f\u5bf9\u4e8e\u90a3\u4e9b\u9065\u8fdc\u7684\u3001\u7ec6\u8282\u7684\u4fe1\u606f\u3002<\/p>\n<p>\u4e3a\u4e86\u6253\u7834\u8fd9\u4e00\u74f6\u9888&#xff0c;\u6ce8\u610f\u529b\u673a\u5236&#xff08;Attention Mechanism&#xff09;\u7684\u66d9\u5149\u521d\u73b0\u3002\u5b83\u6700\u521d\u662f\u4f5c\u4e3a\u5bf9RNN\u4fe1\u606f\u74f6\u9888\u7684\u4e00\u79cd\u8865\u5145\u7597\u6cd5\u800c\u63d0\u51fa\u7684&#xff0c;\u5176\u6838\u5fc3\u601d\u60f3\u662f&#xff1a;\u5728\u751f\u6210\u8f93\u51fa\u7684\u6bcf\u4e00\u6b65&#xff0c;\u4e0d\u518d\u53ea\u4f9d\u8d56\u4e8e\u90a3\u4e2a\u56fa\u5316\u7684\u6700\u7ec8\u5370\u8c61&#xff0c;\u800c\u662f\u5141\u8bb8\u6a21\u578b\u201c\u56de\u5934\u770b\u201d&#xff0c;\u5e76\u52a8\u6001\u5730\u3001\u6709\u9009\u62e9\u5730\u805a\u7126\u4e8e\u8f93\u5165\u5e8f\u5217\u4e2d\u4e0e\u5f53\u524d\u4efb\u52a1\u6700\u76f8\u5173\u7684\u90e8\u5206\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u4e00\u573a\u66f4\u5f7b\u5e95\u7684\u9769\u547d\u6b63\u5728\u915d\u917f\u30022017\u5e74&#xff0c;\u4e00\u7bc7\u540d\u4e3a\u300aAttention Is All You Need\u300b\u7684\u8bba\u6587\u6a2a\u7a7a\u51fa\u4e16&#xff0c;\u5b83\u63d0\u51fa\u4e86\u4e00\u4e2a\u77f3\u7834\u5929\u60ca\u7684\u89c2\u70b9&#xff1a;\u6211\u4eec\u6216\u8bb8\u6839\u672c\u4e0d\u9700\u8981RNN\u7684\u5faa\u73af\u7ed3\u6784\u3002\u6211\u4eec\u53ef\u4ee5\u5b8c\u5168\u629b\u5f03\u201c\u5faa\u73af\u201d&#xff0c;\u4ec5\u4f9d\u9760\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u6765\u6784\u5efa\u4e00\u4e2a\u6027\u80fd\u66f4\u5f3a\u3001\u5e76\u884c\u80fd\u529b\u66f4\u597d\u3001\u80fd\u591f\u5b8c\u7f8e\u6355\u6349\u957f\u8ddd\u79bb\u4f9d\u8d56\u7684\u5e8f\u5217\u6a21\u578b\u3002\u8fd9\u4e2a\u6a21\u578b&#xff0c;\u5c31\u662f\u4f1f\u5927\u7684Transformer\u3002<\/p>\n<p>Transformer\u7684\u8bde\u751f&#xff0c;\u5f7b\u5e95\u91cd\u5851\u4e86\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;\u7684\u7248\u56fe\u3002\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5c06\u6df1\u5165\u8fd9\u573a\u9769\u547d\u7684\u6838\u5fc3\u3002\u6211\u4eec\u5c06\u4ece\u6ce8\u610f\u529b\u673a\u5236\u7684\u672c\u8d28\u601d\u60f3\u51fa\u53d1&#xff0c;\u8be6\u7ec6\u89e3\u6784Transformer\u8fd9\u5ea7\u7531\u81ea\u6ce8\u610f\u529b\u3001\u591a\u5934\u6ce8\u610f\u529b\u548c\u4f4d\u7f6e\u7f16\u7801\u7b49\u6838\u5fc3\u90e8\u4ef6\u6784\u6210\u7684\u5b8f\u4f1f\u5efa\u7b51\u3002\u968f\u540e&#xff0c;\u6211\u4eec\u5c06\u89c1\u8bc1\u7531\u5b83\u50ac\u751f\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b&#xff08;\u5982BERT\u3001GPT&#xff09;\u662f\u5982\u4f55\u5f00\u542f\u4e86NLP\u7684\u5168\u65b0\u8303\u5f0f&#xff0c;\u5e76\u6700\u7ec8\u4e00\u7aa5\u5176\u5f3a\u5927\u7684\u5f71\u54cd\u529b&#xff0c;\u662f\u5982\u4f55\u8de8\u8d8a\u6a21\u6001\u7684\u754c\u9650&#xff0c;\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u6380\u8d77\u65b0\u4e00\u8f6e\u7684\u6d6a\u6f6e\u3002<\/p>\n<p>\u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u4e00\u540c\u6323\u8131\u201c\u5faa\u73af\u201d\u7684\u67b7\u9501&#xff0c;\u8fdb\u5165\u8fd9\u4e2a\u7531\u201c\u6ce8\u610f\u529b\u201d\u4e3b\u5bb0\u7684\u3001\u66f4\u9ad8\u6548\u3001\u66f4\u5f3a\u5927\u7684\u65b0\u4e16\u754c\u3002<\/p>\n<h4>9.1 \u6ce8\u610f\u529b\u673a\u5236&#xff1a;\u4ece\u5173\u6ce8\u201c\u5168\u90e8\u201d\u5230\u805a\u7126\u201c\u91cd\u70b9\u201d<\/h4>\n<p>\u5728\u6df1\u5165Transformer\u7684\u5b8f\u4f1f\u67b6\u6784\u4e4b\u524d&#xff0c;\u6211\u4eec\u5fc5\u987b\u9996\u5148\u7406\u89e3\u5176\u6700\u6838\u5fc3\u7684\u7075\u9b42\u2014\u2014\u6ce8\u610f\u529b&#xff08;Attention&#xff09;\u673a\u5236\u3002\u5b83\u5e76\u975e\u4e00\u4e2a\u5177\u4f53\u7684\u6a21\u578b&#xff0c;\u800c\u662f\u4e00\u79cd\u901a\u7528\u7684\u3001\u5f3a\u5927\u7684\u601d\u60f3\u3002\u8fd9\u79cd\u601d\u60f3\u7684\u5f15\u5165&#xff0c;\u662f\u6df1\u5ea6\u5b66\u4e60\u4ece\u5904\u7406\u9759\u6001\u6570\u636e&#xff0c;\u5230\u771f\u6b63\u7406\u89e3\u590d\u6742\u3001\u52a8\u6001\u4e0a\u4e0b\u6587\u5173\u7cfb\u7684\u4e00\u6b21\u5173\u952e\u8dc3\u8fc1\u3002<\/p>\n<h5>9.1.1 \u6838\u5fc3\u601d\u60f3&#xff1a;\u6e90\u4e8e\u4eba\u7c7b\u7684\u8ba4\u77e5\u76f4\u89c9<\/h5>\n<p>\u4eba\u7c7b\u89c6\u89c9\u6ce8\u610f\u529b\u7684\u6bd4\u55bb \u8981\u7406\u89e3\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u6700\u597d\u7684\u8d77\u70b9\u662f\u53cd\u89c2\u6211\u4eec\u81ea\u8eab\u3002\u60f3\u8c61\u4e00\u4e0b&#xff0c;\u5f53\u60a8\u770b\u5230\u4e0b\u9762\u8fd9\u5f20\u590d\u6742\u7684\u56fe\u7247\u65f6&#xff0c;\u60a8\u7684\u5927\u8111\u662f\u5982\u4f55\u5904\u7406\u5b83\u7684&#xff1f;<\/p>\n<p><img decoding=\"async\" alt=\"\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/01\/20260124113235-6974add367c74.png\" \/><\/p>\n<p>\u60a8\u5e76\u4e0d\u4f1a\u5728\u540c\u4e00\u77ac\u95f4&#xff0c;\u4ee5\u540c\u7b49\u7684\u7cbe\u529b\u53bb\u5904\u7406\u56fe\u7247\u4e2d\u7684\u6bcf\u4e00\u4e2a\u50cf\u7d20\u3002\u76f8\u53cd&#xff0c;\u60a8\u7684\u89c6\u89c9\u7cfb\u7edf\u4f1a\u8fdb\u884c\u4e00\u6b21\u5feb\u901f\u7684\u3001\u65e0\u610f\u8bc6\u7684\u626b\u63cf&#xff0c;\u7136\u540e\u60a8\u7684\u6ce8\u610f\u529b\u4f1a\u8fc5\u901f\u805a\u7126\u4e8e\u67d0\u4e9b\u60a8\u8ba4\u4e3a\u91cd\u8981\u7684\u533a\u57df\u3002\u5982\u679c\u60a8\u60f3\u8fc7\u9a6c\u8def&#xff0c;\u60a8\u7684\u6ce8\u610f\u529b\u4f1a\u96c6\u4e2d\u5728\u6765\u5f80\u7684\u8f66\u8f86\u548c\u4ea4\u901a\u4fe1\u53f7\u706f\u4e0a&#xff1b;\u5982\u679c\u60a8\u60f3\u627e\u4e00\u5bb6\u5496\u5561\u5e97&#xff0c;\u60a8\u7684\u6ce8\u610f\u529b\u5219\u4f1a\u9501\u5b9a\u5728\u5404\u79cd\u5e97\u94fa\u7684\u62db\u724c\u4e0a\u3002<\/p>\n<p>\u8fd9\u79cd\u6709\u9009\u62e9\u6027\u5730\u805a\u7126\u4e8e\u5173\u952e\u4fe1\u606f&#xff0c;\u800c\u5ffd\u7565\u6b21\u8981\u4fe1\u606f\u7684\u80fd\u529b&#xff0c;\u5c31\u662f\u4eba\u7c7b\u8ba4\u77e5\u7cfb\u7edf\u9ad8\u6548\u5de5\u4f5c\u7684\u6838\u5fc3\u79d8\u8bc0\u3002\u6211\u4eec\u7684\u5927\u8111\u8d44\u6e90\u662f\u6709\u9650\u7684&#xff0c;\u6ce8\u610f\u529b\u673a\u5236\u5e2e\u52a9\u6211\u4eec\u5c06\u8fd9\u4e9b\u5b9d\u8d35\u7684\u8d44\u6e90&#xff0c;\u52a8\u6001\u5730\u5206\u914d\u5230\u6700\u9700\u8981\u7684\u5730\u65b9\u3002<\/p>\n<p>\u7a81\u7834\u4fe1\u606f\u74f6\u74f6\u9888 \u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u56de\u5230\u4e0a\u4e00\u7ae0\u8ba8\u8bba\u7684RNN\u53ca\u5176\u5728\u5904\u7406\u957f\u5e8f\u5217\u65f6\u7684\u56f0\u5883\u3002\u4e00\u4e2a\u7ecf\u5178\u7684\u57fa\u4e8eRNN\u7684\u201c\u7f16\u7801\u5668-\u89e3\u7801\u5668\u201d\u6a21\u578b\u5728\u8fdb\u884c\u673a\u5668\u7ffb\u8bd1\u65f6&#xff0c;\u7f16\u7801\u5668&#xff08;Encoder&#xff09;\u4f1a\u8bfb\u53d6\u6574\u4e2a\u6e90\u8bed\u8a00\u53e5\u5b50&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a20\u4e2a\u8bcd\u7684\u5fb7\u8bed\u53e5\u5b50&#xff09;&#xff0c;\u5e76\u5c06\u5176\u6240\u6709\u4fe1\u606f\u5f3a\u884c\u538b\u7f29\u6210\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u4e0a\u4e0b\u6587\u5411\u91cf&#xff08;Context Vector&#xff09;\u3002\u7136\u540e&#xff0c;\u89e3\u7801\u5668&#xff08;Decoder&#xff09;\u5fc5\u987b\u4ec5\u51ed\u8fd9\u4e2a\u6d53\u7f29\u540e\u7684\u5411\u91cf&#xff0c;\u53bb\u751f\u6210\u76ee\u6807\u8bed\u8a00\u7684\u53e5\u5b50&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a22\u4e2a\u8bcd\u7684\u82f1\u8bed\u53e5\u5b50&#xff09;\u3002<\/p>\n<p>\u8fd9\u5c31\u50cf\u8981\u6c42\u4e00\u4f4d\u540c\u58f0\u4f20\u8bd1\u5458&#xff0c;\u5fc5\u987b\u5728\u542c\u5b8c\u4e00\u6574\u6bb5\u957f\u8fbe\u4e00\u5206\u949f\u7684\u6f14\u8bb2\u540e&#xff0c;\u624d\u80fd\u5f00\u59cb\u7ffb\u8bd1&#xff0c;\u5e76\u4e14\u5728\u7ffb\u8bd1\u8fc7\u7a0b\u4e2d&#xff0c;\u4e0d\u5141\u8bb8\u4ed6\u518d\u67e5\u9605\u4efb\u4f55\u7b14\u8bb0\u3002\u8fd9\u663e\u7136\u662f\u53cd\u76f4\u89c9\u4e14\u4f4e\u6548\u7684\u3002\u65e0\u8bba\u8fd9\u4e2a\u4e0a\u4e0b\u6587\u5411\u91cf\u7684\u7ef4\u5ea6\u6709\u591a\u5927&#xff0c;\u5b83\u90fd\u4e0d\u53ef\u907f\u514d\u5730\u4f1a\u6210\u4e3a\u4e00\u4e2a\u4fe1\u606f\u74f6\u9888&#xff0c;\u5c24\u5176\u5bf9\u4e8e\u957f\u53e5\u5b50\u800c\u8a00&#xff0c;\u65e9\u671f\u7684\u3001\u7ec6\u8282\u7684\u4fe1\u606f\u5f88\u5bb9\u6613\u5728\u53cd\u590d\u7684\u538b\u7f29\u4e2d\u88ab\u201c\u9057\u5fd8\u201d\u3002<\/p>\n<p>\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u6b63\u662f\u4e3a\u4e86\u6253\u7834\u8fd9\u4e2a\u74f6\u9888\u800c\u8bbe\u8ba1\u7684\u3002<\/p>\n<p>\u5b83\u7684\u6838\u5fc3\u601d\u60f3&#xff0c;\u5c31\u662f\u6a21\u4eff\u4eba\u7c7b\u7684\u8ba4\u77e5\u884c\u4e3a&#xff0c;\u8d4b\u4e88\u89e3\u7801\u5668\u4e00\u79cd\u201c\u56de\u5934\u770b\u201d\u7684\u80fd\u529b\u3002\u5728\u751f\u6210\u76ee\u6807\u53e5\u5b50\u7684\u6bcf\u4e00\u4e2a\u8bcd\u65f6&#xff0c;\u89e3\u7801\u5668\u4e0d\u518d\u53ea\u4f9d\u8d56\u4e8e\u90a3\u4e2a\u5355\u4e00\u7684\u3001\u56fa\u5316\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u3002\u76f8\u53cd&#xff0c;\u5b83\u4f1a\u5f97\u5230\u4e00\u4e2a\u52a8\u6001\u7684\u3001\u4e3a\u5f53\u524d\u6b65\u9aa4\u91cf\u8eab\u5b9a\u5236\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u3002\u8fd9\u4e2a\u52a8\u6001\u5411\u91cf&#xff0c;\u662f\u901a\u8fc7\u5bf9\u6e90\u8bed\u8a00\u53e5\u5b50\u4e2d\u6240\u6709\u8bcd\u7684\u8868\u793a\u8fdb\u884c\u4e00\u6b21\u52a0\u6743\u6c42\u548c\u800c\u5f97\u5230\u7684\u3002<\/p>\n<p>\u800c\u8fd9\u4e2a\u201c\u6743\u201d&#xff0c;\u5c31\u662f\u6ce8\u610f\u529b\u6743\u91cd&#xff08;Attention Weights&#xff09;\u3002\u5b83\u4ee3\u8868\u4e86\u5728\u751f\u6210\u5f53\u524d\u76ee\u6807\u8bcd\u65f6&#xff0c;\u5e94\u8be5\u5bf9\u6e90\u8bed\u8a00\u4e2d\u7684\u54ea\u4e2a\u8bcd&#xff0c;\u6295\u5165\u591a\u5927\u7684\u201c\u6ce8\u610f\u529b\u201d\u3002\u4f8b\u5982&#xff0c;\u5728\u7ffb\u8bd1\u5fb7\u8bed\u53e5\u5b50 &#034;Ich bin ein Student&#034; \u4e3a\u82f1\u8bed\u65f6&#xff0c;\u5f53\u89e3\u7801\u5668\u51c6\u5907\u751f\u6210 &#034;student&#034; \u8fd9\u4e2a\u8bcd\u65f6&#xff0c;\u6ce8\u610f\u529b\u673a\u5236\u4f1a\u4f7f\u5f97\u6e90\u8bed\u53e5\u5b50\u4e2d\u7684 &#034;Student&#034; \u8fd9\u4e2a\u8bcd\u83b7\u5f97\u6781\u9ad8\u7684\u6ce8\u610f\u529b\u6743\u91cd&#xff0c;\u800c\u5176\u4ed6\u8bcd\u7684\u6743\u91cd\u5219\u76f8\u5bf9\u8f83\u4f4e\u3002<\/p>\n<p>\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u6a21\u578b\u4e0d\u518d\u9700\u8981\u5c06\u6240\u6709\u4fe1\u606f\u90fd\u786c\u585e\u8fdb\u4e00\u4e2a\u72ed\u5c0f\u7684\u74f6\u9888\u91cc\u3002\u5b83\u62e5\u6709\u4e86\u76f4\u63a5\u8bbf\u95ee\u6240\u6709\u8f93\u5165\u4fe1\u606f\u6e90\u7684\u80fd\u529b&#xff0c;\u5e76\u5b66\u4f1a\u4e86\u5982\u4f55\u6839\u636e\u5f53\u524d\u7684\u9700\u6c42&#xff0c;\u53bb\u52a8\u6001\u5730\u3001\u6709\u9009\u62e9\u5730\u805a\u7126\u4e8e\u91cd\u70b9\u3002<\/p>\n<h5>9.1.2 Query, Key, Value&#xff1a;\u6ce8\u610f\u529b\u7684\u4e09\u8981\u7d20<\/h5>\n<p>\u4e3a\u4e86\u5c06\u8fd9\u79cd\u76f4\u89c9\u8f6c\u5316\u4e3a\u5177\u4f53\u7684\u6570\u5b66\u6a21\u578b&#xff0c;\u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u975e\u5e38\u4f18\u7f8e\u4e14\u901a\u7528\u7684\u8ba1\u7b97\u6846\u67b6\u3002\u8fd9\u4e2a\u6846\u67b6\u5c06\u6ce8\u610f\u529b\u673a\u5236\u7684\u8ba1\u7b97\u8fc7\u7a0b&#xff0c;\u7c7b\u6bd4\u4e3a\u4e00\u6b21\u6570\u636e\u5e93\u7684\u67e5\u8be2&#xff08;Query&#xff09;\u64cd\u4f5c\u3002\u8fd9\u4e2a\u7c7b\u6bd4\u975e\u5e38\u6df1\u523b&#xff0c;\u662f\u7406\u89e3\u6240\u6709\u6ce8\u610f\u529b\u53d8\u4f53\u7684\u5173\u952e\u3002<\/p>\n<p>\u5728\u8fd9\u4e2a\u6846\u67b6\u4e2d&#xff0c;\u6709\u4e09\u4e2a\u6838\u5fc3\u7684\u89d2\u8272&#xff1a;Query (Q), Key (K), \u548c Value (V)\u3002<\/p>\n<p>Query (Q)&#xff1a;\u67e5\u8be2 Query \u4ee3\u8868\u4e86\u6211\u4eec\u5f53\u524d\u7684\u9700\u6c42\u3001\u610f\u56fe\u6216\u95ee\u9898\u3002\u5b83\u662f\u9a71\u52a8\u6574\u4e2a\u6ce8\u610f\u529b\u8fc7\u7a0b\u7684\u201c\u4e3b\u52a8\u65b9\u201d\u3002\u5728\u4e0d\u540c\u7684\u573a\u666f\u4e0b&#xff0c;Query\u53ef\u4ee5\u662f\u6211\u4eec\u60f3\u8981\u89e3\u7b54\u7684\u95ee\u9898&#xff0c;\u662f\u6211\u4eec\u6b63\u5728\u5904\u7406\u7684\u67d0\u4e2a\u7279\u5b9a\u5bf9\u8c61&#xff0c;\u6216\u8005\u662f\u6211\u4eec\u4e0b\u4e00\u6b65\u884c\u52a8\u7684\u610f\u56fe\u3002<\/p>\n<ul>\n<li>\u5728\u7ecf\u5178\u7684Encoder-Decoder\u6a21\u578b\u4e2d&#xff1a;\u5f53\u89e3\u7801\u5668\u51c6\u5907\u751f\u6210\u7b2c\u00a0i\u00a0\u4e2a\u76ee\u6807\u8bcd\u65f6&#xff0c;\u5176\u4e0a\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001\u00a0h_{i-1}\u00a0\u5c31\u626e\u6f14\u4e86Query\u7684\u89d2\u8272\u3002\u8fd9\u4e2a\u00a0h_{i-1}\u00a0\u8574\u542b\u4e86\u201c\u6211\u5df2\u7ecf\u751f\u6210\u4e86\u8fd9\u4e9b\u8bcd&#xff0c;\u73b0\u5728\u6211\u9700\u8981\u4ec0\u4e48\u4fe1\u606f\u6765\u751f\u6210\u4e0b\u4e00\u4e2a\u8bcd&#xff1f;\u201d\u8fd9\u6837\u7684\u67e5\u8be2\u610f\u56fe\u3002<\/li>\n<\/ul>\n<p>Key (K)&#xff1a;\u952e Key \u4ee3\u8868\u4e86\u4fe1\u606f\u5e93\u4e2d&#xff0c;\u6bcf\u4e00\u6761\u53ef\u4f9b\u67e5\u8be2\u7684\u8bb0\u5f55\u6240\u62e5\u6709\u7684**\u201c\u6807\u7b7e\u201d\u6216\u201c\u7d22\u5f15\u201d\u3002\u5b83\u7684\u4e3b\u8981\u4f5c\u7528&#xff0c;\u662f\u4e0eQuery\u8fdb\u884c\u5339\u914d\u548c\u6bd4\u8f83**&#xff0c;\u4ee5\u8861\u91cf\u5b83\u6240\u4ee3\u8868\u7684\u90a3\u6761\u4fe1\u606f&#xff0c;\u4e0e\u5f53\u524d\u67e5\u8be2\u9700\u6c42\u7684\u76f8\u5173\u6027\u6216\u76f8\u4f3c\u5ea6\u3002<\/p>\n<ul>\n<li>\u5728\u7ecf\u5178\u7684Encoder-Decoder\u6a21\u578b\u4e2d&#xff1a;\u7f16\u7801\u5668\u5728\u5904\u7406\u5b8c\u6574\u4e2a\u6e90\u8bed\u8a00\u53e5\u5b50\u540e&#xff0c;\u6240\u4ea7\u751f\u7684\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001\u00a0(h_1, h_2, &#8230;, h_n)&#xff0c;\u5c31\u626e\u6f14\u4e86Key\u7684\u89d2\u8272\u3002\u6bcf\u4e00\u4e2a\u00a0h_j\u00a0\u90fd\u662f\u4e00\u4e2a\u201c\u952e\u201d&#xff0c;\u4ee3\u8868\u4e86\u6e90\u53e5\u4e2d\u7b2c\u00a0j\u00a0\u4e2a\u8bcd\u53ca\u5176\u4e0a\u4e0b\u6587\u7684\u4fe1\u606f\u6807\u7b7e\u3002<\/li>\n<\/ul>\n<p>Value (V)&#xff1a;\u503c Value \u4ee3\u8868\u4e86\u4fe1\u606f\u5e93\u4e2d&#xff0c;\u6bcf\u4e00\u6761\u8bb0\u5f55\u6240\u5305\u542b\u7684\u771f\u6b63\u7684\u4fe1\u606f\u5185\u5bb9\u3002\u5b83\u662f\u6211\u4eec\u6700\u7ec8\u60f3\u8981\u63d0\u53d6\u548c\u5229\u7528\u7684\u4e1c\u897f\u3002\u901a\u5e38\u60c5\u51b5\u4e0b&#xff0c;Key\u548cValue\u662f\u6210\u5bf9\u51fa\u73b0\u7684&#xff0c;Key\u662f\u4fe1\u606f\u7684\u7d22\u5f15&#xff0c;Value\u662f\u4fe1\u606f\u672c\u8eab\u3002\u5728\u5f88\u591a\u6a21\u578b\u4e2d&#xff0c;\u4e3a\u4e86\u7b80\u5316&#xff0c;Key\u548cValue\u53ef\u4ee5\u662f\u540c\u4e00\u4e2a\u4e1c\u897f\u3002<\/p>\n<ul>\n<li>\u5728\u7ecf\u5178\u7684Encoder-Decoder\u6a21\u578b\u4e2d&#xff1a;Value\u901a\u5e38\u5c31\u7b49\u4e8eKey&#xff0c;\u5373\u7f16\u7801\u5668\u7684\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u9690\u85cf\u72b6\u6001\u00a0(h_1, h_2, &#8230;, h_n)\u3002\u5f53\u6211\u4eec\u901a\u8fc7Query\u548cKey\u7684\u5339\u914d&#xff0c;\u786e\u5b9a\u4e86\u5e94\u8be5\u5bf9\u6e90\u53e5\u4e2d\u7b2c\u00a0j\u00a0\u4e2a\u8bcd\u6295\u5165\u9ad8\u5ea6\u5173\u6ce8\u65f6&#xff0c;\u6211\u4eec\u771f\u6b63\u60f3\u8981\u63d0\u53d6\u7684&#xff0c;\u5c31\u662f\u8fd9\u4e2a\u8bcd\u7684\u8868\u793a\u00a0h_j\u00a0\u672c\u8eab\u3002<\/li>\n<\/ul>\n<p>\u8ba1\u7b97\u4e09\u90e8\u66f2 \u6709\u4e86Q, K, V\u8fd9\u4e09\u4e2a\u8981\u7d20&#xff0c;\u6ce8\u610f\u529b\u7684\u8ba1\u7b97\u8fc7\u7a0b\u5c31\u53ef\u4ee5\u6e05\u6670\u5730\u5206\u4e3a\u4e09\u6b65&#xff1a;<\/p>\n<li>\n<p>\u7b2c\u4e00\u6b65&#xff1a;\u8ba1\u7b97\u76f8\u4f3c\u5ea6&#xff08;Similarity Calculation&#xff09; \u8fd9\u4e00\u6b65\u7684\u6838\u5fc3&#xff0c;\u662f\u7528\u5f53\u524d\u7684Query&#xff0c;\u53bb\u548c\u4fe1\u606f\u5e93\u4e2d\u6240\u6709\u7684Key\u8fdb\u884c\u9010\u4e00\u7684\u5339\u914d&#xff0c;\u8ba1\u7b97\u51fa\u4e00\u4e2a\u6ce8\u610f\u529b\u5206\u6570&#xff08;Attention Score&#xff09;\u3002\u8fd9\u4e2a\u5206\u6570\u8861\u91cf\u4e86Query\u548c\u6bcf\u4e2aKey\u7684\u76f8\u4f3c\u6216\u76f8\u5173\u7a0b\u5ea6\u3002\u8ba1\u7b97\u76f8\u4f3c\u5ea6\u7684\u65b9\u6cd5\u6709\u5f88\u591a&#xff0c;\u6700\u5e38\u89c1\u7684\u6709&#xff1a;<\/p>\n<ul>\n<li>\u70b9\u79ef&#xff08;Dot-Product&#xff09;&#xff1a;score(Q, K_i) &#061; Q^T * K_i\u3002\u8fd9\u662f\u6700\u7b80\u5355\u3001\u6700\u9ad8\u6548\u7684\u65b9\u6cd5&#xff0c;\u4e5f\u662fTransformer\u91c7\u7528\u7684\u6838\u5fc3\u65b9\u6cd5\u3002<\/li>\n<li>\u52a0\u6027\u6ce8\u610f\u529b&#xff08;Additive Attention&#xff09;&#xff1a;score(Q, K_i) &#061; v^T * tanh(W_q*Q &#043; W_k*K_i)\u3002\u5b83\u901a\u8fc7\u4e00\u4e2a\u5e26\u6fc0\u6d3b\u51fd\u6570\u7684\u524d\u9988\u7f51\u7edc\u6765\u8ba1\u7b97\u5206\u6570&#xff0c;\u7406\u8bba\u4e0a\u8868\u8fbe\u80fd\u529b\u66f4\u5f3a&#xff0c;\u4f46\u8ba1\u7b97\u66f4\u590d\u6742\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7b2c\u4e8c\u6b65&#xff1a;\u5f52\u4e00\u5316\u5206\u6570&#xff08;Score Normalization&#xff09; \u6211\u4eec\u5f97\u5230\u4e86\u4e00\u7cfb\u5217\u7684\u6ce8\u610f\u529b\u5206\u6570&#xff0c;\u4f46\u8fd9\u4e9b\u5206\u6570\u7684\u5927\u5c0f\u4e0d\u4e00&#xff0c;\u96be\u4ee5\u76f4\u63a5\u7528\u4f5c\u6743\u91cd\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u9700\u8981\u4f7f\u7528Softmax\u51fd\u6570&#xff0c;\u5c06\u8fd9\u4e9b\u539f\u59cb\u7684\u5206\u6570&#xff0c;\u8f6c\u6362\u6210\u4e00\u7ec4\u548c\u4e3a1\u3001\u975e\u8d1f\u7684\u6ce8\u610f\u529b\u6743\u91cd&#xff08;Attention Weights&#xff09;\u03b1_i\u3002 \u03b1_i &#061; softmax(score_i) &#061; exp(score_i) \/ \u03a3_j(exp(score_j)) \u7ecf\u8fc7Softmax\u5904\u7406\u540e&#xff0c;\u6bcf\u4e00\u4e2a\u03b1_i\u90fd\u4ee3\u8868\u4e86\u5728\u5f53\u524dQuery\u4e0b&#xff0c;\u7b2c i \u4e2aValue\u6240\u5e94\u5360\u7684\u201c\u6ce8\u610f\u529b\u767e\u5206\u6bd4\u201d\u3002\u6240\u6709\u6743\u91cd\u52a0\u8d77\u6765\u7b49\u4e8e1&#xff0c;\u5c31\u50cf\u662f\u6211\u4eec\u603b\u5171100%\u7684\u6ce8\u610f\u529b&#xff0c;\u88ab\u5206\u914d\u5230\u4e86\u4e0d\u540c\u7684\u4fe1\u606f\u6e90\u4e0a\u3002<\/p>\n<\/li>\n<li>\n<p>\u7b2c\u4e09\u6b65&#xff1a;\u52a0\u6743\u6c42\u548c&#xff08;Weighted Sum&#xff09; \u6700\u540e\u4e00\u6b65&#xff0c;\u5c31\u662f\u7528\u4e0a\u4e00\u6b65\u5f97\u5230\u7684\u6ce8\u610f\u529b\u6743\u91cd \u03b1_i&#xff0c;\u53bb\u5bf9\u6240\u6709\u7684Value\u8fdb\u884c\u52a0\u6743\u6c42\u548c&#xff0c;\u5f97\u5230\u6700\u7ec8\u7684\u8f93\u51fa\u3002\u8fd9\u4e2a\u8f93\u51fa&#xff0c;\u901a\u5e38\u88ab\u79f0\u4e3a\u4e0a\u4e0b\u6587\u5411\u91cf&#xff08;Context Vector&#xff09;\u6216\u6ce8\u610f\u529b\u8f93\u51fa&#xff08;Attention Output&#xff09;\u3002 Context &#061; \u03a3_i(\u03b1_i * V_i) \u8fd9\u4e2a\u6700\u7ec8\u7684Context\u5411\u91cf&#xff0c;\u662f\u4e00\u4e2a\u9ad8\u5ea6\u6d53\u7f29\u7684\u3001\u4e3a\u5f53\u524dQuery\u91cf\u8eab\u5b9a\u5236\u7684\u4fe1\u606f\u7cbe\u534e\u3002\u5b83\u4e0d\u518d\u662f\u6e90\u5e8f\u5217\u6240\u6709\u4fe1\u606f\u7684\u7b80\u5355\u5e73\u5747\u6216\u7c97\u66b4\u538b\u7f29&#xff0c;\u800c\u662f\u6839\u636e\u5f53\u524d\u9700\u6c42&#xff0c;\u4ece\u6240\u6709Value\u4e2d\u667a\u80fd\u5730\u3001\u6709\u4fa7\u91cd\u5730\u63d0\u53d6\u51fa\u7684\u4fe1\u606f\u7684\u52a8\u6001\u878d\u5408\u3002<\/p>\n<\/li>\n<p>\u901a\u8fc7\u8fd9\u4f18\u7f8e\u7684\u4e09\u90e8\u66f2&#xff0c;\u6ce8\u610f\u529b\u673a\u5236\u6323\u8131\u4e86\u4fe1\u606f\u74f6\u9888\u7684\u675f chiffres&#xff0c;\u4e3a\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u8d4b\u4e88\u4e86\u805a\u7126\u4e8e\u91cd\u70b9\u7684\u5f3a\u5927\u80fd\u529b\u3002\u5b83\u4e0d\u4ec5\u662f\u540e\u7eedTransformer\u67b6\u6784\u7684\u57fa\u77f3&#xff0c;\u5176\u601d\u60f3\u672c\u8eab\u4e5f\u5df2\u7ecf\u6e17\u900f\u5230\u8ba1\u7b97\u673a\u89c6\u89c9\u3001\u63a8\u8350\u7cfb\u7edf\u7b49\u4f17\u591a\u9886\u57df&#xff0c;\u6210\u4e3a\u73b0\u4ee3AI\u5de5\u5177\u7bb1\u4e2d\u4e0d\u53ef\u6216\u7f3a\u7684\u4e00\u4ef6\u6cd5\u5b9d\u3002<\/p>\n<h4>9.2 Transformer\u67b6\u6784\u8be6\u89e3&#xff1a;\u6ce8\u610f\u529b\u5c31\u662f\u4e00\u5207<\/h4>\n<p>2017\u5e74&#xff0c;Google\u7684\u7814\u7a76\u8005\u4eec\u53d1\u8868\u4e86\u4e00\u7bc7\u540d\u4e3a\u300aAttention Is All You Need\u300b\u7684\u8bba\u6587&#xff0c;\u8fd9\u7bc7\u8bba\u6587\u7684\u6807\u9898\u672c\u8eab&#xff0c;\u5c31\u662f\u4e00\u53e5\u77f3\u7834\u5929\u60ca\u7684\u5ba3\u8a00\u3002\u5b83\u6240\u63d0\u51fa\u7684Transformer\u6a21\u578b&#xff0c;\u5f7b\u5e95\u98a0\u8986\u4e86\u5e8f\u5217\u5904\u7406\u7684\u4f20\u7edf\u8303\u5f0f&#xff0c;\u5ba3\u544a\u4e86\u4e00\u4e2a\u65b0\u65f6\u4ee3\u7684\u6765\u4e34\u3002<\/p>\n<h5>9.2.1 \u544a\u522bRNN&#xff1a;\u62e5\u62b1\u5e76\u884c\u5316\u7684\u54f2\u5b66\u9769\u547d<\/h5>\n<p>\u6838\u5fc3\u53d8\u9769&#xff1a;\u4ece\u201c\u65f6\u95f4\u94fe\u201d\u5230\u201c\u5173\u7cfb\u7f51\u201d RNN\u7684\u54f2\u5b66&#xff0c;\u662f\u65f6\u95f4\u7684\u7ebf\u6027\u53d9\u4e8b\u3002\u5b83\u50cf\u4e00\u4f4d\u53f2\u5b98&#xff0c;\u6cbf\u7740\u65f6\u95f4\u8f74\u4e00\u683c\u4e00\u683c\u5730\u524d\u8fdb&#xff0c;\u5c06\u8fc7\u53bb\u7684\u4fe1\u606f\u4e0d\u65ad\u7d2f\u79ef\u3001\u63d0\u70bc&#xff0c;\u5f62\u6210\u5bf9\u5f53\u4e0b\u7684\u7406\u89e3\u3002\u8fd9\u79cd\u6a21\u5f0f\u7684\u672c\u8d28&#xff0c;\u662f\u4e00\u6761\u65f6\u95f4\u4f9d\u8d56\u7684\u94fe\u6761\u3002<\/p>\n<p>Transformer\u7684\u54f2\u5b66&#xff0c;\u5219\u662f\u7a7a\u95f4\u7684\u5173\u7cfb\u7f51\u7edc\u3002\u5b83\u5f7b\u5e95\u65a9\u65ad\u4e86\u65f6\u95f4\u7684\u9501\u94fe&#xff0c;\u5c06\u6574\u4e2a\u5e8f\u5217\u89c6\u4e3a\u4e00\u4e2a\u540c\u65f6\u5b58\u5728\u7684\u96c6\u5408\u3002\u5b83\u8ba4\u4e3a&#xff0c;\u5e8f\u5217\u4e2d\u5143\u7d20\u7684\u610f\u4e49&#xff0c;\u5e76\u975e\u4ec5\u4ec5\u7531\u5176\u201c\u524d\u8eab\u201d\u51b3\u5b9a&#xff0c;\u800c\u662f\u7531\u5176\u4e0e\u96c6\u5408\u4e2d\u6240\u6709\u5176\u4ed6\u5143\u7d20\u4e4b\u95f4\u590d\u6742\u3001\u591a\u7ef4\u7684\u5173\u7cfb\u6240\u5171\u540c\u5b9a\u4e49\u3002\u5b83\u4e0d\u518d\u95ee\u201c\u63a5\u4e0b\u6765\u4f1a\u53d1\u751f\u4ec0\u4e48&#xff1f;\u201d&#xff0c;\u800c\u662f\u95ee\u201c\u5728\u8fd9\u4e2a\u6574\u4f53\u4e2d&#xff0c;\u6bcf\u4e2a\u5143\u7d20\u4e0e\u5176\u4ed6\u6240\u6709\u5143\u7d20\u662f\u5982\u4f55\u76f8\u4e92\u5173\u8054\u3001\u76f8\u4e92\u5b9a\u4e49\u7684&#xff1f;\u201d<\/p>\n<p>\u5e76\u884c\u4f18\u52bf&#xff1a;\u8ba1\u7b97\u6548\u7387\u7684\u89e3\u653e \u8fd9\u573a\u54f2\u5b66\u9769\u547d&#xff0c;\u76f4\u63a5\u5e26\u6765\u4e86\u8ba1\u7b97\u6548\u7387\u4e0a\u7684\u5de8\u5927\u89e3\u653e\u3002\u7531\u4e8eTransformer\u6452\u5f03\u4e86RNN\u7684\u5faa\u73af\u4f9d\u8d56&#xff0c;\u5bf9\u4e8e\u4e00\u4e2a\u8f93\u5165\u5e8f\u5217&#xff0c;\u6a21\u578b\u5185\u90e8\u7684\u8ba1\u7b97&#xff08;\u5c24\u5176\u662f\u6838\u5fc3\u7684\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97&#xff09;\u53ef\u4ee5\u5728\u6240\u6709\u4f4d\u7f6e\u4e0a\u540c\u65f6\u5c55\u5f00\u3002\u8fd9\u610f\u5473\u7740&#xff0c;\u65e0\u8bba\u5e8f\u5217\u6709\u591a\u957f&#xff0c;\u7406\u8bba\u4e0a\u6211\u4eec\u90fd\u53ef\u4ee5\u5728\u4e00\u4e2a\u8ba1\u7b97\u6b65\u9aa4\u5185&#xff0c;\u5b8c\u6210\u6240\u6709\u5143\u7d20\u4e4b\u95f4\u5173\u7cfb\u7684\u5efa\u6a21\u3002\u8fd9\u79cd\u5185\u5728\u7684\u5e76\u884c\u6027&#xff0c;\u5b8c\u7f8e\u5730\u5951\u5408\u4e86\u73b0\u4ee3GPU&#xff08;\u56fe\u5f62\u5904\u7406\u5668&#xff09;\u5927\u89c4\u6a21\u5e76\u884c\u8ba1\u7b97\u7684\u67b6\u6784&#xff0c;\u4f7f\u5f97\u8bad\u7ec3\u66f4\u5927\u3001\u66f4\u6df1\u7684\u6a21\u578b&#xff0c;\u5904\u7406\u66f4\u957f\u7684\u5e8f\u5217\u6210\u4e3a\u53ef\u80fd&#xff0c;\u6781\u5927\u5730\u7f29\u77ed\u4e86\u5b9e\u9a8c\u5468\u671f&#xff0c;\u52a0\u901f\u4e86\u6574\u4e2a\u9886\u57df\u7684\u8fed\u4ee3\u3002<\/p>\n<h5>9.2.2 \u81ea\u6ce8\u610f\u529b\u673a\u5236&#xff08;Self-Attention&#xff09;&#xff1a;\u5e8f\u5217\u5185\u90e8\u7684\u6df1\u5ea6\u201c\u51dd\u89c6\u201d<\/h5>\n<p>\u81ea\u6ce8\u610f\u529b&#xff0c;\u662fTransformer\u8fd9\u5ea7\u795e\u6bbf\u7684\u4e2d\u592e\u652f\u67f1&#xff0c;\u662f\u5176\u6240\u6709\u529b\u91cf\u7684\u6e90\u6cc9\u3002\u5b83\u662f\u6ce8\u610f\u529b\u673a\u5236\u7684\u4e00\u79cd\u7279\u6b8a\u5f62\u5f0f&#xff0c;\u4e5f\u662f\u5176\u6700\u5f3a\u5927\u7684\u5e94\u7528\u3002<\/p>\n<p>\u201c\u81ea\u5df1\u5bf9\u81ea\u5df1\u201d\u7684\u6ce8\u610f\u529b&#xff1a;\u6982\u5ff5\u7684\u5347\u534e \u5728\u7ecf\u5178\u7684\u6ce8\u610f\u529b\u673a\u5236\u4e2d&#xff0c;Query\u6765\u81ea\u4e00\u4e2a\u5730\u65b9&#xff08;\u5982\u89e3\u7801\u5668&#xff09;&#xff0c;\u800cKey\u548cValue\u6765\u81ea\u53e6\u4e00\u4e2a\u5730\u65b9&#xff08;\u5982\u7f16\u7801\u5668&#xff09;\u3002\u5728\u81ea\u6ce8\u610f\u529b\u4e2d&#xff0c;\u8fd9\u79cd\u533a\u5206\u6d88\u5931\u4e86\u3002Query, Key, \u548c Value \u8fd9\u4e09\u4e2a\u89d2\u8272&#xff0c;\u90fd\u6765\u6e90\u4e8e\u540c\u4e00\u4e2a\u8f93\u5165\u5e8f\u5217\u672c\u8eab\u3002<\/p>\n<p>\u8fd9\u610f\u5473\u7740\u4ec0\u4e48&#xff1f;\u8fd9\u610f\u5473\u7740\u5e8f\u5217\u4e2d\u7684\u6bcf\u4e00\u4e2a\u5143\u7d20&#xff0c;\u90fd\u4f1a\u8f6e\u6d41\u626e\u6f14**\u201c\u67e5\u8be2\u8005&#xff08;Query&#xff09;\u201d\u7684\u89d2\u8272&#xff0c;\u53bb\u5ba1\u89c6\u548c\u63a2\u5bfb\u5e8f\u5217\u4e2d\u5305\u62ec\u81ea\u8eab\u5728\u5185\u7684\u6240\u6709\u5176\u4ed6\u5143\u7d20&#xff08;\u5b83\u4eec\u540c\u65f6\u626e\u6f14\u7740Key\u548cValue&#xff09;**\u3002<\/p>\n<p>\u8ba9\u6211\u4eec\u7528\u4e00\u4e2a\u53e5\u5b50\u6765\u5177\u4f53\u5316\u8fd9\u4e2a\u8fc7\u7a0b&#xff1a;The animal didn&#039;t cross the street because it was too tired.<\/p>\n<p>\u5f53\u6a21\u578b\u5904\u7406\u5230 it \u8fd9\u4e2a\u8bcd\u65f6&#xff0c;\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4f1a\u8ba9 it \u751f\u6210\u4e00\u4e2aQuery&#xff0c;\u7136\u540e\u8fd9\u4e2aQuery\u4f1a\u53bb\u548c\u53e5\u5b50\u4e2d\u6240\u6709\u8bcd&#xff08;The, animal, didn&#039;t, &#8230;, tired, .)\u7684Key\u8fdb\u884c\u5339\u914d\u3002\u901a\u8fc7\u8ba1\u7b97&#xff0c;it \u7684Query\u53ef\u80fd\u4f1a\u53d1\u73b0&#xff0c;\u4e0e animal \u7684Key\u76f8\u4f3c\u5ea6\u6700\u9ad8&#xff0c;\u4e0e street \u7684Key\u76f8\u4f3c\u5ea6\u8f83\u4f4e\u3002\u4e8e\u662f&#xff0c;\u5728\u6700\u7ec8\u751f\u6210 it \u7684\u65b0\u8868\u793a\u65f6&#xff0c;\u6a21\u578b\u4f1a\u5206\u914d\u7ed9 animal \u7684Value\u4e00\u4e2a\u6781\u9ad8\u7684\u6ce8\u610f\u529b\u6743\u91cd&#xff0c;\u800c\u7ed9 street \u7684Value\u4e00\u4e2a\u5f88\u4f4e\u7684\u6743\u91cd\u3002<\/p>\n<p>\u4f5c\u7528&#xff1a;\u6784\u5efa\u52a8\u6001\u7684\u3001\u4e0a\u4e0b\u6587\u611f\u77e5\u7684\u8bcd\u8868\u793a \u901a\u8fc7\u8fd9\u79cd\u201c\u5185\u90e8\u51dd\u89c6\u201d&#xff0c;\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4e3a\u5e8f\u5217\u4e2d\u7684\u6bcf\u4e00\u4e2a\u8bcd&#xff0c;\u90fd\u6784\u5efa\u4e86\u4e00\u4e2a\u9ad8\u5ea6\u52a8\u6001\u3001\u6df1\u5ea6\u4e0a\u4e0b\u6587\u611f\u77e5\u7684\u8868\u793a\u3002\u4e00\u4e2a\u8bcd\u7684\u5411\u91cf\u8868\u793a&#xff0c;\u4e0d\u518d\u662f\u4e00\u4e2a\u5728\u8bcd\u5178\u4e2d\u56fa\u5b9a\u7684\u3001\u9759\u6001\u7684\u503c&#xff08;\u5982Word2Vec&#xff09;&#xff0c;\u800c\u662f\u6839\u636e\u5b83\u5728\u5f53\u524d\u53e5\u5b50\u4e2d\u4e0e\u5176\u4ed6\u6240\u6709\u8bcd\u7684\u5177\u4f53\u5173\u7cfb&#xff0c;\u88ab\u52a8\u6001\u5730\u201c\u91cd\u65b0\u5851\u9020\u201d\u4e86\u3002\u5b83\u5b8c\u7f8e\u5730\u89e3\u51b3\u4e86\u8bed\u8a00\u4e2d\u7684**\u4e00\u8bcd\u591a\u4e49&#xff08;Polysemy&#xff09;\u548c\u6307\u4ee3\u6d88\u89e3&#xff08;Coreference Resolution&#xff09;**\u7b49\u6838\u5fc3\u96be\u9898\u3002<\/p>\n<p>\u7f29\u653e\u70b9\u79ef\u6ce8\u610f\u529b&#xff08;Scaled Dot-Product Attention&#xff09;&#xff1a;\u4f18\u96c5\u7684\u5b9e\u73b0 Transformer\u4e2d\u4f7f\u7528\u7684&#xff0c;\u662f\u4e00\u79cd\u88ab\u79f0\u4e3a\u7f29\u653e\u70b9\u79ef\u6ce8\u610f\u529b\u7684\u5177\u4f53\u5b9e\u73b0\u3002\u5176\u8ba1\u7b97\u8fc7\u7a0b\u4e0e\u6211\u4eec9.1\u8282\u63cf\u8ff0\u7684\u4e09\u90e8\u66f2\u5b8c\u5168\u4e00\u81f4&#xff0c;\u4f46\u589e\u52a0\u4e86\u4e00\u4e2a\u770b\u4f3c\u5fae\u5c0f\u5374\u81f3\u5173\u91cd\u8981\u7684\u7ec6\u8282\u2014\u2014\u7f29\u653e&#xff08;Scaling&#xff09;\u3002<\/p>\n<p>\u8ba1\u7b97\u516c\u5f0f&#xff1a;Attention(Q, K, V) &#061; softmax( (Q * K^T) \/ \u221ad_k ) * V<\/p>\n<p>\u8fd9\u91cc\u7684 d_k \u662fKey\u5411\u91cf&#xff08;\u4e5f\u662fQuery\u5411\u91cf&#xff09;\u7684\u7ef4\u5ea6\u3002\u4e3a\u4ec0\u4e48\u8981\u8fdb\u884c\u8fd9\u4e2a\u7f29\u653e\u5462&#xff1f; \u7814\u7a76\u8005\u53d1\u73b0&#xff0c;\u5f53\u5411\u91cf\u7ef4\u5ea6 d_k \u8f83\u5927\u65f6&#xff0c;Q * K^T \u7684\u70b9\u79ef\u7ed3\u679c\u7684\u65b9\u5dee\u4e5f\u4f1a\u968f\u4e4b\u589e\u5927&#xff0c;\u8fd9\u610f\u5473\u7740\u70b9\u79ef\u7684\u7ed3\u679c\u53ef\u80fd\u4f1a\u53d8\u5f97\u975e\u5e38\u5927\u6216\u975e\u5e38\u5c0f\u3002\u5982\u679c\u5c06\u8fd9\u4e9b\u60ac\u6b8a\u7684\u6570\u503c\u76f4\u63a5\u8f93\u5165Softmax\u51fd\u6570&#xff0c;\u4f1a\u5bfc\u81f4Softmax\u7684\u68af\u5ea6\u53d8\u5f97\u6781\u5176\u5fae\u5c0f&#xff08;\u5373\u8fdb\u5165\u4e86\u68af\u5ea6\u9971\u548c\u533a&#xff09;&#xff0c;\u8fd9\u4f1a\u4e25\u91cd\u963b\u788d\u6a21\u578b\u7684\u5b66\u4e60\u3002 \u901a\u8fc7\u9664\u4ee5 \u221ad_k \u8fd9\u4e2a\u7f29\u653e\u56e0\u5b50&#xff0c;\u53ef\u4ee5\u5c06\u70b9\u79ef\u7ed3\u679c\u7684\u65b9\u5dee\u7a33\u5b9a\u57281\u5de6\u53f3&#xff0c;\u65e0\u8bba\u7ef4\u5ea6 d_k \u5982\u4f55\u53d8\u5316&#xff0c;\u90fd\u80fd\u4fdd\u8bc1Softmax\u51fd\u6570\u5de5\u4f5c\u5728\u4e00\u4e2a\u66f4\u5065\u5eb7\u3001\u68af\u5ea6\u66f4\u7a33\u5b9a\u7684\u533a\u57df\u3002\u8fd9\u662f\u4e00\u4e2a\u5178\u578b\u7684\u3001\u5c55\u73b0\u4e86\u6df1\u5ea6\u5b66\u4e60\u7814\u7a76\u4e2d\u7406\u8bba\u6d1e\u5bdf\u4e0e\u5de5\u7a0b\u5b9e\u8df5\u76f8\u7ed3\u5408\u4e4b\u7f8e\u7684\u4f8b\u5b50\u3002<\/p>\n<h5>9.2.3 \u591a\u5934\u6ce8\u610f\u529b&#xff08;Multi-Head Attention&#xff09;&#xff1a;\u4ece\u4e0d\u540c\u201c\u8ba4\u77e5\u901a\u9053\u201d\u5ba1\u89c6\u5173\u7cfb<\/h5>\n<p>\u5982\u679c\u8bf4\u81ea\u6ce8\u610f\u529b\u673a\u5236\u8ba9\u6a21\u578b\u5b66\u4f1a\u4e86\u201c\u51dd\u89c6\u201d&#xff0c;\u90a3\u4e48\u591a\u5934\u6ce8\u610f\u529b\u5219\u8d4b\u4e88\u4e86\u6a21\u578b\u4e00\u53cc\u201c\u590d\u773c\u201d&#xff0c;\u4f7f\u5176\u80fd\u591f\u4ece\u591a\u4e2a\u4e0d\u540c\u7684\u89d2\u5ea6&#xff0c;\u540c\u65f6\u8fdb\u884c\u51dd\u89c6\u3002<\/p>\n<p>\u5355\u4e00\u6ce8\u610f\u529b\u7684\u5c40\u9650&#xff1a;\u8ba4\u77e5\u7684\u201c\u504f\u89c1\u201d \u53ea\u7528\u4e00\u5957Q, K, V\u77e9\u9635\u8fdb\u884c\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97&#xff0c;\u5c31\u597d\u6bd4\u6211\u4eec\u53ea\u7528\u4e00\u79cd\u6807\u51c6\u6216\u4e00\u4e2a\u201c\u8ba4\u77e5\u901a\u9053\u201d\u53bb\u7406\u89e3\u8bcd\u4e0e\u8bcd\u4e4b\u95f4\u7684\u5173\u7cfb\u3002\u4f8b\u5982&#xff0c;\u5b83\u53ef\u80fd\u5b66\u4f1a\u4e86\u5173\u6ce8\u53e5\u6cd5\u4e0a\u7684\u4e3b\u8c13\u5173\u7cfb&#xff0c;\u4f46\u5374\u5ffd\u7565\u4e86\u8bed\u4e49\u4e0a\u7684\u540c\u4e49\u5173\u7cfb\u3002\u8fd9\u79cd\u5355\u4e00\u7684\u89c6\u89d2&#xff0c;\u9650\u5236\u4e86\u6a21\u578b\u6355\u6349\u4fe1\u606f\u4e30\u5bcc\u6027\u7684\u80fd\u529b\u3002<\/p>\n<p>\u5de5\u4f5c\u539f\u7406&#xff1a;\u5206\u800c\u6cbb\u4e4b&#xff0c;\u878d\u4f1a\u8d2f\u901a **\u591a\u5934\u6ce8\u610f\u529b&#xff08;Multi-Head Attention&#xff09;**\u7684\u89e3\u51b3\u65b9\u6848&#xff0c;\u662f\u4e00\u79cd\u4f18\u96c5\u7684\u201c\u5206\u800c\u6cbb\u4e4b\u201d\u7b56\u7565\u3002\u5047\u8bbe\u6211\u4eec\u8bbe\u5b9a\u4e86 h \u4e2a\u201c\u5934\u201d&#xff08;\u4f8b\u5982&#xff0c;\u5728Transformer Base\u6a21\u578b\u4e2d&#xff0c;h&#061;8&#xff09;\u3002<\/p>\n<li>\u6295\u5f71&#xff08;Projection&#xff09;&#xff1a;\u9996\u5148&#xff0c;\u5b83\u4e3a\u6bcf\u4e00\u4e2a\u5934&#xff0c;\u90fd\u51c6\u5907\u4e86\u4e00\u7ec4\u72ec\u7acb\u7684\u3001\u53ef\u5b66\u4e60\u7684\u7ebf\u6027\u53d8\u6362\u77e9\u9635&#xff08;\u6743\u91cd\u77e9\u9635&#xff09;\u3002\u5b83\u5c06\u539f\u59cb\u7684\u3001\u9ad8\u7ef4\u7684Q, K, V\u8f93\u5165&#xff0c;\u5206\u522b\u901a\u8fc7\u8fd9\u4e9b\u77e9\u9635&#xff0c;\u6295\u5f71\u5230\u00a0h\u00a0\u4e2a\u4e0d\u540c\u7684\u3001\u4f4e\u7ef4\u7684\u8868\u793a\u5b50\u7a7a\u95f4\u4e2d\u3002\u4e5f\u5c31\u662f\u8bf4&#xff0c;\u6211\u4eec\u5f97\u5230\u4e86\u00a0h\u00a0\u7ec4\u4f4e\u7ef4\u7684\u00a0(Q_i, K_i, V_i)&#xff0c;\u5176\u4e2d\u00a0i\u00a0\u4ece1\u5230\u00a0h\u3002<\/li>\n<li>\u5e76\u884c\u6ce8\u610f\u529b\u8ba1\u7b97&#xff08;Parallel Attention&#xff09;&#xff1a;\u7136\u540e&#xff0c;\u5728\u8fd9\u00a0h\u00a0\u4e2a\u5b50\u7a7a\u95f4\u4e2d&#xff0c;\u5e76\u884c\u5730\u3001\u72ec\u7acb\u5730\u8fdb\u884c\u00a0h\u00a0\u6b21\u7f29\u653e\u70b9\u79ef\u6ce8\u610f\u529b\u8ba1\u7b97\u3002\u6bcf\u4e00\u4e2a\u5934&#xff0c;\u90fd\u5c06\u8f93\u51fa\u4e00\u4e2a\u5b83\u81ea\u5df1\u201c\u89c6\u89d2\u201d\u4e0b\u7684\u6ce8\u610f\u529b\u7ed3\u679c\u3002<\/li>\n<li>\u62fc\u63a5\u4e0e\u878d\u5408&#xff08;Concatenation &amp; Final Projection&#xff09;&#xff1a;\u6700\u540e&#xff0c;\u5c06\u8fd9\u00a0h\u00a0\u4e2a\u5934\u8f93\u51fa\u7684\u7ed3\u679c\u5411\u91cf\u62fc\u63a5&#xff08;Concatenate&#xff09;\u8d77\u6765&#xff0c;\u5f62\u6210\u4e00\u4e2a\u5927\u7684\u5411\u91cf\u3002\u518d\u5c06\u8fd9\u4e2a\u5927\u5411\u91cf\u901a\u8fc7\u6700\u540e\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684\u7ebf\u6027\u53d8\u6362\u77e9\u9635&#xff0c;\u5c06\u5176\u878d\u5408\u5e76\u6295\u5f71\u56de\u6a21\u578b\u6240\u671f\u671b\u7684\u539f\u59cb\u8f93\u51fa\u7ef4\u5ea6\u3002<\/li>\n<p>\u4f18\u52bf&#xff1a;\u4e30\u5bcc\u7684\u7279\u5f81\u8868\u793a \u591a\u5934\u673a\u5236\u7684\u5a01\u529b\u5728\u4e8e&#xff0c;\u5b83\u5141\u8bb8\u6a21\u578b\u5728\u4e0d\u540c\u7684\u8868\u793a\u5b50\u7a7a\u95f4\u4e2d&#xff0c;\u540c\u65f6\u5b66\u4e60\u548c\u5173\u6ce8\u4e0d\u540c\u7c7b\u578b\u7684\u5173\u7cfb\u3002\u4f8b\u5982&#xff0c;\u5728\u5904\u7406\u53e5\u5b50 The animal didn&#039;t cross the street because it was too tired. \u65f6&#xff1a;<\/p>\n<ul>\n<li>\u4e00\u4e2a\u5934\u53ef\u80fd\u5b66\u4f1a\u4e86\u5173\u6ce8\u6307\u4ee3\u5173\u7cfb&#xff0c;\u5c06\u00a0it\u00a0\u548c\u00a0animal\u00a0\u5f3a\u70c8\u5730\u5173\u8054\u8d77\u6765\u3002<\/li>\n<li>\u53e6\u4e00\u4e2a\u5934\u53ef\u80fd\u5b66\u4f1a\u4e86\u5173\u6ce8\u56e0\u679c\u5173\u7cfb&#xff0c;\u5c06\u00a0tired\u00a0\u548c\u00a0didn&#039;t cross\u00a0\u5173\u8054\u8d77\u6765\u3002<\/li>\n<li>\u8fd8\u6709\u4e00\u4e2a\u5934\u53ef\u80fd\u53ea\u662f\u5173\u6ce8\u4e00\u4e9b\u4f4d\u7f6e\u4e0a\u7684\u8fd1\u90bb\u5173\u7cfb\u3002 \u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u591a\u5934\u6ce8\u610f\u529b\u6781\u5927\u5730\u4e30\u5bcc\u4e86\u6a21\u578b\u6355\u6349\u548c\u6574\u5408\u4fe1\u606f\u7684\u80fd\u529b&#xff0c;\u4f7f\u5f97\u6700\u7ec8\u7684\u8f93\u51fa\u8868\u793a&#xff0c;\u662f\u591a\u4e2a\u4e0d\u540c\u201c\u4e13\u5bb6\u201d\u89c6\u89d2\u4e0b\u7684\u4fe1\u606f\u878d\u5408&#xff0c;\u4ece\u800c\u66f4\u52a0\u5168\u9762\u548c\u5f3a\u5927\u3002<\/li>\n<\/ul>\n<h5>9.2.4 \u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;&#xff1a;\u5728\u65e0\u5e8f\u7684\u4e16\u754c\u4e2d\u627e\u56de\u201c\u987a\u5e8f\u611f\u201d<\/h5>\n<p>\u8fd9\u662f\u7406\u89e3Transformer\u7684\u6700\u540e\u4e00\u4e2a\u5173\u952e&#xff0c;\u4e5f\u662f\u4e00\u4e2a\u53cd\u76f4\u89c9\u7684\u3001\u5929\u624d\u822c\u7684\u8bbe\u8ba1\u3002<\/p>\n<p>\u5e76\u884c\u5316\u5e26\u6765\u7684\u201c\u5931\u5fc6\u75c7\u201d \u6211\u4eec\u4e4b\u524d\u76db\u8d5eTransformer\u7684\u5e76\u884c\u8ba1\u7b97\u80fd\u529b&#xff0c;\u4f46\u8fd9\u79cd\u80fd\u529b\u4e5f\u5e26\u6765\u4e86\u4e00\u4e2a\u81f4\u547d\u7684\u526f\u4f5c\u7528&#xff1a;\u6a21\u578b\u672c\u8eab\u65e0\u6cd5\u611f\u77e5\u5230\u5e8f\u5217\u7684\u987a\u5e8f\u3002\u7531\u4e8e\u6240\u6709\u8bcd\u90fd\u662f\u540c\u65f6\u88ab\u5904\u7406\u7684&#xff0c;\u5bf9\u4e8e\u4e00\u4e2a\u7eaf\u7cb9\u7684\u81ea\u6ce8\u610f\u529b\u7f51\u7edc\u6765\u8bf4&#xff0c;\u201c\u732b \u8ffd \u8001\u9f20\u201d\u548c\u201c\u8001\u9f20 \u8ffd \u732b\u201d\u8fd9\u4e24\u4e2a\u8f93\u5165\u662f\u5b8c\u5168\u7b49\u4ef7\u7684&#xff0c;\u56e0\u4e3a\u5b83\u53ea\u770b\u5230\u4e86\u8bcd\u7684\u96c6\u5408&#xff0c;\u800c\u4e22\u5931\u4e86\u5b83\u4eec\u7684\u6392\u5217\u987a\u5e8f\u3002\u8fd9\u5bf9\u4e8e\u7406\u89e3\u8bed\u8a00\u6765\u8bf4\u662f\u4e0d\u53ef\u63a5\u53d7\u7684\u3002<\/p>\n<p>\u89e3\u51b3\u65b9\u6848&#xff1a;\u4e3a\u8bcd\u8bed\u6ce8\u5165\u201c\u4f4d\u7f6e\u4fe1\u53f7\u201d Transformer\u7684\u89e3\u51b3\u65b9\u6848&#xff0c;\u4e0d\u662f\u5728\u6a21\u578b\u7ed3\u6784\u4e2d\u53bb\u5904\u7406\u987a\u5e8f&#xff0c;\u800c\u662f\u5728\u8f93\u5165\u7aef&#xff0c;\u76f4\u63a5\u4e3a\u6bcf\u4e2a\u8bcd\u7684\u5d4c\u5165\u5411\u91cf&#xff0c;\u6ce8\u5165&#xff08;\u52a0\u4e0a&#xff09;\u4e00\u4e2a\u4ee3\u8868\u5176\u4f4d\u7f6e\u4fe1\u606f\u7684\u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;\u5411\u91cf\u3002 \u8fd9\u4e2a\u4f4d\u7f6e\u7f16\u7801\u5411\u91cf&#xff0c;\u5e76\u4e0d\u662f\u4e00\u4e2a\u9700\u8981\u5b66\u4e60\u7684\u53c2\u6570&#xff0c;\u800c\u662f\u6839\u636e\u8bcd\u5728\u5e8f\u5217\u4e2d\u7684\u7edd\u5bf9\u4f4d\u7f6e pos&#xff0c;\u901a\u8fc7\u4e00\u7ec4\u56fa\u5b9a\u7684\u3001\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u6b63\u5f26&#xff08;sin&#xff09;\u548c\u4f59\u5f26&#xff08;cos&#xff09;\u51fd\u6570\u6765\u751f\u6210\u7684\u3002 PE(pos, 2i) &#061; sin(pos \/ 10000^(2i\/d_model)) PE(pos, 2i&#043;1) &#061; cos(pos \/ 10000^(2i\/d_model)) \u5176\u4e2d&#xff0c;pos\u662f\u8bcd\u7684\u4f4d\u7f6e\u7d22\u5f15&#xff08;0, 1, 2, &#8230;&#xff09;&#xff0c;i\u662f\u7f16\u7801\u5411\u91cf\u7684\u7ef4\u5ea6\u7d22\u5f15&#xff0c;d_model\u662f\u6a21\u578b\u7684\u603b\u7ef4\u5ea6\u3002<\/p>\n<p>\u8bbe\u8ba1\u7684\u5de7\u5999\u4e4b\u5904&#xff1a;\u76f8\u5bf9\u4f4d\u7f6e\u7684\u7f16\u7801 \u4e3a\u4ec0\u4e48\u9009\u62e9\u5982\u6b64\u590d\u6742\u7684\u4e09\u89d2\u51fd\u6570&#xff0c;\u800c\u4e0d\u662f\u7b80\u5355\u5730\u7ed9\u6bcf\u4e2a\u4f4d\u7f6e\u5206\u914d\u4e00\u4e2a\u6570\u5b57&#xff08;\u59820.1, 0.2, &#8230;&#xff09;\u5462&#xff1f;\u8fd9\u6b63\u662f\u5176\u8bbe\u8ba1\u7684\u7cbe\u5999\u6240\u5728&#xff1a;<\/p>\n<li>\u552f\u4e00\u6027&#xff1a;\u5b83\u4e3a\u6bcf\u4e2a\u4f4d\u7f6e\u90fd\u751f\u6210\u4e86\u4e00\u4e2a\u72ec\u4e00\u65e0\u4e8c\u7684\u7f16\u7801\u3002<\/li>\n<li>\u76f8\u5bf9\u4f4d\u7f6e\u4fe1\u606f&#xff1a;\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u5bf9\u4e8e\u4efb\u610f\u56fa\u5b9a\u7684\u504f\u79fb\u91cf\u00a0k&#xff0c;PE(pos&#043;k)\u00a0\u90fd\u53ef\u4ee5\u8868\u793a\u4e3a\u00a0PE(pos)\u00a0\u7684\u4e00\u4e2a\u7ebf\u6027\u51fd\u6570\u3002\u8fd9\u610f\u5473\u7740&#xff0c;\u8bcd\u4e0e\u8bcd\u4e4b\u95f4\u7684\u76f8\u5bf9\u4f4d\u7f6e\u5173\u7cfb&#xff0c;\u88ab\u7f16\u7801\u5728\u4e86\u8fd9\u4e9b\u5411\u91cf\u7684\u7ebf\u6027\u5173\u7cfb\u4e4b\u4e2d\u3002\u6a21\u578b\u4e0d\u9700\u8981\u53bb\u8bb0\u5fc6\u6bcf\u4e2a\u8bcd\u7684\u7edd\u5bf9\u4f4d\u7f6e&#xff0c;\u800c\u662f\u53ef\u4ee5\u5f88\u5bb9\u6613\u5730\u901a\u8fc7\u6bd4\u8f83\u5b83\u4eec\u7684\u7f16\u7801\u5411\u91cf&#xff0c;\u6765\u5b66\u4e60\u5230\u5b83\u4eec\u4e4b\u95f4\u7684\u76f8\u5bf9\u8ddd\u79bb\u548c\u987a\u5e8f\u3002<\/li>\n<li>\u6cdb\u5316\u80fd\u529b&#xff1a;\u8fd9\u79cd\u57fa\u4e8e\u51fd\u6570\u751f\u6210\u7684\u65b9\u5f0f&#xff0c;\u4f7f\u5f97\u6a21\u578b\u7406\u8bba\u4e0a\u53ef\u4ee5\u5904\u7406\u6bd4\u8bad\u7ec3\u4e2d\u9047\u5230\u7684\u66f4\u957f\u7684\u5e8f\u5217\u3002<\/li>\n<p>\u901a\u8fc7\u5c06\u8bcd\u5d4c\u5165\u548c\u4f4d\u7f6e\u7f16\u7801\u76f8\u52a0&#xff0c;Transformer\u6210\u529f\u5730\u5728\u4e0d\u7834\u574f\u5e76\u884c\u8ba1\u7b97\u7684\u524d\u63d0\u4e0b&#xff0c;\u5c06\u5e8f\u5217\u7684\u987a\u5e8f\u4fe1\u606f\u65e0\u7f1d\u5730\u878d\u5165\u5230\u4e86\u6a21\u578b\u7684\u8f93\u5165\u4e4b\u4e2d\u3002<\/p>\n<h5>9.2.5 \u6574\u4f53\u67b6\u6784&#xff1a;\u7f16\u7801\u5668-\u89e3\u7801\u5668\u7684\u7cbe\u5bc6\u534f\u4f5c<\/h5>\n<p>\u73b0\u5728&#xff0c;\u6211\u4eec\u53ef\u4ee5\u5c06\u6240\u6709\u8fd9\u4e9b\u90e8\u4ef6\u7ec4\u88c5\u8d77\u6765&#xff0c;\u4e00\u7aa5Transformer\u5b8c\u6574\u7684\u7f16\u7801\u5668-\u89e3\u7801\u5668&#xff08;Encoder-Decoder&#xff09;\u67b6\u6784\u3002<\/p>\n<p>\u7f16\u7801\u5668&#xff08;The Encoder&#xff09; \u7f16\u7801\u5668\u7684\u4efb\u52a1\u662f\u8bfb\u53d6\u5e76\u7406\u89e3\u6574\u4e2a\u8f93\u5165\u5e8f\u5217\u3002\u5b83\u7531N\u4e2a&#xff08;\u539f\u8bba\u6587\u4e2dN&#061;6&#xff09;\u5b8c\u5168\u76f8\u540c\u7684\u7f16\u7801\u5668\u5c42&#xff08;Encoder Layer&#xff09;\u5806\u53e0\u800c\u6210\u3002 \u6bcf\u4e00\u4e2a\u7f16\u7801\u5668\u5c42&#xff0c;\u90fd\u7531\u4e24\u4e2a\u6838\u5fc3\u7684\u5b50\u6a21\u5757\u7ec4\u6210&#xff1a;<\/p>\n<li>\u4e00\u4e2a\u591a\u5934\u81ea\u6ce8\u610f\u529b\u6a21\u5757&#xff08;Multi-Head Self-Attention&#xff09;\u3002<\/li>\n<li>\u4e00\u4e2a\u7b80\u5355\u7684\u3001\u4f4d\u7f6e\u5168\u8fde\u63a5\u7684\u524d\u9988\u795e\u7ecf\u7f51\u7edc&#xff08;Position-wise Feed-Forward Network&#xff09;\u3002\u8fd9\u4e2a\u524d\u9988\u7f51\u7edc\u7531\u4e24\u4e2a\u7ebf\u6027\u5c42\u548c\u4e00\u4e2aReLU\u6fc0\u6d3b\u51fd\u6570\u7ec4\u6210&#xff0c;\u5b83\u88ab\u72ec\u7acb\u5730\u5e94\u7528\u4e8e\u6bcf\u4e00\u4e2a\u4f4d\u7f6e\u7684\u8f93\u51fa\u4e0a\u3002 \u5728\u6bcf\u4e2a\u5b50\u6a21\u5757\u7684\u540e\u9762&#xff0c;\u90fd\u8ddf\u968f\u7740\u4e00\u4e2a**\u6b8b\u5dee\u8fde\u63a5&#xff08;Residual Connection&#xff09;\u548c\u5c42\u5f52\u4e00\u5316&#xff08;Layer Normalization&#xff09;**\u64cd\u4f5c\u3002\u8fd9\u4e0eResNet\u4e2d\u7684\u601d\u60f3\u4e00\u81f4&#xff0c;\u6781\u5927\u5730\u5e2e\u52a9\u4e86\u6df1\u5c42\u7f51\u7edc\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u548c\u4fe1\u606f\u6d41\u52a8\u3002<\/li>\n<p>\u89e3\u7801\u5668&#xff08;The Decoder&#xff09; \u89e3\u7801\u5668\u7684\u4efb\u52a1\u662f\u6839\u636e\u7f16\u7801\u5668\u5bf9\u6e90\u5e8f\u5217\u7684\u7406\u89e3&#xff0c;\u6765\u751f\u6210\u76ee\u6807\u5e8f\u5217\u3002\u5b83\u540c\u6837\u7531N\u4e2a&#xff08;N&#061;6&#xff09;\u5b8c\u5168\u76f8\u540c\u7684\u89e3\u7801\u5668\u5c42&#xff08;Decoder Layer&#xff09;\u5806\u53e0\u800c\u6210\u3002 \u6bcf\u4e00\u4e2a\u89e3\u7801\u5668\u5c42&#xff0c;\u6bd4\u7f16\u7801\u5668\u5c42\u8981\u590d\u6742\u4e00\u4e9b&#xff0c;\u5b83\u7531\u4e09\u4e2a\u6838\u5fc3\u7684\u5b50\u6a21\u5757\u7ec4\u6210&#xff1a;<\/p>\n<li>\u4e00\u4e2a\u5e26\u63a9\u7801\u7684\u591a\u5934\u81ea\u6ce8\u610f\u529b\u6a21\u5757&#xff08;Masked Multi-Head Self-Attention&#xff09;\u3002\u8fd9\u91cc\u7684\u201c\u63a9\u7801\u201d\u81f3\u5173\u91cd\u8981\u3002\u5728\u8bad\u7ec3\u89e3\u7801\u5668\u65f6&#xff0c;\u4e3a\u4e86\u6a21\u62df\u771f\u5b9e\u7684\u751f\u6210\u8fc7\u7a0b&#xff0c;\u6211\u4eec\u5fc5\u987b\u9632\u6b62\u6a21\u578b\u5728\u9884\u6d4b\u7b2c\u00a0i\u00a0\u4e2a\u8bcd\u65f6&#xff0c;\u201c\u5077\u770b\u201d\u5230\u7b2c\u00a0i\u00a0\u4e2a\u8bcd\u4e4b\u540e\u7684\u7b54\u6848\u3002\u63a9\u7801\u7684\u4f5c\u7528&#xff0c;\u5c31\u662f\u5728\u81ea\u6ce8\u610f\u529b\u8ba1\u7b97\u4e2d&#xff0c;\u5c06\u6240\u6709\u672a\u6765\u4f4d\u7f6e\u7684\u6ce8\u610f\u529b\u5206\u6570\u8bbe\u7f6e\u4e3a\u8d1f\u65e0\u7a77&#xff0c;\u8fd9\u6837\u5728Softmax\u4e4b\u540e&#xff0c;\u5b83\u4eec\u7684\u6743\u91cd\u5c31\u53d8\u6210\u4e860\u3002<\/li>\n<li>\u4e00\u4e2a\u7f16\u7801\u5668-\u89e3\u7801\u5668\u6ce8\u610f\u529b\u6a21\u5757&#xff08;Encoder-Decoder Attention&#xff09;\u3002\u8fd9\u662f\u8fde\u63a5\u7f16\u7801\u5668\u548c\u89e3\u7801\u5668\u7684\u6865\u6881\u3002\u5728\u8fd9\u4e2a\u6a21\u5757\u4e2d&#xff0c;Query\u6765\u81ea\u4e8e\u89e3\u7801\u5668\u81ea\u8eab&#xff08;\u524d\u4e00\u4e2a\u81ea\u6ce8\u610f\u529b\u6a21\u5757\u7684\u8f93\u51fa&#xff09;&#xff0c;\u800cKey\u548cValue\u5219\u6765\u81ea\u4e8e\u7f16\u7801\u5668\u6700\u7ec8\u7684\u8f93\u51fa\u3002\u8fd9\u5b8c\u7f8e\u5730\u5b9e\u73b0\u4e86\u6211\u4eec\u6700\u521d\u63cf\u8ff0\u7684\u6ce8\u610f\u529b\u673a\u5236&#xff1a;\u89e3\u7801\u5668\u6839\u636e\u81ea\u5df1\u5f53\u524d\u7684\u72b6\u6001&#xff0c;\u53bb\u201c\u67e5\u8be2\u201d\u6e90\u5e8f\u5217\u7684\u8868\u793a&#xff0c;\u5e76\u4ece\u4e2d\u63d0\u53d6\u6240\u9700\u7684\u4fe1\u606f\u3002<\/li>\n<li>\u4e00\u4e2a\u4f4d\u7f6e\u5168\u8fde\u63a5\u7684\u524d\u9988\u795e\u7ecf\u7f51\u7edc&#xff0c;\u4e0e\u7f16\u7801\u5668\u4e2d\u7684\u7ed3\u6784\u76f8\u540c\u3002 \u540c\u6837&#xff0c;\u89e3\u7801\u5668\u7684\u6bcf\u4e2a\u5b50\u6a21\u5757\u540e\u9762&#xff0c;\u4e5f\u90fd\u8ddf\u968f\u7740\u6b8b\u5dee\u8fde\u63a5\u548c\u5c42\u5f52\u4e00\u5316\u3002<\/li>\n<p>\u901a\u8fc7\u8fd9\u79cd\u7cbe\u5bc6\u7684\u3001\u7531\u591a\u5c42\u81ea\u6ce8\u610f\u529b\u548c\u524d\u9988\u7f51\u7edc\u6784\u6210\u7684\u7ed3\u6784&#xff0c;Transformer\u4e0d\u4ec5\u5f7b\u5e95\u6446\u8131\u4e86\u5faa\u73af\u7684\u675f\u7f1a&#xff0c;\u66f4\u662f\u5728\u5e8f\u5217\u5efa\u6a21\u7684\u80fd\u529b\u548c\u6548\u7387\u4e0a&#xff0c;\u8fbe\u5230\u4e86\u524d\u6240\u672a\u6709\u7684\u65b0\u9ad8\u5ea6&#xff0c;\u4e3a\u540e\u7eed\u7684AI\u9769\u547d\u5960\u5b9a\u4e86\u575a\u4e0d\u53ef\u6467\u7684\u57fa\u77f3\u3002<\/p>\n<h4>9.3 BERT\u3001GPT\u53ca\u5176\u4ed6&#xff1a;\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u7684\u65b0\u8303\u5f0f<\/h4>\n<p>Transformer\u7684\u8bde\u751f&#xff0c;\u4e3a\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;\u9886\u57df\u63d0\u4f9b\u4e86\u4e00\u4ef6\u524d\u6240\u672a\u6709\u7684\u201c\u795e\u5668\u201d\u3002\u7136\u800c&#xff0c;\u5982\u4f55\u5c06\u8fd9\u4ef6\u795e\u5668\u7684\u5a01\u529b\u53d1\u6325\u5230\u6781\u81f4&#xff0c;\u8fd8\u9700\u8981\u4e00\u573a\u65b9\u6cd5\u8bba\u4e0a\u7684\u9769\u547d\u3002\u8fd9\u573a\u9769\u547d&#xff0c;\u5c31\u662f**\u201c\u9884\u8bad\u7ec3-\u5fae\u8c03\u201d&#xff08;Pre-training and Fine-tuning&#xff09;**\u65b0\u8303\u5f0f\u7684\u5d1b\u8d77&#xff0c;\u800cBERT\u548cGPT&#xff0c;\u6b63\u662f\u8fd9\u573a\u9769\u547d\u4e2d\u6700\u8000\u773c\u7684\u4e24\u9762\u65d7\u5e1c\u3002<\/p>\n<h5>9.3.1 \u8303\u5f0f\u7684\u8f6c\u53d8&#xff1a;\u4ece\u201c\u4e3a\u4efb\u52a1\u4ece\u96f6\u5f00\u59cb\u201d\u5230\u201c\u7ad9\u5728\u5de8\u4eba\u7684\u77e5\u8bc6\u4e0a\u201d<\/h5>\n<p>\u4f20\u7edf\u8303\u5f0f\u7684\u56f0\u5883&#xff1a;\u6570\u636e\u9965\u6e34\u4e0e\u77e5\u8bc6\u5272\u88c2 \u5728\u9884\u8bad\u7ec3\u6a21\u578b\u51fa\u73b0\u4e4b\u524d&#xff0c;NLP\u4efb\u52a1\u7684\u4e3b\u6d41\u505a\u6cd5\u662f\u201c\u4e3a\u4efb\u52a1\u4ece\u96f6\u5f00\u59cb&#xff08;Training from Scratch&#xff09;\u201d\u3002\u5373\u4fbf\u662f\u4f7f\u7528\u4e86\u5f3a\u5927\u7684Transformer\u67b6\u6784&#xff0c;\u9488\u5bf9\u4e00\u4e2a\u5177\u4f53\u7684\u4efb\u52a1&#xff08;\u5982\u60c5\u611f\u5206\u6790&#xff09;&#xff0c;\u6211\u4eec\u4f9d\u7136\u9700\u8981&#xff1a;<\/p>\n<li>\u6536\u96c6\u5927\u91cf\u4e0e\u8be5\u4efb\u52a1\u76f8\u5173\u7684\u3001\u6709\u6807\u7b7e\u7684\u8bad\u7ec3\u6570\u636e\u3002<\/li>\n<li>\u968f\u673a\u521d\u59cb\u5316\u4e00\u4e2a\u6a21\u578b\u3002<\/li>\n<li>\u7528\u8fd9\u4e9b\u6807\u6ce8\u6570\u636e&#xff0c;\u4ece\u5934\u5f00\u59cb\u8bad\u7ec3\u6a21\u578b\u7684\u5168\u90e8\u53c2\u6570\u3002<\/li>\n<p>\u8fd9\u79cd\u8303\u5f0f\u5b58\u5728\u4e24\u4e2a\u81f4\u547d\u7684\u7f3a\u9677&#xff1a;<\/p>\n<ul>\n<li>\u6570\u636e\u9965\u6e34&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u901a\u5e38\u9700\u8981\u6d77\u91cf\u7684\u6807\u6ce8\u6570\u636e\u624d\u80fd\u8868\u73b0\u826f\u597d\u3002\u4f46\u5bf9\u4e8e\u8bb8\u591aNLP\u4efb\u52a1\u800c\u8a00&#xff0c;\u83b7\u53d6\u9ad8\u8d28\u91cf\u7684\u6807\u6ce8\u6570\u636e\u6210\u672c\u9ad8\u6602\u4e14\u8017\u65f6\u3002<\/li>\n<li>\u77e5\u8bc6\u5272\u88c2&#xff1a;\u6bcf\u4e2a\u4efb\u52a1\u90fd\u4ece\u96f6\u5f00\u59cb\u8bad\u7ec3&#xff0c;\u610f\u5473\u7740\u6a21\u578b\u5b66\u5230\u7684\u77e5\u8bc6\u662f\u5b64\u7acb\u7684\u3001\u65e0\u6cd5\u8fc1\u79fb\u7684\u3002\u4e3a\u60c5\u611f\u5206\u6790\u8bad\u7ec3\u7684\u6a21\u578b&#xff0c;\u5bf9\u8bed\u6cd5\u3001\u5e38\u8bc6\u3001\u4e16\u754c\u77e5\u8bc6\u4e00\u65e0\u6240\u77e5&#xff1b;\u4e3a\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u8bad\u7ec3\u7684\u6a21\u578b&#xff0c;\u4e5f\u65e0\u6cd5\u7406\u89e3\u53e5\u5b50\u7684\u60c5\u611f\u3002\u6211\u4eec\u6bcf\u6b21\u90fd\u5728\u91cd\u590d\u5730\u6559\u6a21\u578b\u4e00\u4e9b\u6700\u57fa\u7840\u7684\u8bed\u8a00\u77e5\u8bc6&#xff0c;\u6548\u7387\u6781\u5176\u4f4e\u4e0b\u3002<\/li>\n<\/ul>\n<p>\u65b0\u8303\u5f0f\u7684\u66d9\u5149&#xff1a;\u501f\u9274\u8ba1\u7b97\u673a\u89c6\u89c9\u7684\u6210\u529f\u7ecf\u9a8c \u8fd9\u573a\u8303\u5f0f\u9769\u547d\u7684\u7075\u611f&#xff0c;\u5f88\u5927\u7a0b\u5ea6\u4e0a\u501f\u9274\u4e86\u8ba1\u7b97\u673a\u89c6\u89c9&#xff08;CV&#xff09;\u9886\u57df\u7684\u6210\u529f\u7ecf\u9a8c\u3002\u5728CV\u9886\u57df&#xff0c;\u7814\u7a76\u8005\u4eec\u65e9\u5df2\u53d1\u73b0\u5728ImageNet\u8fd9\u6837\u7684\u5927\u89c4\u6a21\u6570\u636e\u96c6\u4e0a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff08;\u5982ResNet&#xff09;&#xff0c;\u5176\u5b66\u5230\u7684\u5377\u79ef\u5c42\u80fd\u591f\u6355\u6349\u5230\u901a\u7528\u7684\u89c6\u89c9\u7279\u5f81&#xff08;\u5982\u8fb9\u7f18\u3001\u7eb9\u7406\u3001\u5f62\u72b6&#xff09;\u3002\u5f53\u9762\u5bf9\u4e00\u4e2a\u65b0\u7684\u3001\u6570\u636e\u91cf\u8f83\u5c0f\u7684\u4e0b\u6e38\u4efb\u52a1&#xff08;\u5982\u732b\u72d7\u5206\u7c7b&#xff09;\u65f6&#xff0c;\u7814\u7a76\u8005\u65e0\u9700\u4ece\u96f6\u8bad\u7ec3&#xff0c;\u53ea\u9700\u5728\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u57fa\u7840\u4e0a\u8fdb\u884c\u5fae\u8c03&#xff08;Fine-tuning&#xff09;&#xff0c;\u7528\u5c11\u91cf\u4efb\u52a1\u76f8\u5173\u6570\u636e\u8c03\u6574\u4e00\u4e0b\u9ad8\u5c42\u53c2\u6570&#xff0c;\u5c31\u80fd\u53d6\u5f97\u975e\u5e38\u597d\u7684\u6548\u679c\u3002<\/p>\n<p>NLP\u7684\u65b0\u8303\u5f0f&#xff1a;\u4e24\u9636\u6bb5\u5b66\u4e60 NLP\u7684\u7814\u7a76\u8005\u4eec\u5c06\u8fd9\u4e00\u601d\u60f3\u5f15\u5165\u4e86\u8bed\u8a00\u9886\u57df&#xff0c;\u5f62\u6210\u4e86\u201c\u9884\u8bad\u7ec3-\u5fae\u8c03\u201d\u7684\u4e24\u9636\u6bb5\u65b0\u8303\u5f0f&#xff1a;<\/p>\n<li>\u9884\u8bad\u7ec3&#xff08;Pre-training&#xff09;\u9636\u6bb5&#xff1a;\u8fd9\u662f\u65b0\u8303\u5f0f\u7684\u6838\u5fc3\u3002\u6211\u4eec\u5229\u7528\u4e92\u8054\u7f51\u4e0a\u553e\u624b\u53ef\u5f97\u7684\u3001\u6d77\u91cf\u7684\u3001\u65e0\u6807\u7b7e\u7684\u6587\u672c\u6570\u636e&#xff08;\u5982\u7ef4\u57fa\u767e\u79d1\u3001\u65b0\u95fb\u3001\u4e66\u7c4d\u7b49&#xff09;&#xff0c;\u6765\u8bad\u7ec3\u4e00\u4e2a\u5de8\u5927\u7684Transformer\u6a21\u578b\u3002\u5173\u952e\u5728\u4e8e&#xff0c;\u6211\u4eec\u4e3a\u8fd9\u4e2a\u9636\u6bb5\u8bbe\u8ba1\u4e86\u5de7\u5999\u7684\u81ea\u76d1\u7763\u5b66\u4e60&#xff08;Self-supervised Learning&#xff09;\u4efb\u52a1\u3002\u6a21\u578b\u4e0d\u9700\u8981\u4eba\u5de5\u6807\u7b7e&#xff0c;\u800c\u662f\u4ece\u6587\u672c\u81ea\u8eab\u4e2d\u521b\u9020\u5b66\u4e60\u4fe1\u53f7&#xff08;\u4f8b\u5982&#xff0c;\u901a\u8fc7\u9884\u6d4b\u88ab\u906e\u76d6\u7684\u8bcd&#xff09;\u3002\u901a\u8fc7\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u6a21\u578b\u88ab\u8feb\u53bb\u5b66\u4e60\u8bed\u8a00\u5185\u5728\u7684\u3001\u901a\u7528\u7684\u89c4\u5f8b&#xff0c;\u5305\u62ec\u8bed\u6cd5\u7ed3\u6784\u3001\u8bed\u4e49\u4fe1\u606f\u3001\u4e0a\u4e0b\u6587\u5173\u7cfb&#xff0c;\u4e43\u81f3\u6d77\u91cf\u7684\u4e16\u754c\u77e5\u8bc6\u548c\u5e38\u8bc6\u3002\u8fd9\u4e2a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5c31\u50cf\u4e00\u4e2a\u5df2\u7ecf\u8bfb\u8fc7\u201c\u4e07\u5377\u4e66\u201d\u7684\u3001\u77e5\u8bc6\u6e0a\u535a\u7684\u201c\u901a\u624d\u201d\u3002<\/li>\n<li>\u5fae\u8c03&#xff08;Fine-tuning&#xff09;\u9636\u6bb5&#xff1a;\u5f53\u6211\u4eec\u9700\u8981\u89e3\u51b3\u4e00\u4e2a\u5177\u4f53\u7684\u4e0b\u6e38\u4efb\u52a1\u65f6&#xff08;\u5982\u6587\u672c\u5206\u7c7b\u3001\u95ee\u7b54\u7cfb\u7edf&#xff09;&#xff0c;\u6211\u4eec\u4e0d\u518d\u4ece\u96f6\u5f00\u59cb\u3002\u6211\u4eec\u76f4\u63a5\u52a0\u8f7d\u8fd9\u4e2a\u9884\u8bad\u7ec3\u597d\u7684\u201c\u901a\u624d\u201d\u6a21\u578b&#xff0c;\u5e76\u5728\u5176\u9876\u90e8\u6dfb\u52a0\u4e00\u4e2a\u7b80\u5355\u7684\u3001\u4efb\u52a1\u76f8\u5173\u7684\u8f93\u51fa\u5c42\u3002\u7136\u540e&#xff0c;\u6211\u4eec\u7528\u5c11\u91cf\u7684\u3001\u6709\u6807\u7b7e\u7684\u4efb\u52a1\u6570\u636e&#xff0c;\u5bf9\u6a21\u578b\u7684\u53c2\u6570\u8fdb\u884c\u201c\u5fae\u8c03\u201d&#xff0c;\u4f7f\u5176\u9002\u5e94\u7279\u5b9a\u4efb\u52a1\u7684\u9700\u6c42\u3002\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u50cf\u662f\u8ba9\u4e00\u4f4d\u535a\u5b66\u7684\u901a\u624d&#xff0c;\u9488\u5bf9\u4e00\u4e2a\u4e13\u4e1a\u9886\u57df\u8fdb\u884c\u77ed\u6682\u7684\u5c97\u524d\u57f9\u8bad&#xff0c;\u4ed6\u80fd\u5f88\u5feb\u4e0a\u624b\u5e76\u6210\u4e3a\u4e13\u5bb6\u3002<\/li>\n<p>\u8fd9\u4e00\u8303\u5f0f\u7684\u8f6c\u53d8&#xff0c;\u5f7b\u5e95\u89e3\u51b3\u4e86\u4f20\u7edf\u65b9\u6cd5\u7684\u4e24\u5927\u56f0\u5883\u3002\u5b83\u6781\u5927\u5730\u964d\u4f4e\u4e86\u5bf9\u4e0b\u6e38\u4efb\u52a1\u6807\u6ce8\u6570\u636e\u7684\u4f9d\u8d56&#xff0c;\u5e76\u4f7f\u5f97\u6a21\u578b\u4e4b\u95f4\u80fd\u591f\u5171\u4eab\u548c\u8fc1\u79fb\u4ece\u6d77\u91cf\u6570\u636e\u4e2d\u5b66\u5230\u7684\u901a\u7528\u8bed\u8a00\u77e5\u8bc6&#xff0c;\u6807\u5fd7\u7740NLP\u8fdb\u5165\u4e86\u4e00\u4e2a\u5168\u65b0\u7684\u5de5\u4e1a\u5316\u65f6\u4ee3\u3002<\/p>\n<h5>9.3.2 BERT&#xff1a;\u6df1\u9083\u7684\u201c\u5b8c\u5f62\u586b\u7a7a\u201d\u5927\u5e08<\/h5>\n<p>**BERT&#xff08;Bidirectional Encoder Representations from Transformers&#xff09;**\u7531Google\u57282018\u5e74\u53d1\u5e03&#xff0c;\u5b83\u7684\u51fa\u73b0&#xff0c;\u50cf\u4e00\u573a\u98ce\u66b4\u5e2d\u5377\u4e86\u6574\u4e2aNLP\u9886\u57df&#xff0c;\u5728\u51e0\u4e4e\u6240\u6709\u7684NLP\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u90fd\u5237\u65b0\u4e86\u8bb0\u5f55\u3002BERT\u6210\u529f\u7684\u79d8\u8bc0&#xff0c;\u5728\u4e8e\u5176\u5bf9\u201c\u6df1\u5ea6\u53cc\u5411\u201d\u7406\u89e3\u7684\u6781\u81f4\u8ffd\u6c42\u3002<\/p>\n<p>\u6838\u5fc3\u67b6\u6784&#xff1a;\u53ea\u53d6Transformer\u7684\u201c\u7406\u89e3\u8005\u201d BERT\u7684\u67b6\u6784\u975e\u5e38\u7eaf\u7cb9&#xff1a;\u5b83\u53ea\u4f7f\u7528\u4e86Transformer\u7684\u7f16\u7801\u5668&#xff08;Encoder&#xff09;\u90e8\u5206\u3002\u5b83\u662f\u4e00\u4e2a\u5929\u751f\u7684\u201c\u7406\u89e3\u8005\u201d\u800c\u975e\u201c\u751f\u6210\u8005\u201d\u3002\u5b83\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u4e3a\u8f93\u5165\u7684\u4efb\u610f\u6587\u672c&#xff0c;\u751f\u6210\u4e00\u4e2a\u6df1\u5ea6\u878d\u5408\u4e86\u4e0a\u4e0b\u6587\u4fe1\u606f\u7684\u3001\u9ad8\u8d28\u91cf\u7684\u5411\u91cf\u8868\u793a\u3002<\/p>\n<p>\u9884\u8bad\u7ec3\u4efb\u52a1&#xff1a;\u8feb\u4f7f\u6a21\u578b\u7406\u89e3\u53cc\u5411\u8bed\u5883 BERT\u4e4b\u6240\u4ee5\u5f3a\u5927&#xff0c;\u5173\u952e\u5728\u4e8e\u5176\u4e24\u4e2a\u72ec\u521b\u7684\u9884\u8bad\u7ec3\u4efb\u52a1&#xff1a;<\/p>\n<li>\n<p>\u63a9\u7801\u8bed\u8a00\u6a21\u578b&#xff08;Masked Language Model, MLM&#xff09;&#xff1a;\u8fd9\u662fBERT\u7684\u7075\u9b42\u3002\u4f20\u7edf\u7684\u8bed\u8a00\u6a21\u578b&#xff08;\u5982GPT\u7684\u524d\u8eab&#xff09;\u90fd\u662f\u5355\u5411\u7684&#xff0c;\u53ea\u80fd\u6839\u636e\u5de6\u4fa7\u7684\u6587\u672c\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u3002\u8fd9\u79cd\u5355\u5411\u6027\u9650\u5236\u4e86\u6a21\u578b\u5bf9\u8bed\u5883\u7684\u7406\u89e3\u3002BERT\u5219\u5f00\u521b\u6027\u5730\u4f7f\u7528\u4e86\u201c\u5b8c\u5f62\u586b\u7a7a\u201d\u7684\u65b9\u5f0f\u3002\u5728\u9884\u8bad\u7ec3\u65f6&#xff0c;\u5b83\u968f\u673a\u5730\u5c06\u8f93\u5165\u53e5\u5b50\u4e2d15%\u7684\u8bcd\u5143&#xff08;Token&#xff09;\u7528\u4e00\u4e2a\u7279\u6b8a\u7684 [MASK] \u6807\u8bb0\u66ff\u6362\u6389&#xff0c;\u7136\u540e&#xff0c;\u6a21\u578b\u7684\u4efb\u52a1\u5c31\u662f\u4ec5\u6839\u636e\u672a\u88ab\u906e\u76d6\u7684\u4e0a\u4e0b\u6587&#xff0c;\u53bb\u9884\u6d4b\u8fd9\u4e9b\u88ab\u906e\u76d6\u4f4f\u7684\u539f\u59cb\u8bcd\u5143\u3002 \u4f8b\u5982&#xff0c;\u8f93\u5165 My dog is [MASK] and hairy.&#xff0c;\u6a21\u578b\u9700\u8981\u9884\u6d4b\u51fa [MASK] \u7684\u4f4d\u7f6e\u662f cute\u3002\u4e3a\u4e86\u505a\u51fa\u6b63\u786e\u7684\u9884\u6d4b&#xff0c;\u6a21\u578b\u5fc5\u987b\u540c\u65f6\u5173\u6ce8\u5de6\u4fa7\u7684 My dog is \u548c\u53f3\u4fa7\u7684 and hairy\u3002\u8fd9\u79cd\u673a\u5236&#xff0c;\u8feb\u4f7fBERT\u5fc5\u987b\u5b66\u4e60\u5230\u771f\u6b63\u7684\u3001\u6df1\u5ea6\u7684\u53cc\u5411\u8bed\u5883\u8868\u793a\u3002\u8fd9\u4e0e\u4eba\u7c7b\u505a\u9605\u8bfb\u7406\u89e3\u9898\u7684\u65b9\u5f0f\u4f55\u5176\u76f8\u4f3c\u3002<\/p>\n<\/li>\n<li>\n<p>\u4e0b\u4e00\u53e5\u9884\u6d4b&#xff08;Next Sentence Prediction, NSP&#xff09;&#xff1a;\u4e3a\u4e86\u8ba9\u6a21\u578b\u7406\u89e3\u53e5\u5b50\u4e0e\u53e5\u5b50\u4e4b\u95f4\u7684\u903b\u8f91\u5173\u7cfb&#xff08;\u8fd9\u5bf9\u4e8e\u95ee\u7b54\u3001\u5bf9\u8bdd\u7b49\u4efb\u52a1\u81f3\u5173\u91cd\u8981&#xff09;&#xff0c;BERT\u8fd8\u8bbe\u8ba1\u4e86NSP\u4efb\u52a1\u3002\u5728\u9884\u8bad\u7ec3\u65f6&#xff0c;\u6a21\u578b\u4f1a\u63a5\u6536\u4e00\u5bf9\u53e5\u5b50(A, B)&#xff0c;\u5e76\u9700\u8981\u5224\u65ad\u53e5\u5b50B\u662f\u5426\u662f\u53e5\u5b50A\u5728\u539f\u59cb\u6587\u672c\u4e2d\u7684\u4e0b\u4e00\u53e5\u300250%\u7684\u60c5\u51b5\u4e0bB\u662f\u771f\u5b9e\u7684\u4e0b\u4e00\u53e5&#xff0c;\u53e6\u591650%\u5219\u662f\u4ece\u8bed\u6599\u5e93\u4e2d\u968f\u673a\u9009\u62e9\u7684\u4e00\u4e2a\u53e5\u5b50\u3002\u8fd9\u4e2a\u4efb\u52a1&#xff0c;\u8ba9BERT\u5b66\u4f1a\u4e86\u6355\u6349\u7bc7\u7ae0\u7ea7\u522b\u7684\u8fde\u8d2f\u6027\u3002<\/p>\n<\/li>\n<p>\u901a\u8fc7\u8fd9\u4e24\u4e2a\u4efb\u52a1\u7684\u8054\u5408\u8bad\u7ec3&#xff0c;BERT\u6210\u4e3a\u4e86\u4e00\u4e2a\u5bf9\u8bed\u8a00\u6709\u7740\u6df1\u523b\u53cc\u5411\u7406\u89e3\u80fd\u529b\u7684\u6a21\u578b\u3002\u5728\u5fae\u8c03\u9636\u6bb5&#xff0c;\u6211\u4eec\u53ef\u4ee5\u6839\u636e\u4e0d\u540c\u7684\u4efb\u52a1&#xff0c;\u5bf9BERT\u7684\u8f93\u51fa\u8fdb\u884c\u4e0d\u540c\u7684\u5229\u7528\u3002\u4f8b\u5982&#xff0c;\u5bf9\u4e8e\u53e5\u5b50\u5206\u7c7b\u4efb\u52a1&#xff0c;\u6211\u4eec\u53ef\u4ee5\u53d6\u5176\u7b2c\u4e00\u4e2a\u7279\u6b8a\u6807\u8bb0 [CLS] \u7684\u6700\u7ec8\u8f93\u51fa\u5411\u91cf&#xff0c;\u9001\u5165\u4e00\u4e2a\u5206\u7c7b\u5668&#xff1b;\u5bf9\u4e8e\u5e8f\u5217\u6807\u6ce8\u4efb\u52a1&#xff0c;\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u6bcf\u4e2a\u8bcd\u5143\u7684\u6700\u7ec8\u8f93\u51fa\u5411\u91cf\u3002<\/p>\n<h5>9.3.3 GPT&#xff1a;\u96c4\u8fa9\u7684\u201c\u6587\u5b57\u63a5\u9f99\u201d\u5929\u624d<\/h5>\n<p>\u4e0eBERT\u540c\u5e74&#xff0c;\u7531OpenAI\u53d1\u5e03\u7684**GPT&#xff08;Generative Pre-trained Transformer&#xff09;**\u5219\u8d70\u4e86\u53e6\u4e00\u6761\u622a\u7136\u4e0d\u540c\u7684\u6280\u672f\u8def\u7ebf&#xff0c;\u5e76\u6700\u7ec8\u5f00\u542f\u4e86\u901a\u5f80AIGC&#xff08;\u4eba\u5de5\u667a\u80fd\u751f\u6210\u5185\u5bb9&#xff09;\u7684\u5eb7\u5e84\u5927\u9053\u3002<\/p>\n<p>\u6838\u5fc3\u67b6\u6784&#xff1a;\u53ea\u53d6Transformer\u7684\u201c\u521b\u9020\u8005\u201d \u4e0eBERT\u76f8\u53cd&#xff0c;GPT\u7cfb\u5217\u7684\u67b6\u6784\u53ea\u4f7f\u7528\u4e86Transformer\u7684\u89e3\u7801\u5668&#xff08;Decoder&#xff09;\u90e8\u5206\u3002\u5b83\u662f\u4e00\u4e2a\u5929\u751f\u7684\u201c\u751f\u6210\u8005\u201d\u6216\u201c\u521b\u9020\u8005\u201d\u3002\u5b83\u7684\u6838\u5fc3\u80fd\u529b&#xff0c;\u5728\u4e8e\u6839\u636e\u7ed9\u5b9a\u7684\u4e0a\u6587&#xff0c;\u53bb\u9884\u6d4b\u548c\u751f\u6210\u6700\u6709\u53ef\u80fd\u7684\u4e0b\u4e00\u4e2a\u8bcd\u3002<\/p>\n<p>\u9884\u8bad\u7ec3\u4efb\u52a1&#xff1a;\u7ecf\u5178\u7684\u81ea\u56de\u5f52\u8bed\u8a00\u6a21\u578b GPT\u7684\u9884\u8bad\u7ec3\u4efb\u52a1\u975e\u5e38\u7ecf\u5178\u548c\u76f4\u63a5&#xff0c;\u5b83\u5c31\u662f\u4e00\u4e2a\u6807\u51c6\u7684\u81ea\u56de\u5f52\u8bed\u8a00\u6a21\u578b&#xff08;Autoregressive Language Model&#xff09;\u3002\u5176\u76ee\u6807\u6c38\u8fdc\u662f&#xff1a;\u6839\u636e\u6240\u6709\u5df2\u7ecf\u51fa\u73b0\u7684\u3001\u5de6\u4fa7\u7684\u6587\u672c&#xff0c;\u6765\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u5143\u3002 \u4f8b\u5982&#xff0c;\u7ed9\u5b9a\u4e0a\u6587 The cat sat on the&#xff0c;\u6a21\u578b\u9700\u8981\u9884\u6d4b\u51fa\u4e0b\u4e00\u4e2a\u8bcd\u662f mat \u7684\u6982\u7387\u6700\u9ad8\u3002 \u7531\u4e8eGPT\u91c7\u7528\u4e86Transformer\u7684\u89e3\u7801\u5668\u7ed3\u6784&#xff0c;\u5176\u5185\u90e8\u7684\u201c\u5e26\u63a9\u7801\u7684\u81ea\u6ce8\u610f\u529b\u201d\u673a\u5236&#xff0c;\u5929\u7136\u5730\u4fdd\u8bc1\u4e86\u5728\u9884\u6d4b\u7b2c i \u4e2a\u8bcd\u65f6&#xff0c;\u6a21\u578b\u53ea\u80fd\u770b\u5230\u7b2c i-1 \u4e2a\u8bcd\u53ca\u5176\u4e4b\u524d\u7684\u5185\u5bb9&#xff0c;\u800c\u65e0\u6cd5\u201c\u5077\u770b\u201d\u672a\u6765\u7684\u4fe1\u606f\u3002\u8fd9\u79cd\u4e25\u683c\u7684\u5355\u5411\u6027&#xff08;Uni-directional&#xff09;&#xff0c;\u4f7f\u5f97GPT\u5929\u751f\u5c31\u975e\u5e38\u64c5\u957f\u4e8e\u6267\u884c\u6587\u672c\u751f\u6210\u4efb\u52a1&#xff0c;\u4ece\u5199\u6545\u4e8b\u3001\u5199\u4ee3\u7801\u5230\u56de\u7b54\u95ee\u9898&#xff0c;\u672c\u8d28\u4e0a\u90fd\u662f\u5728\u505a\u201c\u6587\u5b57\u63a5\u9f99\u201d\u3002<\/p>\n<p>BERT vs. GPT&#xff1a;\u54f2\u5b66\u7684\u5206\u6b67<\/p>\n<ul>\n<li>BERT\u662f\u201c\u53cc\u5411\u7f16\u7801\u5668\u201d&#xff0c;\u50cf\u4e00\u4e2a\u9605\u8bfb\u7406\u89e3\u5927\u5e08\u3002\u5b83\u7684\u5f3a\u9879\u5728\u4e8e\u7406\u89e3&#xff0c;\u4e3a\u8f93\u5165\u7684\u6587\u672c\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u3001\u5bcc\u542b\u4e0a\u4e0b\u6587\u7684\u8868\u793a&#xff0c;\u7279\u522b\u9002\u5408\u4e8e**\u81ea\u7136\u8bed\u8a00\u7406\u89e3&#xff08;NLU&#xff09;**\u4efb\u52a1&#xff0c;\u5982\u5206\u7c7b\u3001\u5b9e\u4f53\u8bc6\u522b\u3001\u60c5\u611f\u5206\u6790\u7b49\u3002<\/li>\n<li>GPT\u662f\u201c\u5355\u5411\u89e3\u7801\u5668\u201d&#xff0c;\u50cf\u4e00\u4e2a\u5199\u4f5c\u5929\u624d\u3002\u5b83\u7684\u5f3a\u9879\u5728\u4e8e\u751f\u6210&#xff0c;\u6839\u636e\u7ed9\u5b9a\u7684\u63d0\u793a&#xff08;Prompt&#xff09;\u521b\u9020\u51fa\u8fde\u8d2f\u3001\u6d41\u7545\u7684\u6587\u672c&#xff0c;\u662f**\u81ea\u7136\u8bed\u8a00\u751f\u6210&#xff08;NLG&#xff09;**\u4efb\u52a1\u7684\u738b\u8005\u3002<\/li>\n<\/ul>\n<h5>9.3.4 \u6a21\u578b\u7684\u6f14\u8fdb\u4e0e\u201c\u6d8c\u73b0\u201d&#xff1a;\u4ece\u5fae\u8c03\u5230\u63d0\u793a&#xff0c;\u4ece\u91cf\u53d8\u5230\u8d28\u53d8<\/h5>\n<p>BERT\u548cGPT\u7684\u51fa\u73b0&#xff0c;\u53ea\u662f\u8fd9\u573a\u9769\u547d\u7684\u5e8f\u7ae0\u3002\u968f\u540e\u7684\u51e0\u5e74&#xff0c;\u6574\u4e2a\u9886\u57df\u8fdb\u5165\u4e86\u4e00\u573a\u4ee5\u201c\u89c4\u6a21\u5b9a\u5f8b&#xff08;Scaling Laws&#xff09;\u201d\u4e3a\u4e3b\u5bfc\u7684\u519b\u5907\u7ade\u8d5b\u3002\u7814\u7a76\u8005\u53d1\u73b0&#xff0c;\u6a21\u578b\u7684\u6027\u80fd&#xff0c;\u4e0e\u5176\u53c2\u6570\u89c4\u6a21\u3001\u6570\u636e\u91cf\u3001\u8ba1\u7b97\u91cf\u4e4b\u95f4&#xff0c;\u5b58\u5728\u7740\u53ef\u9884\u6d4b\u7684\u5e42\u5f8b\u5173\u7cfb\u3002\u6a21\u578b\u8d8a\u5927&#xff0c;\u4f3c\u4e4e\u5c31\u8d8a\u201c\u806a\u660e\u201d\u3002<\/p>\n<p>\u6a21\u578b\u5373\u670d\u52a1&#xff08;Model as a Service&#xff09;\u4e0e\u63d0\u793a&#xff08;Prompting&#xff09; \u968f\u7740\u6a21\u578b&#xff08;\u7279\u522b\u662fGPT\u7cfb\u5217&#xff0c;\u5982GPT-3, GPT-4&#xff09;\u7684\u53c2\u6570\u89c4\u6a21\u4ece\u6570\u4ebf\u81a8\u80c0\u5230\u6570\u5343\u4ebf\u751a\u81f3\u4e07\u4ebf&#xff0c;\u5fae\u8c03\u6574\u4e2a\u6a21\u578b\u7684\u6210\u672c\u53d8\u5f97\u6781\u5176\u9ad8\u6602&#xff0c;\u751a\u81f3\u4e0d\u518d\u5fc5\u8981\u3002\u4e00\u79cd\u5168\u65b0\u7684\u3001\u66f4\u8f7b\u4fbf\u7684\u4ea4\u4e92\u8303\u5f0f\u2014\u2014\u63d0\u793a&#xff08;Prompting&#xff09;\u5e94\u8fd0\u800c\u751f\u3002 \u7528\u6237\u53d1\u73b0&#xff0c;\u5bf9\u4e8e\u8fd9\u4e9b\u8d85\u5927\u89c4\u6a21\u7684\u8bed\u8a00\u6a21\u578b&#xff08;Large Language Models, LLMs&#xff09;&#xff0c;\u6211\u4eec\u4e0d\u518d\u9700\u8981\u7528\u6210\u5343\u4e0a\u4e07\u7684\u6807\u6ce8\u6570\u636e\u53bb\u5fae\u8c03\u5b83\u3002\u6211\u4eec\u53ea\u9700\u8981\u5728\u8f93\u5165\u7aef&#xff0c;\u7528\u81ea\u7136\u8bed\u8a00\u7ed9\u51fa\u6e05\u6670\u7684\u6307\u4ee4&#xff08;Instruction&#xff09;\u6216\u4e0a\u4e0b\u6587\u793a\u4f8b&#xff08;In-context Learning&#xff09;&#xff0c;\u6a21\u578b\u5c31\u80fd\u76f4\u63a5\u7406\u89e3\u6211\u4eec\u7684\u610f\u56fe&#xff0c;\u5e76\u5b8c\u6210\u76f8\u5e94\u7684\u4efb\u52a1\u3002 \u4f8b\u5982&#xff0c;\u8981\u8fdb\u884c\u60c5\u611f\u5206\u6790&#xff0c;\u6211\u4eec\u4e0d\u518d\u9700\u8981\u4e00\u4e2a\u60c5\u611f\u5206\u7c7b\u6570\u636e\u96c6&#xff0c;\u53ea\u9700\u5411\u6a21\u578b\u63d0\u95ee&#xff1a; &#034;\u8bf7\u5224\u65ad\u4ee5\u4e0b\u8bc4\u8bba\u7684\u60c5\u611f\u662f\u6b63\u9762\u7684\u8fd8\u662f\u8d1f\u9762\u7684&#xff1a;&#039;\u8fd9\u90e8\u7535\u5f71\u771f\u662f\u592a\u68d2\u4e86&#xff01;&#039;&#034; \u6a21\u578b\u5c31\u80fd\u76f4\u63a5\u56de\u7b54&#xff1a;\u201c\u6b63\u9762\u7684\u201d\u3002\u8fd9\u79cd\u201c\u6a21\u578b\u5373\u670d\u52a1\u201d\u7684\u8303\u5f0f&#xff0c;\u6781\u5927\u5730\u964d\u4f4e\u4e86AI\u7684\u5e94\u7528\u95e8\u69db\u3002<\/p>\n<p>\u80fd\u529b\u7684\u6d8c\u73b0&#xff08;Emergent Abilities&#xff09;&#xff1a;\u5f53\u91cf\u53d8\u5f15\u53d1\u8d28\u53d8 \u66f4\u4ee4\u4eba\u9707\u60ca\u7684\u73b0\u8c61\u662f**\u201c\u6d8c\u73b0\u201d**\u3002\u7814\u7a76\u8005\u53d1\u73b0&#xff0c;\u5f53\u8bed\u8a00\u6a21\u578b\u7684\u89c4\u6a21\u8de8\u8d8a\u67d0\u4e2a\u5de8\u5927\u7684\u9608\u503c\u4e4b\u540e&#xff0c;\u4f1a\u7a81\u7136\u8868\u73b0\u51fa\u5728\u5c0f\u89c4\u6a21\u6a21\u578b\u4e0a\u5b8c\u5168\u4e0d\u5b58\u5728\u7684\u3001\u4ee4\u4eba\u60ca\u53f9\u7684\u65b0\u80fd\u529b\u3002\u8fd9\u4e9b\u80fd\u529b\u4f3c\u4e4e\u4e0d\u662f\u88ab\u76f4\u63a5\u201c\u6559\u4f1a\u201d\u7684&#xff0c;\u800c\u662f\u4ece\u6d77\u91cf\u6570\u636e\u548c\u5de8\u5927\u89c4\u6a21\u4e2d\u201c\u81ea\u53d1\u5730\u201d\u6d8c\u73b0\u51fa\u6765\u7684\u3002 \u5178\u578b\u7684\u6d8c\u73b0\u80fd\u529b\u5305\u62ec&#xff1a;<\/p>\n<ul>\n<li>\u601d\u7ef4\u94fe&#xff08;Chain-of-Thought, CoT&#xff09;\u63a8\u7406&#xff1a;\u6a21\u578b\u4e0d\u518d\u53ea\u662f\u7ed9\u51fa\u7b54\u6848&#xff0c;\u800c\u662f\u80fd\u50cf\u4eba\u7c7b\u4e00\u6837&#xff0c;\u901a\u8fc7\u4e00\u6b65\u6b65\u7684\u903b\u8f91\u63a8\u5bfc\u6765\u89e3\u51b3\u590d\u6742\u95ee\u9898\u3002<\/li>\n<li>\u9ad8\u7ea7\u4ee3\u7801\u751f\u6210&#xff1a;\u80fd\u591f\u7406\u89e3\u590d\u6742\u7684\u7f16\u7a0b\u9700\u6c42&#xff0c;\u5e76\u751f\u6210\u9ad8\u8d28\u91cf\u3001\u53ef\u8fd0\u884c\u7684\u4ee3\u7801\u3002<\/li>\n<li>\u4e16\u754c\u77e5\u8bc6\u7684\u8fd0\u7528&#xff1a;\u80fd\u591f\u56de\u7b54\u9700\u8981\u6df1\u5165\u3001\u591a\u9886\u57df\u77e5\u8bc6\u624d\u80fd\u89e3\u7b54\u7684\u95ee\u9898\u3002<\/li>\n<\/ul>\n<p>\u201c\u6d8c\u73b0\u201d\u73b0\u8c61\u662f\u5f53\u524dAI\u9886\u57df\u6700\u524d\u6cbf\u3001\u4e5f\u6700\u795e\u79d8\u7684\u7814\u7a76\u65b9\u5411\u4e4b\u4e00\u3002\u5b83\u8868\u660e&#xff0c;\u6211\u4eec\u53ef\u80fd\u6b63\u5728\u4ece\u201c\u8bad\u7ec3\u6a21\u578b\u89e3\u51b3\u7279\u5b9a\u4efb\u52a1\u201d\u7684\u65f6\u4ee3&#xff0c;\u8fc8\u5411\u4e00\u4e2a\u201c\u4e0e\u901a\u7528\u4eba\u5de5\u667a\u80fd\u96cf\u5f62\u8fdb\u884c\u4ea4\u4e92\u201d\u7684\u65b0\u65f6\u4ee3\u3002BERT\u548cGPT&#xff0c;\u6b63\u662f\u5f00\u542f\u8fd9\u6247\u5927\u95e8\u7684\u4e24\u628a\u94a5\u5319\u3002<\/p>\n<h4>9.4 Transformer\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u7684\u5e94\u7528&#xff1a;\u65b0\u4e00\u8f6e\u7684\u7edf\u4e00<\/h4>\n<p>\u5728BERT\u548cGPT\u5f15\u9886\u7684\u6d6a\u6f6e\u5e2d\u5377NLP\u4e4b\u540e&#xff0c;\u4e00\u4e2a\u81ea\u7136\u800c\u7136\u7684\u95ee\u9898\u6d6e\u73b0\u5728\u4e86\u7814\u7a76\u8005\u4eec\u7684\u8111\u6d77\u4e2d&#xff1a;Transformer\u90a3\u5f3a\u5927\u7684\u3001\u901a\u7528\u7684\u5e8f\u5217\u5efa\u6a21\u80fd\u529b&#xff0c;\u80fd\u5426\u88ab\u8fc1\u79fb\u5230\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df&#xff1f;\u8fd9\u4e2a\u95ee\u9898\u5728\u5f53\u65f6\u770b\u6765&#xff0c;\u65e2\u8bf1\u4eba\u53c8\u5145\u6ee1\u6311\u6218\u3002\u56e0\u4e3aCV\u9886\u57df\u65e9\u5df2\u88ab\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u201c\u7edf\u6cbb\u201d\u4e86\u8fd1\u5341\u5e74\u3002CNN\u7684\u6210\u529f&#xff0c;\u5f88\u5927\u7a0b\u5ea6\u4e0a\u6e90\u4e8e\u5176\u7cbe\u5de7\u7684\u5f52\u7eb3\u504f\u7f6e&#xff08;Inductive Bias&#xff09;&#xff0c;\u8fd9\u4f7f\u5176\u5728\u5b66\u4e60\u89c6\u89c9\u4efb\u52a1\u65f6&#xff0c;\u62e5\u6709\u4e86\u5929\u7136\u7684\u201c\u5148\u53d1\u4f18\u52bf\u201d\u3002<\/p>\n<p>CNN\u7684\u5f52\u7eb3\u504f\u7f6e&#xff1a;\u4e3a\u89c6\u89c9\u800c\u751f\u7684\u201c\u4e13\u5bb6\u201d<\/p>\n<ul>\n<li>\u5c40\u90e8\u6027&#xff08;Locality&#xff09;&#xff1a;CNN\u901a\u8fc7\u5c0f\u7684\u5377\u79ef\u6838&#xff0c;\u5047\u8bbe\u4e86\u56fe\u50cf\u4e2d\u7684\u4fe1\u606f\u662f\u5c40\u90e8\u76f8\u5173\u7684\u3002\u4e00\u4e2a\u50cf\u7d20\u7684\u610f\u4e49&#xff0c;\u4e3b\u8981\u7531\u5176\u5468\u56f4\u7684\u50cf\u7d20\u51b3\u5b9a\u3002<\/li>\n<li>\u5e73\u79fb\u4e0d\u53d8\u6027&#xff08;Translation Invariance&#xff09;&#xff1a;\u4e00\u4e2a\u5728\u56fe\u50cf\u5de6\u4e0a\u89d2\u88ab\u8bc6\u522b\u4e3a\u201c\u732b\u201d\u7684\u6a21\u5f0f&#xff0c;\u5728\u53f3\u4e0b\u89d2\u540c\u6837\u80fd\u88ab\u8bc6\u522b\u51fa\u6765&#xff0c;\u56e0\u4e3a\u5377\u79ef\u6838\u5728\u6574\u4e2a\u56fe\u50cf\u4e0a\u662f\u5171\u4eab\u7684\u3002 \u8fd9\u4e9b\u504f\u7f6e\u4e0e\u6211\u4eec\u5bf9\u7269\u7406\u4e16\u754c\u7684\u76f4\u89c2\u8ba4\u77e5\u9ad8\u5ea6\u4e00\u81f4&#xff0c;\u4f7f\u5f97CNN\u80fd\u591f\u4ee5\u6781\u9ad8\u7684\u6570\u636e\u6548\u7387&#xff08;Data Efficiency&#xff09;\u5b66\u4e60\u5230\u5f3a\u5927\u7684\u89c6\u89c9\u8868\u793a\u3002\u5b83\u5c31\u50cf\u4e00\u4e2a\u5929\u751f\u7684\u89c6\u89c9\u4e13\u5bb6&#xff0c;\u81ea\u5e26\u4e00\u5957\u7406\u89e3\u4e16\u754c\u7684\u65b9\u6cd5\u8bba\u3002<\/li>\n<\/ul>\n<p>\u800cTransformer&#xff0c;\u5219\u662f\u4e00\u4e2a\u201c\u65e0\u504f\u89c1\u201d\u7684\u3001\u66f4\u52a0\u901a\u7528\u7684\u5b66\u4e60\u5668\u3002\u5b83\u4e0d\u5bf9\u8f93\u5165\u7684\u5143\u7d20\u505a\u4efb\u4f55\u7a7a\u95f4\u4e0a\u7684\u5047\u8bbe&#xff0c;\u53ea\u662f\u4e00\u89c6\u540c\u4ec1\u5730\u5b66\u4e60\u5b83\u4eec\u4e4b\u95f4\u7684\u5173\u8054\u3002\u8ba9\u8fd9\u6837\u4e00\u4e2a\u201c\u901a\u624d\u201d\u53bb\u5904\u7406\u56fe\u50cf&#xff0c;\u65e0\u5f02\u4e8e\u4e00\u573a\u8c6a\u8d4c\u3002<\/p>\n<h5>9.4.1 Vision Transformer (ViT)&#xff1a;\u5f53\u56fe\u50cf\u88ab\u89c6\u4e3a\u201c\u5355\u8bcd\u201d\u5e8f\u5217<\/h5>\n<p>2020\u5e74&#xff0c;Google\u7684\u7814\u7a76\u8005\u4eec\u53d1\u8868\u4e86\u8bba\u6587\u300aAn Image is Worth 16&#215;16 Words: Transformers for Image Recognition at Scale\u300b&#xff0c;\u6b63\u5f0f\u63a8\u51fa\u4e86Vision Transformer (ViT)\u3002\u8fd9\u7bc7\u8bba\u6587\u7684\u8bbe\u8ba1&#xff0c;\u5145\u6ee1\u4e86\u5927\u9053\u81f3\u7b80\u7684\u667a\u6167\u4e0e\u52c7\u6c14\u3002\u5b83\u6ca1\u6709\u8bd5\u56fe\u5c06\u5377\u79ef\u4e0eTransformer\u8fdb\u884c\u590d\u6742\u7684\u878d\u5408&#xff0c;\u800c\u662f\u505a\u51fa\u4e86\u4e00\u4e2a\u6781\u5176\u5927\u80c6\u7684\u5ba3\u8a00&#xff1a;\u6211\u4eec\u53ef\u4ee5\u5b8c\u5168\u629b\u5f03\u5377\u79ef&#xff0c;\u5c06\u56fe\u50cf\u76f4\u63a5\u89c6\u4e3a\u4e00\u4e2a\u201c\u5355\u8bcd\u201d\u5e8f\u5217&#xff0c;\u7136\u540e\u7528\u4e00\u4e2a\u6807\u51c6\u7684Transformer\u6765\u5904\u7406\u5b83\u3002<\/p>\n<p>\u6838\u5fc3\u601d\u60f3&#xff1a;\u4ece\u50cf\u7d20\u7f51\u683c\u5230\u201c\u56fe\u50cf\u8bcd\u201d\u5e8f\u5217 ViT\u7684\u5de5\u4f5c\u6d41\u7a0b&#xff0c;\u4f18\u96c5\u5730\u5c06\u56fe\u50cf\u6570\u636e\u201c\u7ffb\u8bd1\u201d\u6210\u4e86Transformer\u80fd\u591f\u7406\u89e3\u7684\u8bed\u8a00&#xff1a;<\/p>\n<li>\n<p>\u7b2c\u4e00\u6b65&#xff1a;\u56fe\u50cf\u5206\u5757&#xff08;Image Patching&#xff09; \u8fd9\u662fViT\u6700\u6838\u5fc3\u7684\u3001\u4e5f\u662f\u552f\u4e00\u7684\u201c\u56fe\u50cf\u7279\u5b9a\u201d\u9884\u5904\u7406\u6b65\u9aa4\u3002\u5b83\u5c06\u4e00\u5f20\u8f93\u5165\u7684\u56fe\u50cf&#xff08;\u4f8b\u5982 224&#215;224 \u50cf\u7d20&#xff09;&#xff0c;\u5206\u5272\u6210\u4e00\u7cfb\u5217\u56fa\u5b9a\u5927\u5c0f\u7684\u3001\u4e0d\u91cd\u53e0\u7684\u5c0f\u65b9\u5757&#xff08;Patches&#xff09;\u3002\u4f8b\u5982&#xff0c;\u5982\u679c\u6bcf\u4e2aPatch\u7684\u5927\u5c0f\u662f 16&#215;16 \u50cf\u7d20&#xff0c;\u90a3\u4e48\u4e00\u5f20 224&#215;224 \u7684\u56fe\u50cf\u5c31\u4f1a\u88ab\u5206\u5272\u6210 (224\/16) * (224\/16) &#061; 14 * 14 &#061; 196 \u4e2aPatches\u3002 \u8fd9\u4e2a\u7b80\u5355\u7684\u64cd\u4f5c&#xff0c;\u5728\u6982\u5ff5\u4e0a\u5177\u6709\u9769\u547d\u6027\u7684\u610f\u4e49&#xff1a;\u5b83\u5c06\u4e00\u4e2a\u8fde\u7eed\u7684\u3001\u4e8c\u7ef4\u7684\u50cf\u7d20\u7f51\u683c&#xff0c;\u79bb\u6563\u5316\u6210\u4e86\u4e00\u4e2a\u4e00\u7ef4\u7684**\u201c\u56fe\u50cf\u8bcd&#xff08;Image Words&#xff09;\u201d\u5e8f\u5217**\u3002<\/p>\n<\/li>\n<li>\n<p>\u7b2c\u4e8c\u6b65&#xff1a;\u5757\u5d4c\u5165&#xff08;Patch Embedding&#xff09; \u4e0eNLP\u4e2d\u5c06\u5355\u8bcd\u6620\u5c04\u4e3a\u8bcd\u5411\u91cf\u7c7b\u4f3c&#xff0c;ViT\u5c06\u6bcf\u4e00\u4e2aPatch\u5c55\u5e73&#xff08;Flatten&#xff09;\u6210\u4e00\u4e2a\u4e00\u7ef4\u5411\u91cf&#xff08;\u4f8b\u5982&#xff0c;16x16x3 -&gt; 768&#xff09;&#xff0c;\u7136\u540e\u901a\u8fc7\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684\u7ebf\u6027\u6295\u5f71\u5c42&#xff08;Linear Projection&#xff09;&#xff0c;\u5c06\u5176\u5d4c\u5165\u5230\u4e00\u4e2a\u56fa\u5b9a\u7ef4\u5ea6\u7684\u5411\u91cf\u7a7a\u95f4\u4e2d&#xff08;\u4f8b\u5982&#xff0c;D&#061;768&#xff09;\u3002\u8fd9\u4e00\u6b65\u4e4b\u540e&#xff0c;\u6211\u4eec\u5c31\u5f97\u5230\u4e86\u4e00\u4e2a\u7531196\u4e2a\u5411\u91cf\u7ec4\u6210\u7684\u5e8f\u5217&#xff0c;\u8fd9\u5728\u5f62\u5f0f\u4e0a\u4e0eNLP\u4e2d\u7684\u8bcd\u5d4c\u5165\u5e8f\u5217\u5df2\u7ecf\u5b8c\u5168\u4e00\u6837\u4e86\u3002<\/p>\n<\/li>\n<li>\n<p>\u7b2c\u4e09\u6b65&#xff1a;\u878d\u5165\u4f4d\u7f6e\u4e0e\u5206\u7c7b\u4fe1\u606f<\/p>\n<ul>\n<li>\u4f4d\u7f6e\u7f16\u7801&#xff08;Positional Encoding&#xff09;&#xff1a;\u4e0eNLP\u4e2d\u7684Transformer\u4e00\u6837&#xff0c;\u4e3a\u4e86\u8ba9\u6a21\u578b\u611f\u77e5\u5230Patch\u4e4b\u95f4\u7684\u7a7a\u95f4\u987a\u5e8f\u5173\u7cfb&#xff0c;ViT\u4e3a\u6bcf\u4e00\u4e2aPatch\u5d4c\u5165\u5411\u91cf&#xff0c;\u90fd\u52a0\u4e0a\u4e86\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684&#xff08;\u6216\u56fa\u5b9a\u7684\u4e09\u89d2\u51fd\u6570&#xff09;\u4f4d\u7f6e\u7f16\u7801\u3002<\/li>\n<li>\u5206\u7c7b\u4ee4\u724c&#xff08;[CLS] Token&#xff09;&#xff1a;\u501f\u9274BERT\u7684\u8bbe\u8ba1&#xff0c;ViT\u5728Patch\u5e8f\u5217\u7684\u6700\u524d\u9762&#xff0c;\u989d\u5916\u62fc\u63a5\u4e0a\u4e86\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684\u5206\u7c7b\u4ee4\u724c&#xff08;Classification Token&#xff09;\u3002\u8fd9\u4e2a\u4ee4\u724c\u4e0d\u4ee3\u8868\u4efb\u4f55\u4e00\u4e2a\u5177\u4f53\u7684Patch&#xff0c;\u5b83\u7684\u4f5c\u7528&#xff0c;\u662f\u5728\u7ecf\u8fc7Transformer\u7f16\u7801\u5668\u4e4b\u540e&#xff0c;\u5176\u5bf9\u5e94\u7684\u6700\u7ec8\u8f93\u51fa\u5411\u91cf&#xff0c;\u5c06\u88ab\u7528\u4f5c\u6574\u4e2a\u56fe\u50cf\u7684\u5168\u5c40\u805a\u5408\u8868\u793a&#xff0c;\u76f4\u63a5\u9001\u5165\u6700\u7ec8\u7684\u5206\u7c7b\u5668\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7b2c\u56db\u6b65&#xff1a;\u6807\u51c6\u7684Transformer\u7f16\u7801\u5668 \u63a5\u4e0b\u6765\u7684\u6b65\u9aa4&#xff0c;\u5c31\u4e0e\u6211\u4eec\u719f\u6089\u7684BERT\u5b8c\u5168\u4e00\u81f4\u4e86\u3002\u8fd9\u4e2a\u5305\u542b\u4e86\u5206\u7c7b\u4ee4\u724c\u548c\u5e26\u4f4d\u7f6e\u7f16\u7801\u7684Patch\u5e8f\u5217&#xff0c;\u88ab\u76f4\u63a5\u8f93\u5165\u4e00\u4e2a\u6807\u51c6\u7684Transformer\u7f16\u7801\u5668\u4e2d\u3002\u7f16\u7801\u5668\u5185\u90e8\u7684\u591a\u5934\u81ea\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u4f1a\u53bb\u8ba1\u7b97\u6bcf\u4e00\u4e2aPatch\u4e0e\u6240\u6709\u5176\u4ed6Patches\u4e4b\u95f4\u7684\u5173\u8054\u5f3a\u5ea6&#xff0c;\u5e76\u636e\u6b64\u52a8\u6001\u5730\u66f4\u65b0\u6bcf\u4e2aPatch\u7684\u8868\u793a\u3002<\/p>\n<\/li>\n<li>\n<p>\u7b2c\u4e94\u6b65&#xff1a;\u5206\u7c7b \u5f53\u5e8f\u5217\u901a\u8fc7\u6574\u4e2a\u7f16\u7801\u5668\u540e&#xff0c;\u6211\u4eec\u53ea\u53d6\u51fa\u5206\u7c7b\u4ee4\u724c\u6240\u5bf9\u5e94\u7684\u6700\u7ec8\u8f93\u51fa\u5411\u91cf&#xff0c;\u5c06\u5176\u9001\u5165\u4e00\u4e2a\u7b80\u5355\u7684\u591a\u5c42\u611f\u77e5\u673a&#xff08;MLP Head&#xff09;&#xff0c;\u8fdb\u884c\u6700\u7ec8\u7684\u56fe\u50cf\u5206\u7c7b\u3002<\/p>\n<\/li>\n<p>ViT\u7684\u201c\u53cd\u76f4\u89c9\u201d\u4e4b\u5904 ViT\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u610f\u5473\u7740\u56fe\u50cf\u5de6\u4e0a\u89d2\u7684\u4e00\u4e2aPatch&#xff0c;\u53ef\u4ee5\u76f4\u63a5\u4e0e\u53f3\u4e0b\u89d2\u7684\u4e00\u4e2aPatch\u8fdb\u884c\u4fe1\u606f\u4ea4\u4e92&#xff0c;\u5b8c\u5168\u5ffd\u7565\u4e86\u5b83\u4eec\u4e4b\u95f4\u7684\u7a7a\u95f4\u8ddd\u79bb\u3002\u8fd9\u5f7b\u5e95\u6253\u7834\u4e86CNN\u57fa\u4e8e\u201c\u5c40\u90e8\u6027\u201d\u7684\u5047\u8bbe\u3002ViT\u8ba4\u4e3a&#xff0c;\u56fe\u50cf\u4e2d\u5143\u7d20\u7684\u5173\u8054&#xff0c;\u4e0d\u5e94\u53d7\u9650\u4e8e\u7a7a\u95f4\u7684\u90bb\u8fd1\u6027&#xff0c;\u4e00\u4e2a\u7269\u4f53\u7684\u4e0d\u540c\u90e8\u5206&#xff08;\u5982\u732b\u7684\u5934\u548c\u5c3e\u5df4&#xff09;\u5373\u4f7f\u76f8\u8ddd\u5f88\u8fdc&#xff0c;\u4e5f\u53ef\u80fd\u5b58\u5728\u7740\u5f3a\u70c8\u7684\u8bed\u4e49\u5173\u8054\u3002<\/p>\n<h5>9.4.2 \u5f71\u54cd\u4e0e\u672a\u6765&#xff1a;\u4e00\u573a\u5173\u4e8e\u201c\u504f\u89c1\u201d\u4e0e\u201c\u89c4\u6a21\u201d\u7684\u6df1\u523b\u5bf9\u8bdd<\/h5>\n<p>ViT\u7684\u5b9e\u9a8c\u7ed3\u679c&#xff0c;\u9707\u60ca\u4e86\u6574\u4e2aAI\u793e\u533a&#xff0c;\u5e76\u5f15\u53d1\u4e86\u4e00\u573a\u5173\u4e8eCV\u9886\u57df\u672a\u6765\u8d70\u5411\u7684\u6df1\u523b\u8ba8\u8bba\u3002<\/p>\n<p>\u6027\u80fd&#xff1a;\u89c4\u6a21\u5b9a\u5f8b\u7684\u518d\u6b21\u80dc\u5229 \u5b9e\u9a8c\u7ed3\u679c\u63ed\u793a\u4e86\u4e00\u4e2a\u5173\u952e\u7684\u73b0\u8c61&#xff1a;<\/p>\n<p>\u8fd9\u4e2a\u7ed3\u679c\u96c4\u8fa9\u5730\u8bc1\u660e\u4e86&#xff1a;Transformer\u67b6\u6784\u4e2d\u7f3a\u5931\u7684\u201c\u5f52\u7eb3\u504f\u7f6e\u201d&#xff0c;\u662f\u53ef\u4ee5\u901a\u8fc7\u201c\u89c4\u6a21&#xff08;Scale&#xff09;\u201d\u6765\u5f25\u8865\u7684\u3002 \u53ea\u8981\u7ed9\u6a21\u578b\u770b\u8db3\u591f\u591a\u7684\u6570\u636e&#xff0c;\u5b83\u5c31\u80fd\u591f\u81ea\u5df1\u4ece\u6570\u636e\u4e2d\u5b66\u4e60\u5230\u90a3\u4e9b\u539f\u672c\u9700\u8981\u6211\u4eec\u901a\u8fc7\u7ed3\u6784\u8bbe\u8ba1\u6765\u201c\u786c\u7f16\u7801\u201d\u7684\u89c6\u89c9\u89c4\u5f8b\u3002\u8fd9\u518d\u6b21\u9a8c\u8bc1\u4e86\u201c\u89c4\u6a21\u5b9a\u5f8b\u201d\u7684\u666e\u9002\u6027\u3002<\/p>\n<ul>\n<li>\u5f53\u5728\u4e2d\u5c0f\u578b\u6570\u636e\u96c6&#xff08;\u5982ImageNet-1k&#xff0c;\u7ea6120\u4e07\u5f20\u56fe&#xff09;\u4e0a\u4ece\u96f6\u5f00\u59cb\u8bad\u7ec3\u65f6&#xff0c;ViT\u7684\u6027\u80fd\u4e0d\u5982\u540c\u7b49\u89c4\u6a21\u7684CNN\u6a21\u578b&#xff08;\u5982ResNet&#xff09;\u3002\u8fd9\u5370\u8bc1\u4e86CNN\u5f52\u7eb3\u504f\u7f6e\u5728\u6570\u636e\u6548\u7387\u4e0a\u7684\u4f18\u52bf\u3002\u5728\u6570\u636e\u6709\u9650\u7684\u60c5\u51b5\u4e0b&#xff0c;ViT\u8fd9\u4e2a\u201c\u901a\u624d\u201d\u5f88\u96be\u4ece\u96f6\u5b66\u5230\u6709\u6548\u7684\u89c6\u89c9\u77e5\u8bc6\u3002<\/li>\n<li>\u7136\u800c&#xff0c;\u5f53\u5728\u8d85\u5927\u89c4\u6a21\u7684\u3001\u79c1\u6709\u7684\u6570\u636e\u96c6&#xff08;\u5982Google\u5185\u90e8\u7684JFT-300M&#xff0c;\u7ea63\u4ebf\u5f20\u56fe&#xff09;\u4e0a\u8fdb\u884c\u9884\u8bad\u7ec3&#xff0c;\u7136\u540e\u518d\u5728\u4e0b\u6e38\u4efb\u52a1\u4e0a\u8fdb\u884c\u5fae\u8c03\u65f6&#xff0c;ViT\u7684\u6027\u80fd\u8fbe\u5230\u751a\u81f3\u8d85\u8d8a\u4e86\u5f53\u65f6\u6700\u9876\u5c16\u7684CNN\u6a21\u578b\u3002<\/li>\n<\/ul>\n<p>\u67b6\u6784\u7684\u7edf\u4e00&#xff1a;\u901a\u5f80\u591a\u6a21\u6001\u667a\u80fd\u7684\u57fa\u77f3 ViT\u7684\u6210\u529f&#xff0c;\u5176\u610f\u4e49\u8fdc\u4e0d\u6b62\u4e8e\u5728\u56fe\u50cf\u5206\u7c7b\u4e0a\u53d6\u5f97SOTA&#xff08;State-of-the-Art&#xff09;\u7684\u6027\u80fd\u3002\u5b83\u6700\u91cd\u8981\u7684\u8d21\u732e&#xff0c;\u5728\u4e8e\u5c55\u793a\u4e86Transformer\u4f5c\u4e3a\u4e00\u79cd\u901a\u7528\u8ba1\u7b97\u67b6\u6784\u7684\u5de8\u5927\u6f5c\u529b\u3002 \u5728\u6b64\u4e4b\u524d&#xff0c;\u6211\u4eec\u5904\u7406\u4e0d\u540c\u6a21\u6001\u7684\u6570\u636e&#xff0c;\u9700\u8981\u8bbe\u8ba1\u4e0d\u540c\u7684\u201c\u4e13\u5bb6\u201d\u6a21\u578b&#xff1a;\u7528CNN\u5904\u7406\u56fe\u50cf&#xff0c;\u7528RNN\/Transformer\u5904\u7406\u6587\u672c&#xff0c;\u7528\u5176\u4ed6\u6a21\u578b\u5904\u7406\u8bed\u97f3\u3002ViT\u7684\u51fa\u73b0&#xff0c;\u5219\u9884\u793a\u7740\u4e00\u79cd\u67b6\u6784\u5927\u4e00\u7edf\u7684\u53ef\u80fd\u6027&#xff1a;\u6211\u4eec\u6216\u8bb8\u53ef\u4ee5\u7528\u540c\u4e00\u79cd\u6838\u5fc3\u67b6\u6784&#xff08;Transformer&#xff09;&#xff0c;\u6765\u7edf\u4e00\u5904\u7406\u6587\u672c\u3001\u56fe\u50cf\u3001\u8bed\u97f3\u3001\u89c6\u9891\u3001\u8868\u683c\u7b49\u51e0\u4e4e\u6240\u6709\u7c7b\u578b\u7684\u6570\u636e\u3002 \u8fd9\u79cd\u7edf\u4e00&#xff0c;\u4e3a\u771f\u6b63\u7684**\u591a\u6a21\u6001\u4eba\u5de5\u667a\u80fd&#xff08;Multimodal AI&#xff09;**\u5960\u5b9a\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002\u6211\u4eec\u53ef\u4ee5\u60f3\u8c61&#xff0c;\u672a\u6765\u7684\u6a21\u578b&#xff0c;\u5176\u8f93\u5165\u4e0d\u518d\u662f\u5355\u4e00\u7684\u6587\u672c\u6216\u56fe\u50cf&#xff0c;\u800c\u662f\u4e00\u4e2a\u7531\u201c\u6587\u672c\u8bcd\u5143\u201d\u3001\u201c\u56fe\u50cf\u8bcd\u5143\u201d\u3001\u201c\u8bed\u97f3\u8bcd\u5143\u201d\u6df7\u5408\u7ec4\u6210\u7684\u5e8f\u5217\u3002\u6a21\u578b\u5185\u90e8\u7684\u81ea\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u5c06\u80fd\u591f\u65e0\u5dee\u522b\u5730\u5b66\u4e60\u8fd9\u4e9b\u4e0d\u540c\u6a21\u6001\u201c\u8bcd\u5143\u201d\u4e4b\u95f4\u7684\u8de8\u6a21\u6001\u5173\u8054\u3002\u4f8b\u5982&#xff0c;\u5b83\u80fd\u7406\u89e3\u56fe\u50cf\u4e2d\u7684\u4e00\u53ea\u72d7&#xff0c;\u4e0e\u6587\u672c\u4e2d\u201c\u6bdb\u8338\u8338\u7684\u670b\u53cb\u201d\u4e4b\u95f4\u7684\u6307\u4ee3\u5173\u7cfb\u3002\u8fd9\u6b63\u662f\u901a\u5f80\u66f4\u63a5\u8fd1\u4eba\u7c7b\u667a\u80fd\u7684\u3001\u66f4\u901a\u7528\u7684\u4eba\u5de5\u667a\u80fd\u7684\u5fc5\u7531\u4e4b\u8def\u3002<\/p>\n<p>\u540e\u7eed\u53d1\u5c55&#xff1a;CNN\u4e0eTransformer\u7684\u878d\u5408\u4e0e\u8d85\u8d8a ViT\u7684\u51fa\u73b0\u5e76\u975e\u7ec8\u70b9&#xff0c;\u800c\u662f\u5f00\u542f\u4e86\u4e00\u573aCV\u67b6\u6784\u7684\u201c\u6587\u827a\u590d\u5174\u201d\u3002\u540e\u7eed\u7684\u7814\u7a76&#xff0c;\u6cbf\u7740\u4e24\u6761\u4e3b\u7ebf\u5c55\u5f00&#xff1a;<\/p>\n<li>\u6539\u8fdbViT&#xff1a;\u7814\u7a76\u8005\u4eec\u53d1\u73b0&#xff0c;\u7eaf\u7cb9\u7684ViT\u5728\u5904\u7406\u5bc6\u96c6\u9884\u6d4b\u4efb\u52a1&#xff08;\u5982\u5206\u5272\u3001\u68c0\u6d4b&#xff09;\u65f6\u5b58\u5728\u56f0\u96be&#xff0c;\u56e0\u4e3a\u5b83\u7684\u7279\u5f81\u56fe\u5206\u8fa8\u7387\u5355\u4e00\u3002\u56e0\u6b64&#xff0c;\u51fa\u73b0\u4e86\u5982Swin Transformer\u7b49\u6a21\u578b&#xff0c;\u5b83\u4eec\u901a\u8fc7\u5f15\u5165\u5c42\u7ea7\u5316\u7ed3\u6784&#xff08;Hierarchical Structure&#xff09;\u548c\u7a97\u53e3\u5316\u6ce8\u610f\u529b&#xff08;Windowed Attention&#xff09;&#xff0c;\u5c06CNN\u4e2d\u591a\u5c3a\u5ea6\u7279\u5f81\u56fe\u7684\u4f18\u70b9\u4e0eTransformer\u7684\u957f\u8ddd\u79bb\u4f9d\u8d56\u5efa\u6a21\u80fd\u529b\u5de7\u5999\u5730\u7ed3\u5408\u8d77\u6765&#xff0c;\u5728\u5404\u79cd\u89c6\u89c9\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u5168\u9762\u7684\u9886\u5148\u3002<\/li>\n<li>\u9769\u65b0CNN&#xff1a;ViT\u7684\u6210\u529f&#xff0c;\u4e5f\u53cd\u8fc7\u6765\u6fc0\u53d1\u4e86\u5bf9CNN\u67b6\u6784\u7684\u91cd\u65b0\u601d\u8003\u3002\u7814\u7a76\u8005\u4eec\u5f00\u59cb\u501f\u9274Transformer\u7684\u8bbe\u8ba1\u601d\u60f3&#xff08;\u5982\u5927\u5377\u79ef\u6838\u3001\u7c7b\u4f3cInverted Bottleneck\u7684\u7ed3\u6784&#xff09;\u6765\u73b0\u4ee3\u5316CNN&#xff0c;\u8bde\u751f\u4e86\u5982ConvNeXt\u7b49\u6a21\u578b&#xff0c;\u8bc1\u660e\u4e86\u7eaf\u7cb9\u7684\u5377\u79ef\u7f51\u7edc\u5728\u7ecf\u8fc7\u7cbe\u5fc3\u8bbe\u8ba1\u540e&#xff0c;\u4f9d\u7136\u80fd\u4e0eTransformer\u5206\u5ead\u6297\u793c\u3002<\/li>\n<p>\u8fd9\u573a\u7531ViT\u6380\u8d77\u7684\u6d6a\u6f6e&#xff0c;\u81f3\u4eca\u4ecd\u5728\u5954\u6d8c\u3002\u5b83\u4e0d\u4ec5\u4e3a\u8ba1\u7b97\u673a\u89c6\u89c9\u5e26\u6765\u4e86\u65b0\u7684SOTA\u6a21\u578b&#xff0c;\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u5b83\u6253\u7834\u4e86\u6211\u4eec\u5bf9\u201c\u5982\u4f55\u770b\u4e16\u754c\u201d\u7684\u56fa\u6709\u601d\u7ef4&#xff0c;\u63a8\u52a8\u4e86\u6574\u4e2aAI\u9886\u57df\u5411\u7740\u66f4\u901a\u7528\u3001\u66f4\u7edf\u4e00\u3001\u66f4\u5f3a\u5927\u7684\u591a\u6a21\u6001\u672a\u6765&#xff0c;\u8fc8\u51fa\u4e86\u51b3\u5b9a\u6027\u7684\u4e00\u6b65\u3002<\/p>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u8e0f\u4e0a\u4e86\u4e00\u573a\u4ece\u201c\u5faa\u73af\u201d\u5230\u201c\u6ce8\u610f\u201d&#xff0c;\u518d\u5230\u201c\u7edf\u4e00\u201d\u7684\u9769\u547d\u6027\u65c5\u7a0b\u3002\u6211\u4eec\u9996\u5148\u6df1\u5165\u5256\u6790\u4e86\u6ce8\u610f\u529b\u673a\u5236\u8fd9\u4e00\u6e90\u4e8e\u4eba\u7c7b\u8ba4\u77e5\u76f4\u89c9\u7684\u6df1\u523b\u601d\u60f3&#xff0c;\u7406\u89e3\u4e86\u5176\u5982\u4f55\u901a\u8fc7Query-Key-Value\u6846\u67b6&#xff0c;\u6253\u7834\u4e86RNN\u7684\u4fe1\u606f\u74f6\u9888\u3002<\/p>\n<p>\u63a5\u7740&#xff0c;\u6211\u4eec\u8be6\u7ec6\u89e3\u6784\u4e86\u73b0\u4ee3AI\u7684\u57fa\u77f3\u2014\u2014Transformer\u67b6\u6784\u3002\u6211\u4eec\u89c1\u8bc1\u4e86\u5b83\u5982\u4f55\u901a\u8fc7\u81ea\u6ce8\u610f\u529b\u548c\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236&#xff0c;\u5f7b\u5e95\u6446\u8131\u4e86\u5e8f\u5217\u5316\u5904\u7406\u7684\u675f\u7f1a&#xff0c;\u5b9e\u73b0\u4e86\u9ad8\u6548\u7684\u5e76\u884c\u8ba1\u7b97&#xff1b;\u5e76\u901a\u8fc7\u5de7\u5999\u7684\u4f4d\u7f6e\u7f16\u7801&#xff0c;\u627e\u56de\u4e86\u4e22\u5931\u7684\u987a\u5e8f\u611f\u3002<\/p>\n<p>\u968f\u540e&#xff0c;\u6211\u4eec\u63a2\u8ba8\u4e86\u7531Transformer\u50ac\u751f\u7684**\u201c\u9884\u8bad\u7ec3-\u5fae\u8c03\u201d\u65b0\u8303\u5f0f\u3002\u6211\u4eec\u5bf9\u6bd4\u4e86BERT\u8fd9\u4f4d\u6df1\u9083\u7684\u201c\u53cc\u5411\u7406\u89e3\u5927\u5e08\u201d\u548cGPT\u8fd9\u4f4d\u96c4\u8fa9\u7684\u201c\u5355\u5411\u751f\u6210\u5929\u624d\u201d&#xff0c;\u5e76\u76ee\u7779\u4e86\u5728\u201c\u89c4\u6a21\u5b9a\u5f8b\u201d\u7684\u9a71\u52a8\u4e0b&#xff0c;\u5927\u8bed\u8a00\u6a21\u578b\u5982\u4f55\u4ece\u9700\u8981\u201c\u5fae\u8c03\u201d\u7684\u4e13\u5bb6&#xff0c;\u6f14\u5316\u4e3a\u53ea\u9700\u201c\u63d0\u793a\u201d\u7684\u901a\u624d&#xff0c;\u5e76\u5c55\u73b0\u51fa\u60ca\u4eba\u7684\u201c\u6d8c\u73b0\u201d\u80fd\u529b**\u3002<\/p>\n<p>\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u76ee\u5149\u6295\u5411\u4e86\u89c6\u89c9\u9886\u57df&#xff0c;\u89c1\u8bc1\u4e86Vision Transformer (ViT)\u5982\u4f55\u52c7\u6562\u5730\u5c06\u56fe\u50cf\u201c\u7ffb\u8bd1\u201d\u4e3a\u5e8f\u5217&#xff0c;\u6311\u6218\u4e86CNN\u957f\u8fbe\u5341\u5e74\u7684\u7edf\u6cbb\u5730\u4f4d\u3002ViT\u7684\u6210\u529f&#xff0c;\u4e0d\u4ec5\u8bc1\u660e\u4e86\u89c4\u6a21\u53ef\u4ee5\u5f25\u8865\u5f52\u7eb3\u504f\u7f6e\u7684\u4e0d\u8db3&#xff0c;\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u5b83\u4e3a\u7528\u4e00\u79cd\u7edf\u4e00\u7684\u67b6\u6784\u5904\u7406\u591a\u6a21\u6001\u6570\u636e\u94fa\u5e73\u4e86\u9053\u8def&#xff0c;\u9884\u793a\u7740\u4e00\u4e2a\u66f4\u901a\u7528\u3001\u66f4\u5f3a\u5927\u7684\u4eba\u5de5\u667a\u80fd\u65f6\u4ee3\u7684\u5230\u6765\u3002<\/p>\n<p>\u638c\u63e1\u4e86Transformer&#xff0c;\u8bfb\u8005\u4fbf\u638c\u63e1\u4e86\u901a\u5f80\u73b0\u4ee3AI\u51e0\u4e4e\u6240\u6709\u524d\u6cbf\u9886\u57df\u7684\u94a5\u5319\u3002\u5728\u63a5\u4e0b\u6765\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u5c06\u63a2\u7d22\u66f4\u591a\u7531\u8fd9\u4e9b\u57fa\u7840\u6a21\u578b\u884d\u751f\u51fa\u7684\u3001\u4ee4\u4eba\u5174\u594b\u7684\u5e94\u7528\u9886\u57df\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u5341\u7ae0&#xff1a;\u751f\u6210\u5f0f\u6a21\u578b \u2014\u2014 \u521b\u9020\u4e0e\u60f3\u8c61<\/h3>\n<p>\u4ece\u201c\u7406\u89e3\u201d\u5230\u201c\u521b\u9020\u201d\u7684\u98de\u8dc3<\/p>\n<p>\u5728\u672c\u4e66\u524d\u9762\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u6295\u5165\u4e86\u5927\u91cf\u7684\u7cbe\u529b&#xff0c;\u53bb\u6559\u4f1a\u673a\u5668\u5982\u4f55\u201c\u7406\u89e3\u201d\u8fd9\u4e2a\u4e16\u754c\u3002\u6211\u4eec\u8bad\u7ec3\u6a21\u578b\u53bb\u8bc6\u522b\u56fe\u50cf\u4e2d\u7684\u7269\u4f53\u3001\u7406\u89e3\u6587\u672c\u4e2d\u7684\u60c5\u611f\u3001\u9884\u6d4b\u65f6\u95f4\u5e8f\u5217\u7684\u8d70\u5411\u3002\u8fd9\u4e9b\u4efb\u52a1\u7684\u6838\u5fc3&#xff0c;\u662f\u8ba9\u6a21\u578b\u5b66\u4f1a\u5206\u8fa8\u4e0e\u5224\u65ad\u3002\u8fd9\u7c7b\u6a21\u578b&#xff0c;\u6211\u4eec\u79f0\u4e4b\u4e3a\u5224\u522b\u5f0f\u6a21\u578b&#xff08;Discriminative Models&#xff09;\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u4e00\u4e2a\u4f18\u79c0\u7684\u5224\u522b\u5f0f\u6a21\u578b&#xff0c;\u6bd4\u4f5c\u4e00\u4f4d\u5b66\u8bc6\u6e0a\u535a\u7684**\u201c\u827a\u672f\u8bc4\u8bba\u5bb6\u201d\u3002\u4ed6\u80fd\u6e05\u6670\u5730\u544a\u8bc9\u60a8&#xff0c;\u4e00\u5e45\u753b\u662f\u5c5e\u4e8e\u68b5\u9ad8\u7684\u540e\u5370\u8c61\u6d3e&#xff0c;\u8fd8\u662f\u5c5e\u4e8e\u83ab\u5948\u7684\u5370\u8c61\u6d3e\u3002\u4ed6\u5b66\u4e60\u7684\u662f\u4e0d\u540c\u7c7b\u522b\u4e4b\u95f4\u7684\u51b3\u7b56\u8fb9\u754c**\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u4eba\u5de5\u667a\u80fd\u7684\u68a6\u60f3\u4e0d\u6b62\u4e8e\u6b64\u3002\u9664\u4e86\u7406\u89e3\u548c\u5206\u8fa8&#xff0c;\u6211\u4eec\u66f4\u6e34\u671b\u8d4b\u4e88\u673a\u5668\u4e00\u79cd\u66f4\u6df1\u5c42\u6b21\u7684\u80fd\u529b\u2014\u2014\u521b\u9020\u4e0e\u60f3\u8c61\u3002\u6211\u4eec\u5e0c\u671b\u6a21\u578b\u4e0d\u4ec5\u4ec5\u662f\u201c\u8bc4\u8bba\u5bb6\u201d&#xff0c;\u66f4\u80fd\u6210\u4e3a\u201c\u827a\u672f\u5b66\u5f92\u201d&#xff0c;\u751a\u81f3\u201c\u827a\u672f\u5bb6\u201d\u3002\u6211\u4eec\u5e0c\u671b\u5b83\u5728\u770b\u5c3d\u4e86\u68b5\u9ad8\u7684\u6240\u6709\u753b\u4f5c\u4e4b\u540e&#xff0c;\u80fd\u591f\u9886\u609f\u5176\u98ce\u683c\u7684\u7cbe\u9ad3&#xff0c;\u7136\u540e\u753b\u51fa\u4e00\u5e45\u5168\u65b0\u7684\u3001\u4e16\u754c\u4e0a\u4ece\u672a\u5b58\u5728\u8fc7\u7684\u3001\u4f46\u53c8\u5145\u6ee1\u4e86\u65cb\u8f6c\u661f\u7a7a\u548c\u70bd\u70ed\u5411\u65e5\u8475\u98ce\u683c\u7684\u753b\u4f5c**\u3002\u8fd9\u79cd\u5b66\u4e60\u6570\u636e\u5185\u5728\u89c4\u5f8b\u5e76\u521b\u9020\u65b0\u6570\u636e\u7684\u6a21\u578b&#xff0c;\u5c31\u662f\u751f\u6210\u5f0f\u6a21\u578b&#xff08;Generative Models&#xff09;\u3002<\/p>\n<p>\u751f\u6210\u5f0f\u6a21\u578b\u7684\u7ec8\u6781\u76ee\u6807&#xff0c;\u662f\u5b66\u4e60\u771f\u5b9e\u6570\u636e\u80cc\u540e\u90a3\u4e2a\u96be\u4ee5\u6349\u6478\u7684\u3001\u9ad8\u7ef4\u7684\u6982\u7387\u5206\u5e03&#xff0c;\u5e76\u80fd\u4ece\u4e2d\u8fdb\u884c\u91c7\u6837&#xff08;Sampling&#xff09;&#xff0c;\u4ece\u800c\u5b9e\u73b0\u201c\u65e0\u4e2d\u751f\u6709\u201d\u7684\u9b54\u529b\u3002\u8fd9\u6247\u95e8\u4e00\u65e6\u6253\u5f00&#xff0c;\u4eba\u5de5\u667a\u80fd\u4fbf\u4e0d\u518d\u4ec5\u4ec5\u662f\u5206\u6790\u5de5\u5177&#xff0c;\u800c\u662f\u6f14\u5316\u4e3a\u4e86\u53ef\u4ee5\u4e0e\u6211\u4eec\u5171\u540c\u521b\u4f5c\u7684\u3001\u5f3a\u5927\u7684\u521b\u9020\u529b\u4f19\u4f34\u3002<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5c06\u4e00\u540c\u63a2\u7d22\u4e09\u6761\u901a\u5f80\u201c\u521b\u9020\u201d\u7684\u4f1f\u5927\u8def\u5f84&#xff0c;\u5b83\u4eec\u4ee3\u8868\u4e86\u4e09\u79cd\u4e0d\u540c\u7684\u54f2\u5b66\u601d\u60f3\u4e0e\u6280\u672f\u5b9e\u73b0&#xff1a;<\/p>\n<ul>\n<li>\u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;&#xff1a;\u4e00\u573a\u7531\u201c\u4f2a\u9020\u8005\u201d\u4e0e\u201c\u9274\u8d4f\u5bb6\u201d\u4e0a\u6f14\u7684\u3001\u6c38\u65e0\u4f11\u6b62\u7684\u8fdb\u5316\u535a\u5f08\u3002<\/li>\n<li>\u53d8\u5206\u81ea\u7f16\u7801\u5668&#xff08;VAE&#xff09;&#xff1a;\u4e00\u6b21\u5145\u6ee1\u6982\u7387\u7f8e\u5b66\u3001\u8bd5\u56fe\u4e3a\u7eb7\u7e41\u4e16\u76f8\u6784\u5efa\u4e00\u4e2a\u4f18\u96c5\u201c\u6f5c\u5728\u57fa\u56e0\u5e93\u201d\u7684\u5c1d\u8bd5\u3002<\/li>\n<li>\u6269\u6563\u6a21\u578b&#xff08;Diffusion Models&#xff09;&#xff1a;\u4e00\u95e8\u5982\u540c\u65f6\u5149\u5012\u6d41\u822c&#xff0c;\u4ece\u4e00\u7247\u6df7\u6c8c\u566a\u58f0\u4e2d&#xff0c;\u9010\u6b65\u201c\u96d5\u523b\u201d\u51fa\u9ad8\u6e05\u6770\u4f5c\u7684\u9006\u5411\u827a\u672f\u3002<\/li>\n<\/ul>\n<p>\u51c6\u5907\u597d\u8fdb\u5165\u8fd9\u4e2a\u7531\u6570\u636e\u3001\u6982\u7387\u548c\u7b97\u6cd5\u6784\u6210\u7684\u3001\u5145\u6ee1\u65e0\u9650\u53ef\u80fd\u7684\u65b0\u4e16\u754c\u4e86\u5417&#xff1f;\u8ba9\u6211\u4eec\u4e00\u540c\u5f00\u542f\u8fd9\u8d9f\u65c5\u7a0b&#xff0c;\u89c1\u8bc1\u673a\u5668\u5982\u4f55\u5f00\u59cb\u201c\u505a\u68a6\u201d\u3002<\/p>\n<hr \/>\n<h4>10.1 \u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;&#xff1a;\u4e00\u573a\u6c38\u6052\u7684\u8fdb\u5316\u535a\u5f08<\/h4>\n<p>2014\u5e74&#xff0c;\u5f53\u4f0a\u6069\u00b7\u53e4\u5fb7\u8d39\u6d1b&#xff08;Ian Goodfellow&#xff09;\u548c\u4ed6\u7684\u540c\u4e8b\u4eec\u63d0\u51fa\u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;\u65f6&#xff0c;\u6216\u8bb8\u8fde\u4ed6\u4eec\u81ea\u5df1\u4e5f\u672a\u66fe\u9884\u6599\u5230&#xff0c;\u8fd9\u4e2a\u6e90\u4e8e\u4e00\u4e2a\u5de7\u5999\u601d\u60f3\u5b9e\u9a8c\u7684\u6a21\u578b&#xff0c;\u5c06\u4f1a\u5728\u672a\u6765\u6570\u5e74\u5185&#xff0c;\u6210\u4e3a\u751f\u6210\u5f0fAI\u9886\u57df\u6700\u8000\u773c\u7684\u660e\u661f&#xff0c;\u5e76\u50ac\u751f\u51fa\u65e0\u6570\u4ee4\u4eba\u60ca\u53f9\u7684\u201c\u6df1\u5ea6\u4f2a\u9020\u201d\u6770\u4f5c\u3002GAN\u7684\u9b45\u529b&#xff0c;\u4e0d\u5728\u4e8e\u5176\u590d\u6742\u7684\u6570\u5b66&#xff0c;\u800c\u5728\u4e8e\u5176\u80cc\u540e\u90a3\u7b80\u5355\u3001\u6df1\u523b\u3001\u4e14\u5145\u6ee1\u620f\u5267\u6027\u7684\u535a\u5f08\u601d\u60f3\u3002<\/p>\n<h5>10.1.1 \u6838\u5fc3\u601d\u60f3&#xff1a;\u6e90\u4e8e\u535a\u5f08\u8bba\u7684\u201c\u4e8c\u4eba\u96f6\u548c\u6e38\u620f\u201d<\/h5>\n<ul>\n<li>\n<p>\u4f2a\u9020\u8005\u4e0e\u9274\u8d4f\u5bb6\u7684\u6bd4\u55bb \u8981\u7406\u89e3GAN&#xff0c;\u6211\u4eec\u5fc5\u987b\u5148\u5728\u8111\u6d77\u4e2d\u4e0a\u6f14\u4e00\u51fa\u7cbe\u5f69\u7684\u5bf9\u624b\u620f\u3002\u8fd9\u573a\u620f\u6709\u4e24\u4e2a\u4e3b\u89d2&#xff1a;<\/p>\n<ul>\n<li>\u751f\u6210\u5668&#xff08;Generator, G&#xff09;&#xff1a;\u4e00\u4f4d\u6280\u827a\u7cbe\u6e5b\u7684**\u201c\u827a\u672f\u4f2a\u9020\u8005\u201d**\u3002\u4ed6\u4ece\u672a\u89c1\u8fc7\u771f\u6b63\u7684\u4f20\u4e16\u540d\u753b&#xff0c;\u4f46\u4ed6\u62e5\u6709\u4e00\u672c\u201c\u7075\u611f\u4e4b\u4e66\u201d&#xff08;\u4e00\u4e2a\u968f\u673a\u566a\u58f0\u6e90&#xff09;\u3002\u4ed6\u7684\u6bd5\u751f\u8ffd\u6c42&#xff0c;\u5c31\u662f\u5229\u7528\u8fd9\u4e9b\u7075\u611f&#xff0c;\u51ed\u7a7a\u521b\u9020\u51fa\u80fd\u591f\u4ee5\u5047\u4e71\u771f\u7684\u8d5d\u54c1\u753b\u4f5c\u3002<\/li>\n<li>\u5224\u522b\u5668&#xff08;Discriminator, D&#xff09;&#xff1a;\u4e00\u4f4d\u773c\u5149\u6bd2\u8fa3\u7684**\u201c\u827a\u672f\u9274\u8d4f\u5bb6\u201d**\u3002\u4ed6\u7684\u5de5\u4f5c&#xff0c;\u5c31\u662f\u9274\u5b9a\u9001\u5230\u4ed6\u9762\u524d\u7684\u753b\u4f5c\u3002\u4ed6\u65e2\u80fd\u63a5\u89e6\u5230\u535a\u7269\u9986\u91cc\u7684\u771f\u54c1&#xff0c;\u4e5f\u4f1a\u770b\u5230\u4f2a\u9020\u8005\u9001\u6765\u7684\u8d5d\u54c1\u3002\u4ed6\u7684\u76ee\u6807&#xff0c;\u662f\u5c3d\u53ef\u80fd\u7cbe\u51c6\u5730\u5206\u8fa8\u51fa\u54ea\u4e9b\u662f\u201c\u771f\u8ff9\u201d&#xff0c;\u54ea\u4e9b\u662f\u201c\u8d5d\u54c1\u201d\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5bf9\u6297\u8bad\u7ec3&#xff08;Adversarial Training&#xff09;&#xff1a;\u5728\u535a\u5f08\u4e2d\u5171\u540c\u8fdb\u5316 GAN\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u8fd9\u4e24\u4f4d\u4e3b\u89d2\u4e4b\u95f4\u4e00\u573a\u6c38\u65e0\u4f11\u6b62\u7684\u3001\u76f8\u4e92\u9a71\u52a8\u7684\u201c\u4e8c\u4eba\u96f6\u548c\u6e38\u620f\u201d\u3002<\/p>\n<li>\u7b2c\u4e00\u56de\u5408&#xff1a;\u9274\u8d4f\u5bb6\u5b66\u4e60\u3002\u5728\u6e38\u620f\u521d\u671f&#xff0c;\u4f2a\u9020\u8005\u6280\u827a\u62d9\u52a3&#xff0c;\u5176\u4f5c\u54c1&#xff08;\u5982\u4e00\u5806\u6742\u4e71\u7684\u50cf\u7d20&#xff09;\u6f0f\u6d1e\u767e\u51fa\u3002\u6211\u4eec\u56fa\u5b9a\u4f4f\u4f2a\u9020\u8005\u7684\u6c34\u5e73&#xff0c;\u7136\u540e\u8ba9\u9274\u8d4f\u5bb6\u540c\u65f6\u89c2\u770b\u771f\u54c1\u548c\u8fd9\u4e9b\u7c97\u52a3\u7684\u8d5d\u54c1\u3002\u9274\u8d4f\u5bb6\u4f1a\u5f88\u5feb\u5b66\u4f1a\u5206\u8fa8\u4e8c\u8005\u7684\u533a\u522b&#xff0c;\u4f8b\u5982&#xff0c;\u201c\u771f\u54c1\u90fd\u6709\u6e05\u6670\u7684\u8f6e\u5ed3&#xff0c;\u800c\u8d5d\u54c1\u6ca1\u6709\u201d\u3002<\/li>\n<li>\u7b2c\u4e8c\u56de\u5408&#xff1a;\u4f2a\u9020\u8005\u8fdb\u5316\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u56fa\u5b9a\u4f4f\u8fd9\u4f4d\u521a\u521a\u53d8\u5f97\u66f4\u201c\u806a\u660e\u201d\u7684\u9274\u8d4f\u5bb6\u3002\u4f2a\u9020\u8005\u5f00\u59cb\u8c03\u6574\u81ea\u5df1\u7684\u6280\u6cd5&#xff0c;\u4ed6\u7684\u76ee\u6807\u53ea\u6709\u4e00\u4e2a&#xff1a;\u201c\u6b3a\u9a97\u201d\u8fd9\u4f4d\u9274\u8d4f\u5bb6\u3002\u4ed6\u4f1a\u6839\u636e\u9274\u8d4f\u5bb6\u7ed9\u51fa\u7684\u201c\u53cd\u9988\u201d&#xff08;\u68af\u5ea6\u4fe1\u606f&#xff09;&#xff0c;\u53bb\u751f\u6210\u90a3\u4e9b\u6700\u6709\u53ef\u80fd\u88ab\u9274\u8d4f\u5bb6\u8bef\u5224\u4e3a\u201c\u771f\u201d\u7684\u4f5c\u54c1\u3002\u4f8b\u5982&#xff0c;\u4ed6\u4f1a\u5b66\u7740\u53bb\u521b\u9020\u5e26\u6709\u6e05\u6670\u8f6e\u5ed3\u7684\u56fe\u50cf\u3002<\/li>\n<li>\u87ba\u65cb\u4e0a\u5347&#xff1a;\u8fd9\u4e2a\u8fc7\u7a0b\u5468\u800c\u590d\u59cb\u3002\u9274\u8d4f\u5bb6\u5728\u89c1\u8bc6\u4e86\u66f4\u903c\u771f\u7684\u8d5d\u54c1\u540e&#xff0c;\u88ab\u8feb\u63d0\u5347\u81ea\u5df1\u7684\u9274\u5b9a\u6807\u51c6&#xff0c;\u53bb\u53d1\u73b0\u66f4\u7ec6\u5fae\u7684\u7834\u7efd&#xff1b;\u800c\u4f2a\u9020\u8005\u5728\u9762\u5bf9\u66f4\u6311\u5254\u7684\u9274\u8d4f\u5bb6\u65f6&#xff0c;\u4e5f\u88ab\u8feb\u78e8\u7ec3\u81ea\u5df1\u7684\u6280\u827a&#xff0c;\u521b\u9020\u51fa\u66f4\u5b8c\u7f8e\u7684\u8d5d\u54c1\u3002<\/li>\n<p>\u5728\u8fd9\u4e2a\u5171\u540c\u8fdb\u5316\u3001\u87ba\u65cb\u4e0a\u5347\u7684\u8fc7\u7a0b\u4e2d&#xff0c;\u53cc\u65b9\u7684\u80fd\u529b\u90fd\u5f97\u5230\u4e86\u6781\u5927\u7684\u63d0\u5347\u3002\u6700\u7ec8&#xff0c;\u7406\u60f3\u7684\u7ed3\u5c40\u662f&#xff0c;\u4f2a\u9020\u8005\u7684\u6280\u827a\u8fbe\u5230\u4e86\u51fa\u795e\u5165\u5316\u7684\u5730\u6b65&#xff0c;\u4ed6\u521b\u9020\u51fa\u7684\u8d5d\u54c1\u4e0e\u771f\u54c1\u522b\u65e0\u4e8c\u81f4&#xff0c;\u4ee5\u81f3\u4e8e\u6700\u9876\u7ea7\u7684\u9274\u8d4f\u5bb6\u4e5f\u53ea\u670950%\u7684\u6982\u7387\u80fd\u731c\u5bf9&#xff08;\u76f8\u5f53\u4e8e\u968f\u673a\u731c\u6d4b&#xff09;\u3002\u6b64\u65f6&#xff0c;\u6211\u4eec\u8bf4\u7cfb\u7edf\u8fbe\u5230\u4e86\u7eb3\u4ec0\u5747\u8861&#xff08;Nash Equilibrium&#xff09;\u3002\u800c\u8fd9\u4e2a\u6280\u827a\u9ad8\u8d85\u7684\u4f2a\u9020\u8005\u2014\u2014\u751f\u6210\u5668G&#xff0c;\u5c31\u6210\u4e86\u6211\u4eec\u68a6\u5bd0\u4ee5\u6c42\u7684\u3001\u5f3a\u5927\u7684\u751f\u6210\u6a21\u578b\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>10.1.2 \u6a21\u578b\u67b6\u6784\u4e0e\u8bad\u7ec3\u8fc7\u7a0b<\/h5>\n<p>\u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u5c06\u8fd9\u4e2a\u6bd4\u55bb\u7ffb\u8bd1\u6210\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u8a00\u3002<\/p>\n<ul>\n<li>\n<p>\u751f\u6210\u5668&#xff08;Generator&#xff09;<\/p>\n<ul>\n<li>\u8f93\u5165&#xff1a;\u4e00\u4e2a\u4ece\u7b80\u5355\u6982\u7387\u5206\u5e03&#xff08;\u5982100\u7ef4\u7684\u6807\u51c6\u9ad8\u65af\u5206\u5e03&#xff09;\u4e2d\u968f\u673a\u91c7\u6837\u7684\u6f5c\u5728\u5411\u91cf&#xff08;Latent Vector&#xff09;z\u3002\u8fd9\u4e2a\u5411\u91cfz&#xff0c;\u5c31\u662f\u4f2a\u9020\u8005\u7684\u201c\u7075\u611f\u79cd\u5b50\u201d\u3002\u6bcf\u4e00\u4e2a\u4e0d\u540c\u7684z&#xff0c;\u90fd\u5c06\u5bf9\u5e94\u4e00\u4e2a\u4e0d\u540c\u7684\u751f\u6210\u7ed3\u679c\u3002<\/li>\n<li>\u7ed3\u6784&#xff1a;\u751f\u6210\u5668\u7684\u7f51\u7edc\u7ed3\u6784&#xff0c;\u901a\u5e38\u662f\u4e00\u4e2a**\u201c\u653e\u5927\u201d\u7684\u8fc7\u7a0b\u3002\u5b83\u4e0e\u6211\u4eec\u719f\u6089\u7684CNN\u5206\u7c7b\u5668\u6b63\u597d\u76f8\u53cd\u3002\u5b83\u91c7\u7528\u8f6c\u7f6e\u5377\u79ef&#xff08;Transposed Convolution&#xff0c;\u5e38\u88ab\u4e0d\u51c6\u786e\u5730\u79f0\u4e3a\u53cd\u5377\u79ef&#xff09;\u6216\u4e0a\u91c7\u6837&#xff08;Upsampling&#xff09;**\u64cd\u4f5c&#xff0c;\u5c06\u8f93\u5165\u7684\u4f4e\u7ef4z\u5411\u91cf&#xff0c;\u4e00\u5c42\u5c42\u5730\u653e\u5927\u5c3a\u5bf8\u3001\u4e30\u5bcc\u7ec6\u8282&#xff0c;\u6700\u7ec8\u751f\u6210\u4e00\u4e2a\u4e0e\u771f\u5b9e\u6570\u636e\u7ef4\u5ea6\u76f8\u540c\u7684\u4eba\u9020\u6570\u636e&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u5f2064x64x3\u7684\u5f69\u8272\u56fe\u50cf&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5224\u522b\u5668&#xff08;Discriminator&#xff09;<\/p>\n<ul>\n<li>\u8f93\u5165&#xff1a;\u4e00\u7b14\u6570\u636e&#xff0c;\u5b83\u670950%\u7684\u6982\u7387\u662f\u6765\u81ea\u771f\u5b9e\u6570\u636e\u96c6\u7684\u771f\u6837\u672c\u00a0x&#xff0c;\u670950%\u7684\u6982\u7387\u662f\u6765\u81ea\u751f\u6210\u5668\u7684\u5047\u6837\u672c\u00a0G(z)\u3002<\/li>\n<li>\u7ed3\u6784&#xff1a;\u5224\u522b\u5668\u901a\u5e38\u5c31\u662f\u4e00\u4e2a\u6807\u51c6\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u5206\u7c7b\u5668\u3002\u5b83\u63a5\u6536\u4e00\u5f20\u56fe\u50cf&#xff0c;\u901a\u8fc7\u4e00\u7cfb\u5217\u5377\u79ef\u3001\u6c60\u5316\u5c42\u63d0\u53d6\u7279\u5f81\u3002<\/li>\n<li>\u8f93\u51fa&#xff1a;\u4e0e\u666e\u901a\u591a\u5206\u7c7b\u5668\u4e0d\u540c&#xff0c;\u5224\u522b\u5668\u7684\u8f93\u51fa\u5c42\u53ea\u6709\u4e00\u4e2a\u795e\u7ecf\u5143&#xff0c;\u5e76\u4f7f\u7528Sigmoid\u6fc0\u6d3b\u51fd\u6570\u3002\u5b83\u8f93\u51fa\u4e00\u4e2a\u4ecb\u4e8e0\u548c1\u4e4b\u95f4\u7684\u6807\u91cf&#xff08;\u6982\u7387\u503c&#xff09;&#xff0c;\u4ee3\u8868\u6a21\u578b\u5224\u65ad\u8f93\u5165\u6570\u636e\u4e3a**\u201c\u771f\u201d**\u7684\u6982\u7387\u3002\u8f93\u51fa1\u4ee3\u8868\u201c\u7edd\u5bf9\u662f\u771f\u201d&#xff0c;\u8f93\u51fa0\u4ee3\u8868\u201c\u7edd\u5bf9\u662f\u5047\u201d\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4ef7\u503c\u51fd\u6570&#xff08;Value Function&#xff09;\u4e0e\u4f18\u5316\u76ee\u6807 GAN\u7684\u5bf9\u6297\u8fc7\u7a0b&#xff0c;\u53ef\u4ee5\u901a\u8fc7\u4e00\u4e2a\u7edf\u4e00\u7684**\u6700\u5c0f\u6700\u5927\u5316&#xff08;Minimax&#xff09;**\u76ee\u6807\u51fd\u6570\u6765\u63cf\u8ff0&#xff1a; min_G max_D V(D, G) &#061; E_{x~p_data(x)}[log D(x)] &#043; E_{z~p_z(z)}[log(1 &#8211; D(G(z)))]<\/p>\n<p>\u8ba9\u6211\u4eec\u6765\u62c6\u89e3\u8fd9\u4e2a\u770b\u4f3c\u590d\u6742\u7684\u516c\u5f0f&#xff1a;<\/p>\n<ul>\n<li>\u8bad\u7ec3\u5224\u522b\u5668D&#xff08;max_D V(D, G)&#xff09;&#xff1a;\u5f53\u8bad\u7ec3\u5224\u522b\u5668\u65f6&#xff0c;\u6211\u4eec\u5e0c\u671b\u6700\u5927\u5316\u8fd9\u4e2a\u4ef7\u503c\u51fd\u6570\u3002\n<ul>\n<li>\u5bf9\u4e8e\u771f\u5b9e\u6837\u672cx&#xff0c;D(x)\u5e94\u8be5\u8d8b\u8fd1\u4e8e1&#xff0c;log D(x)\u8d8b\u8fd1\u4e8e0\u3002<\/li>\n<li>\u5bf9\u4e8e\u751f\u6210\u6837\u672cG(z)&#xff0c;D(G(z))\u5e94\u8be5\u8d8b\u8fd1\u4e8e0&#xff0c;log(1 &#8211; D(G(z)))\u8d8b\u8fd1\u4e8e0\u3002<\/li>\n<li>\u56e0\u6b64&#xff0c;\u6700\u5927\u5316V(D, G)&#xff0c;\u5c31\u662f\u5728\u8bad\u7ec3\u5224\u522b\u5668\u5c06\u771f\u5b9e\u6837\u672c\u5224\u65ad\u4e3a1&#xff0c;\u5c06\u751f\u6210\u6837\u672c\u5224\u65ad\u4e3a0\u3002\u8fd9\u6b63\u662f\u201c\u9274\u8d4f\u5bb6\u201d\u7684\u804c\u8d23\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u8bad\u7ec3\u751f\u6210\u5668G&#xff08;min_G V(D, G)&#xff09;&#xff1a;\u5f53\u8bad\u7ec3\u751f\u6210\u5668\u65f6&#xff0c;\u6211\u4eec\u5e0c\u671b\u6700\u5c0f\u5316\u8fd9\u4e2a\u4ef7\u503c\u51fd\u6570\u3002\u751f\u6210\u5668\u65e0\u6cd5\u5f71\u54cd\u7b2c\u4e00\u9879log D(x)&#xff0c;\u5b83\u53ea\u80fd\u5f71\u54cd\u7b2c\u4e8c\u9879\u3002\u5b83\u5e0c\u671b\u81ea\u5df1\u751f\u6210\u7684\u6837\u672cG(z)&#xff0c;\u88ab\u5224\u522b\u5668D\u5224\u65ad\u4e3a\u201c\u771f\u201d&#xff0c;\u5373D(G(z))\u8d8b\u8fd1\u4e8e1\u3002\u5f53D(G(z))\u8d8b\u8fd1\u4e8e1\u65f6&#xff0c;log(1 &#8211; D(G(z)))\u4f1a\u8d8b\u8fd1\u4e8e\u8d1f\u65e0\u7a77&#xff0c;\u4ece\u800c\u6700\u5c0f\u5316\u4e86V(D, G)\u3002\n<ul>\n<li>\u5b9e\u8df5\u4e2d\u7684\u6280\u5de7&#xff1a;\u5728\u5b9e\u8df5\u4e2d&#xff0c;\u6700\u5c0f\u5316log(1 &#8211; D(G(z)))\u5728\u65e9\u671f\u68af\u5ea6\u8f83\u5c0f&#xff0c;\u8bad\u7ec3\u56f0\u96be\u3002\u56e0\u6b64&#xff0c;\u901a\u5e38\u4f1a\u5c06\u5176\u66ff\u6362\u4e3a\u4e00\u4e2a\u7b49\u4ef7\u7684\u3001\u4f46\u68af\u5ea6\u8868\u73b0\u66f4\u597d\u7684\u76ee\u6807&#xff1a;\u6700\u5927\u5316\u00a0log D(G(z))\u3002\u8fd9\u5728\u76f4\u89c9\u4e0a\u66f4\u5bb9\u6613\u7406\u89e3&#xff1a;\u751f\u6210\u5668\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u6700\u5927\u5316\u5176\u751f\u6210\u6837\u672c\u88ab\u5224\u522b\u5668\u5224\u65ad\u4e3a\u201c\u771f\u201d\u7684\u5bf9\u6570\u6982\u7387\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>10.1.3 \u6311\u6218\u4e0e\u6f14\u8fdb<\/h5>\n<p>GAN\u7684\u8bad\u7ec3\u8fc7\u7a0b\u5982\u540c\u4e00\u573a\u7cbe\u5bc6\u7684\u821e\u8e48&#xff0c;\u7a0d\u6709\u4e0d\u614e\u5c31\u4f1a\u5931\u8861\u3002\u5176\u8bad\u7ec3\u7684\u4e0d\u7a33\u5b9a\u6027\u662f\u51fa\u4e86\u540d\u7684&#xff0c;\u4e3b\u8981\u4f53\u73b0\u5728\u4ee5\u4e0b\u51e0\u4e2a\u65b9\u9762&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u6a21\u5f0f\u5d29\u6e83&#xff08;Mode Collapse&#xff09; \u8fd9\u662fGAN\u6700\u5e38\u89c1\u7684\u5931\u8d25\u6a21\u5f0f\u4e4b\u4e00\u3002\u60f3\u8c61\u4e00\u4e0b&#xff0c;\u5982\u679c\u4f2a\u9020\u8005\u5076\u7136\u53d1\u73b0&#xff0c;\u4ed6\u753b\u7684\u67d0\u4e00\u79cd\u201c\u732b\u201d&#xff08;\u6bd4\u5982\u4e00\u53ea\u7279\u5b9a\u7684\u5e03\u5076\u732b&#xff09;\u80fd\u591f\u7a33\u5b9a\u5730\u9a97\u8fc7\u5f53\u524d\u7684\u9274\u8d4f\u5bb6\u3002\u51fa\u4e8e\u201c\u5077\u61d2\u201d&#xff0c;\u4ed6\u53ef\u80fd\u4f1a\u505c\u6b62\u63a2\u7d22\u5176\u4ed6\u753b\u6cd5&#xff0c;\u5f00\u59cb\u53cd\u590d\u5730\u3001\u53ea\u751f\u6210\u8fd9\u4e00\u79cd\u5e03\u5076\u732b\u3002\u7ed3\u679c\u5c31\u662f&#xff0c;\u751f\u6210\u5668\u867d\u7136\u80fd\u4ea7\u751f\u9ad8\u8d28\u91cf\u7684\u56fe\u50cf&#xff0c;\u4f46\u751f\u6210\u7ed3\u679c\u6781\u5176\u7f3a\u4e4f\u591a\u6837\u6027\u3002\u5b83\u53ea\u5b66\u4f1a\u4e86\u771f\u5b9e\u6570\u636e\u5206\u5e03\u4e2d\u7684\u4e00\u4e2a\u6216\u5c11\u6570\u51e0\u4e2a\u201c\u6a21\u5f0f&#xff08;Mode&#xff09;\u201d\u3002<\/p>\n<\/li>\n<li>\n<p>\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8 \u5bf9\u6297\u8bad\u7ec3\u7684\u5e73\u8861\u975e\u5e38\u8106\u5f31\u3002<\/p>\n<ul>\n<li>\u5982\u679c\u5224\u522b\u5668\u8fc7\u4e8e\u5f3a\u5927&#xff0c;\u5b83\u80fd\u8f7b\u6613\u5730\u5206\u8fa8\u771f\u5047&#xff0c;\u7ed9\u51fa\u7684\u6982\u7387\u603b\u662f\u63a5\u8fd11\u62160\u3002\u8fd9\u4f1a\u5bfc\u81f4log(1 &#8211; D(G(z)))\u7684\u68af\u5ea6\u53d8\u5f97\u975e\u5e38\u5c0f&#xff0c;\u751f\u6210\u5668\u63a5\u6536\u4e0d\u5230\u6709\u6548\u7684\u5b66\u4e60\u4fe1\u53f7&#xff0c;\u5982\u540c\u4e00\u4e2a\u5b66\u5f92\u9762\u5bf9\u4e00\u4f4d\u4ece\u4e0d\u6307\u70b9\u3001\u53ea\u4f1a\u8bf4\u201c\u4e0d\u5bf9\u201d\u7684\u4e25\u5e08&#xff0c;\u5b8c\u5168\u65e0\u6cd5\u8fdb\u6b65\u3002<\/li>\n<li>\u53cd\u4e4b&#xff0c;\u5982\u679c\u5224\u522b\u5668\u8fc7\u4e8e\u5f31\u5c0f&#xff0c;\u65e0\u6cd5\u63d0\u4f9b\u6709\u533a\u5206\u5ea6\u7684\u53cd\u9988&#xff0c;\u751f\u6210\u5668\u4e5f\u4f1a\u8ff7\u5931\u65b9\u5411\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898&#xff0c;\u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u5927\u91cf\u7684GAN\u53d8\u4f53&#xff0c;\u5176\u4e2d\u4e00\u4e9b\u662f\u5177\u6709\u91cc\u7a0b\u7891\u610f\u4e49\u7684&#xff1a;<\/p>\n<ul>\n<li>\n<p>DCGAN (Deep Convolutional GAN) DCGAN\u662f\u7b2c\u4e00\u4e2a\u5c06\u6df1\u5ea6\u5377\u79ef\u7f51\u7edc\u6210\u529f\u4e14\u7a33\u5b9a\u5730\u5e94\u7528\u4e8eGAN\u7684\u6a21\u578b\u3002\u5b83\u5e76\u975e\u63d0\u51fa\u4e86\u5168\u65b0\u7684\u7406\u8bba&#xff0c;\u800c\u662f\u4e3aGAN\u7684\u8bad\u7ec3\u63d0\u4f9b\u4e86\u4e00\u5957\u6781\u5176\u5b9d\u8d35\u7684\u7f51\u7edc\u7ed3\u6784\u8bbe\u8ba1\u548c\u8d85\u53c2\u6570\u8bbe\u7f6e\u7684\u201c\u6700\u4f73\u5b9e\u8df5\u201d&#xff0c;\u4f8b\u5982&#xff1a;\u7528\u8f6c\u7f6e\u5377\u79ef\u66ff\u4ee3\u4e0a\u91c7\u6837\u3001\u5728\u751f\u6210\u5668\u548c\u5224\u522b\u5668\u4e2d\u90fd\u4f7f\u7528\u6279\u5f52\u4e00\u5316&#xff08;Batch Normalization&#xff09;\u3001\u79fb\u9664\u5168\u8fde\u63a5\u5c42\u3001\u5728\u751f\u6210\u5668\u4e2d\u4f7f\u7528ReLU\u6fc0\u6d3b\u51fd\u6570\u800c\u5728\u5224\u522b\u5668\u4e2d\u4f7f\u7528LeakyReLU\u7b49\u3002\u8fd9\u4e9b\u51c6\u5219&#xff0c;\u6781\u5927\u5730\u63d0\u5347\u4e86GAN\u7684\u8bad\u7ec3\u7a33\u5b9a\u6027\u548c\u751f\u6210\u8d28\u91cf&#xff0c;\u662fGAN\u4ece\u7406\u8bba\u8d70\u5411\u5b9e\u7528\u7684\u5173\u952e\u4e00\u6b65\u3002<\/p>\n<\/li>\n<li>\n<p>WGAN (Wasserstein GAN) WGAN\u4ece\u7406\u8bba\u5c42\u9762&#xff0c;\u4e3a\u89e3\u51b3GAN\u7684\u8bad\u7ec3\u4e0d\u7a33\u5b9a\u6027\u63d0\u4f9b\u4e86\u6df1\u523b\u7684\u6d1e\u89c1\u3002\u5b83\u6307\u51fa&#xff0c;\u539f\u59cbGAN\u6240\u4f18\u5316\u7684\u76ee\u6807&#xff0c;\u7b49\u4ef7\u4e8e\u6700\u5c0f\u5316\u771f\u5b9e\u5206\u5e03\u4e0e\u751f\u6210\u5206\u5e03\u4e4b\u95f4\u7684JS\u6563\u5ea6\u3002\u800c\u5f53\u4e24\u4e2a\u5206\u5e03\u6ca1\u6709\u91cd\u53e0\u65f6&#xff0c;JS\u6563\u5ea6\u662f\u4e00\u4e2a\u5e38\u6570&#xff0c;\u8fd9\u4f1a\u5bfc\u81f4\u68af\u5ea6\u6d88\u5931\u3002 WGAN\u63d0\u51fa&#xff0c;\u8f6c\u800c\u4f7f\u7528Wasserstein\u8ddd\u79bb&#xff08;\u53c8\u79f0\u201c\u63a8\u571f\u673a\u8ddd\u79bb\u201d&#xff09;\u6765\u5ea6\u91cf\u4e24\u4e2a\u5206\u5e03\u7684\u5dee\u5f02\u3002Wasserstein\u8ddd\u79bb\u5373\u4f7f\u5728\u5206\u5e03\u4e0d\u91cd\u53e0\u65f6&#xff0c;\u4e5f\u80fd\u63d0\u4f9b\u6709\u610f\u4e49\u7684\u3001\u5e73\u6ed1\u7684\u68af\u5ea6\u3002\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u4e00\u70b9&#xff0c;WGAN\u5bf9\u5224\u522b\u5668&#xff08;\u5728WGAN\u4e2d\u88ab\u79f0\u4e3a\u201c\u8bc4\u8bba\u5bb6&#xff08;Critic&#xff09;\u201d&#xff09;\u7684\u7ed3\u6784\u8fdb\u884c\u4e86\u4fee\u6539&#xff08;\u5982\u79fb\u9664\u6700\u540e\u7684Sigmoid\u5c42&#xff09;&#xff0c;\u5e76\u5f15\u5165\u4e86\u6743\u91cd\u88c1\u526a&#xff08;Weight Clipping&#xff09;\u6216\u68af\u5ea6\u60e9\u7f5a&#xff08;Gradient Penalty&#xff09;\u7b49\u6280\u5de7&#xff0c;\u6765\u6ee1\u8db3\u7406\u8bba\u6240\u9700\u7684\u201c\u5229\u666e\u5e0c\u8328\u8fde\u7eed\u6027\u201d\u6761\u4ef6\u3002WGAN\u7684\u51fa\u73b0&#xff0c;\u6781\u5927\u5730\u7f13\u89e3\u4e86\u6a21\u5f0f\u5d29\u6e83\u95ee\u9898&#xff0c;\u5e76\u4f7f\u5f97\u8bad\u7ec3\u8fc7\u7a0b\u4e0e\u751f\u6210\u8d28\u91cf\u4e4b\u95f4\u7684\u5173\u8054\u6027\u66f4\u5f3a\u3002<\/p>\n<\/li>\n<li>\n<p>Conditional GAN (cGAN) \u539f\u59cb\u7684GAN\u662f\u4e00\u79cd\u65e0\u6761\u4ef6\u7684&#xff08;Unconditional&#xff09;\u751f\u6210\u6a21\u578b&#xff0c;\u6211\u4eec\u65e0\u6cd5\u63a7\u5236\u5b83\u5177\u4f53\u751f\u6210\u4ec0\u4e48\u5185\u5bb9\u3002\u6761\u4ef6\u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;cGAN&#xff09;\u5219\u901a\u8fc7\u5f15\u5165\u6761\u4ef6\u4fe1\u606fy&#xff08;\u4f8b\u5982&#xff0c;\u56fe\u50cf\u7684\u7c7b\u522b\u6807\u7b7e\u3001\u4e00\u6bb5\u6587\u672c\u63cf\u8ff0&#xff09;&#xff0c;\u5b9e\u73b0\u4e86\u53ef\u63a7\u751f\u6210\u3002 \u5176\u5b9e\u73b0\u65b9\u5f0f\u975e\u5e38\u76f4\u89c2&#xff1a;\u5c06\u6761\u4ef6\u4fe1\u606fy&#xff0c;\u540c\u65f6\u8f93\u5165\u7ed9\u751f\u6210\u5668\u548c\u5224\u522b\u5668\u3002<\/p>\n<ul>\n<li>\u751f\u6210\u5668\u63a5\u6536\u566a\u58f0z\u548c\u6761\u4ef6y&#xff0c;\u5b66\u4e60\u751f\u6210\u7b26\u5408y\u63cf\u8ff0\u7684\u6837\u672cG(z|y)\u3002<\/li>\n<li>\u5224\u522b\u5668\u63a5\u6536\u6837\u672c&#xff08;\u771f\u6216\u5047&#xff09;\u548c\u6761\u4ef6y&#xff0c;\u5b66\u4e60\u5224\u65ad\u8be5\u6837\u672c\u662f\u5426\u65e2\u201c\u771f\u5b9e\u201d\u53c8\u201c\u4e0e\u6761\u4ef6y\u76f8\u7b26\u201d\u3002 cGAN\u7684\u51fa\u73b0&#xff0c;\u4f7f\u5f97GAN\u7684\u5e94\u7528\u573a\u666f\u88ab\u6781\u5927\u5730\u62d3\u5bbd\u4e86\u3002\u6211\u4eec\u53ef\u4ee5\u6307\u5b9a\u6a21\u578b\u201c\u8bf7\u751f\u6210\u4e00\u53ea\u864e\u6591\u732b\u7684\u56fe\u50cf\u201d&#xff0c;\u6216\u8005\u5728\u540e\u7eed\u66f4\u590d\u6742\u7684\u6a21\u578b\u4e2d&#xff0c;\u6839\u636e\u4e00\u6bb5\u8be6\u7ec6\u7684\u6587\u5b57\u63cf\u8ff0\u6765\u751f\u6210\u56fe\u50cf\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h4>10.2 \u53d8\u5206\u81ea\u7f16\u7801\u5668&#xff08;VAE&#xff09;&#xff1a;\u6982\u7387\u751f\u6210\u7684\u7f8e\u5b66<\/h4>\n<p>\u4e0eGAN\u90a3\u5145\u6ee1\u620f\u5267\u6027\u7684\u8bde\u751f\u6545\u4e8b\u4e0d\u540c&#xff0c;VAE\u690d\u6839\u4e8e\u66f4\u6df1\u539a\u7684\u6982\u7387\u56fe\u6a21\u578b\u548c\u53d8\u5206\u63a8\u65ad\u7684\u571f\u58e4\u3002\u5b83\u4e0d\u8ffd\u6c42\u751f\u6210\u6700\u9510\u5229\u3001\u6700\u4ee5\u5047\u4e71\u771f\u7684\u56fe\u50cf&#xff0c;\u800c\u662f\u81f4\u529b\u4e8e\u6784\u5efa\u4e00\u4e2a\u5e73\u6ed1\u3001\u8fde\u7eed\u3001\u4e14\u5bcc\u6709\u610f\u4e49\u7684\u6f5c\u5728\u7a7a\u95f4&#xff08;Latent Space&#xff09;\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u8fd9\u4e2a\u6f5c\u5728\u7a7a\u95f4&#xff0c;\u60f3\u8c61\u6210\u4e00\u4e2a\u7269\u79cd\u7684\u201c\u57fa\u56e0\u5e93\u201d\u3002\u5e93\u4e2d\u7684\u6bcf\u4e00\u4e2a\u70b9&#xff0c;\u90fd\u5bf9\u5e94\u4e00\u4e2a\u201c\u4e2a\u4f53\u201d&#xff1b;\u800c\u70b9\u4e0e\u70b9\u4e4b\u95f4\u7684\u5e73\u6ed1\u8fc7\u6e21&#xff0c;\u5219\u5bf9\u5e94\u7740\u7269\u79cd\u7684\u5e73\u6ed1\u6f14\u5316\u3002VAE\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u4e3a\u6211\u4eec\u6240\u89c2\u5bdf\u5230\u7684\u7eb7\u7e41\u590d\u6742\u7684\u6570\u636e&#xff08;\u5982\u4eba\u8138\u56fe\u50cf&#xff09;&#xff0c;\u6784\u5efa\u51fa\u8fd9\u6837\u4e00\u4e2a\u4f18\u96c5\u7684\u201c\u6f5c\u5728\u57fa\u56e0\u5e93\u201d\u3002<\/p>\n<h5>10.2.1 \u6838\u5fc3\u601d\u60f3&#xff1a;\u4ece\u201c\u5b8c\u7f8e\u590d\u523b\u201d\u5230\u201c\u795e\u4f3c\u5373\u53ef\u201d<\/h5>\n<p>\u8981\u7406\u89e3VAE\u7684\u7cbe\u9ad3&#xff0c;\u6211\u4eec\u5fc5\u987b\u5148\u56de\u987e\u5b83\u7684\u524d\u8eab\u2014\u2014\u6807\u51c6\u81ea\u7f16\u7801\u5668&#xff08;Autoencoder, AE&#xff09;\u3002<\/p>\n<ul>\n<li>\n<p>\u56de\u987e\u81ea\u7f16\u7801\u5668&#xff08;Autoencoder, AE&#xff09; \u4e00\u4e2a\u6807\u51c6\u7684\u81ea\u7f16\u7801\u5668&#xff0c;\u7531\u4e24\u90e8\u5206\u7ec4\u6210&#xff1a;<\/p>\n<ul>\n<li>\u7f16\u7801\u5668&#xff08;Encoder&#xff09;&#xff1a;\u5b83\u50cf\u4e00\u4e2a\u201c\u538b\u7f29\u7a0b\u5e8f\u201d&#xff0c;\u63a5\u6536\u4e00\u4e2a\u9ad8\u7ef4\u7684\u8f93\u5165\u6570\u636ex&#xff08;\u5982\u4e00\u5f20\u56fe\u7247&#xff09;&#xff0c;\u5e76\u5c06\u5176\u538b\u7f29\u6210\u4e00\u4e2a\u4f4e\u7ef4\u7684\u6f5c\u5728\u7f16\u7801&#xff08;Latent Code&#xff09;z\u3002<\/li>\n<li>\u89e3\u7801\u5668&#xff08;Decoder&#xff09;&#xff1a;\u5b83\u50cf\u4e00\u4e2a\u201c\u89e3\u538b\u7a0b\u5e8f\u201d&#xff0c;\u63a5\u6536\u6f5c\u5728\u7f16\u7801z&#xff0c;\u5e76\u5c1d\u8bd5\u5c06\u5176\u91cd\u6784\u51fa\u4e0e\u539f\u59cb\u8f93\u5165x\u4e00\u6a21\u4e00\u6837\u7684x&#039;\u3002 AE\u7684\u8bad\u7ec3\u76ee\u6807&#xff0c;\u5c31\u662f\u6700\u5c0f\u5316\u91cd\u6784\u8bef\u5dee&#xff08;\u5373x\u4e0ex&#039;\u4e4b\u95f4\u7684\u5dee\u8ddd&#xff09;\u3002\u5b83\u662f\u4e00\u4e2a\u5f3a\u5927\u7684\u975e\u7ebf\u6027\u964d\u7ef4\u5de5\u5177\u3002\u7136\u800c&#xff0c;AE\u7684\u6f5c\u5728\u7a7a\u95f4z&#xff0c;\u662f\u4e0d\u8fde\u7eed\u3001\u4e0d\u89c4\u5219\u7684\u3002\u5982\u679c\u5728\u5b83\u7684\u6f5c\u5728\u7a7a\u95f4\u4e2d\u968f\u673a\u53d6\u4e00\u4e2a\u70b9&#xff0c;\u7136\u540e\u7528\u89e3\u7801\u5668\u8fdb\u884c\u89e3\u7801&#xff0c;\u901a\u5e38\u53ea\u4f1a\u5f97\u5230\u4e00\u5806\u6beb\u65e0\u610f\u4e49\u7684\u4e71\u7801\u3002\u8fd9\u662f\u56e0\u4e3aAE\u53ea\u5b66\u4f1a\u4e86\u5982\u4f55\u5bf9\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u7279\u5b9a\u70b9\u8fdb\u884c\u7f16\u7801\u548c\u89e3\u7801&#xff0c;\u800c\u5bf9\u8fd9\u4e9b\u70b9\u4e4b\u95f4\u7684\u201c\u7a7a\u767d\u533a\u57df\u201d\u4e00\u65e0\u6240\u77e5\u3002\u56e0\u6b64&#xff0c;\u6807\u51c6AE\u65e0\u6cd5\u7528\u4e8e\u751f\u6210\u65b0\u6837\u672c\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>VAE\u7684\u6982\u7387\u5316\u6539\u9020&#xff1a;\u4e3a\u6f5c\u5728\u7a7a\u95f4\u6ce8\u5165\u7075\u9b42 VAE\u7684\u9769\u547d\u6027\u521b\u4e3e&#xff0c;\u5c31\u662f\u5bf9\u7f16\u7801\u8fc7\u7a0b\u8fdb\u884c\u4e86\u6982\u7387\u5316\u7684\u6539\u9020\u3002\u5b83\u4e0d\u518d\u8ba4\u4e3a\u4e00\u4e2a\u8f93\u5165x\u5e94\u8be5\u5bf9\u5e94\u4e00\u4e2a\u786e\u5b9a\u7684\u3001\u552f\u4e00\u7684\u6f5c\u5728\u7f16\u7801\u70b9z\u3002\u76f8\u53cd&#xff0c;\u5b83\u8ba4\u4e3a\u4e00\u4e2a\u8f93\u5165x\u5e94\u8be5\u5bf9\u5e94\u6f5c\u5728\u7a7a\u95f4\u4e2d\u7684\u4e00\u4e2a\u6982\u7387\u5206\u5e03\u3002 \u5177\u4f53\u6765\u8bf4&#xff0c;VAE\u7684\u7f16\u7801\u5668\u4e0d\u518d\u76f4\u63a5\u8f93\u51fa\u4e00\u4e2a\u5411\u91cfz&#xff0c;\u800c\u662f\u8f93\u51fa\u4e24\u4e2a\u5411\u91cf&#xff1a;\u8fd9\u4e2a\u6982\u7387\u5206\u5e03\u7684\u5747\u503c\u5411\u91cf\u03bc\u548c\u5bf9\u6570\u65b9\u5dee\u5411\u91cflog(\u03c3\u00b2)\u3002\u8fd9\u4e24\u4e2a\u5411\u91cf&#xff0c;\u5171\u540c\u5b9a\u4e49\u4e86\u4e00\u4e2a\u9ad8\u65af\u5206\u5e03N(\u03bc, \u03c3\u00b2)\u3002\u7136\u540e&#xff0c;\u6211\u4eec\u4ece\u8fd9\u4e2a\u4e13\u5c5e\u4e8ex\u7684\u5206\u5e03\u4e2d&#xff0c;\u968f\u673a\u91c7\u6837\u4e00\u4e2a\u70b9z&#xff0c;\u518d\u9001\u5165\u89e3\u7801\u5668\u8fdb\u884c\u91cd\u6784\u3002<\/p>\n<\/li>\n<li>\n<p>\u6f5c\u5728\u7a7a\u95f4\u7684\u6b63\u5219\u5316&#xff1a;\u6784\u5efa\u201c\u521b\u4e16\u5730\u56fe\u201d \u4ec5\u4ec5\u8fdb\u884c\u6982\u7387\u5316\u6539\u9020\u8fd8\u4e0d\u591f\u3002VAE\u6700\u5173\u952e\u7684\u4e00\u6b65&#xff0c;\u662f\u901a\u8fc7\u4e00\u4e2a\u5de7\u5999\u8bbe\u8ba1\u7684\u635f\u5931\u51fd\u6570&#xff0c;\u5bf9\u7f16\u7801\u5668\u65bd\u52a0\u4e86\u4e00\u4e2a\u5f3a\u5927\u7684\u7ea6\u675f&#xff1a;\u5b83\u8981\u6c42\u6240\u6709\u8f93\u5165x\u6240\u7f16\u7801\u51fa\u7684\u8fd9\u4e9b\u6982\u7387\u5206\u5e03N(\u03bc, \u03c3\u00b2)&#xff0c;\u90fd\u5fc5\u987b\u5c3d\u53ef\u80fd\u5730\u9760\u8fd1\u6807\u51c6\u6b63\u6001\u5206\u5e03N(0, 1)&#xff08;\u4e00\u4e2a\u4ee5\u539f\u70b9\u4e3a\u4e2d\u5fc3\u3001\u5404\u4e2a\u7ef4\u5ea6\u76f8\u4e92\u72ec\u7acb\u7684\u9ad8\u65af\u5206\u5e03&#xff09;\u3002 \u8fd9\u4e2a\u7ea6\u675f&#xff0c;\u5c31\u50cf\u4e00\u80a1\u5f3a\u5927\u7684\u5f15\u529b&#xff0c;\u5c06\u6f5c\u5728\u7a7a\u95f4\u4e2d\u539f\u672c\u79bb\u6563\u3001\u6df7\u4e71\u7684\u5404\u4e2a\u201c\u5c9b\u5c7f\u201d&#xff08;\u4e0d\u540c\u6570\u636e\u70b9\u7684\u7f16\u7801\u5206\u5e03&#xff09;&#xff0c;\u62c9\u626f\u3001\u89c4\u6574\u6210\u4e00\u7247\u8fde\u7eed\u3001\u5e73\u6ed1\u3001\u4e14\u7ed3\u6784\u826f\u597d\u7684\u201c\u5927\u9646\u201d\u3002\u8fd9\u7247\u5927\u9646\u7684\u4e2d\u5fc3\u662f\u539f\u70b9&#xff0c;\u5e76\u4e14\u5927\u90e8\u5206\u533a\u57df\u90fd\u88ab\u6570\u636e\u6240\u8986\u76d6&#xff0c;\u6ca1\u6709\u65e0\u6cd5\u89e3\u7801\u7684\u201c\u65e0\u4eba\u533a\u201d\u3002 \u5982\u6b64\u4e00\u6765&#xff0c;\u6211\u4eec\u4fbf\u5f97\u5230\u4e86\u4e00\u5f20\u5b8c\u7f8e\u7684\u201c\u521b\u4e16\u5730\u56fe\u201d\u3002\u4efb\u4f55\u4ece\u8fd9\u5f20\u5730\u56fe\u7684\u4e2d\u5fc3\u533a\u57df&#xff08;\u5373\u6807\u51c6\u6b63\u6001\u5206\u5e03&#xff09;\u968f\u673a\u91c7\u6837\u51fa\u7684\u4e00\u4e2a\u70b9z&#xff0c;\u901a\u8fc7\u89e3\u7801\u5668&#xff0c;\u90fd\u80fd\u751f\u6210\u4e00\u4e2a\u770b\u8d77\u6765\u201c\u5408\u7406\u201d\u7684\u3001\u5168\u65b0\u7684\u6837\u672c\u3002\u751f\u6210\u80fd\u529b&#xff0c;\u7531\u6b64\u8bde\u751f\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>10.2.2 \u6a21\u578b\u67b6\u6784\u4e0e\u635f\u5931\u51fd\u6570<\/h5>\n<ul>\n<li>\n<p>\u7f16\u7801\u5668&#xff08;Encoder&#xff09; \u5176\u7ed3\u6784\u4e0e\u6807\u51c6AE\u7684\u7f16\u7801\u5668\u7c7b\u4f3c&#xff08;\u5982\u4e00\u7cfb\u5217\u5377\u79ef\u5c42&#xff09;&#xff0c;\u4f46\u5b83\u7684\u6700\u7ec8\u8f93\u51fa\u662f\u4e24\u4e2a\u72ec\u7acb\u7684\u5411\u91cf&#xff1a;\u5747\u503c\u5411\u91cf\u03bc\u548c\u5bf9\u6570\u65b9\u5dee\u5411\u91cflog(\u03c3\u00b2)\u3002\u8f93\u51fa\u5bf9\u6570\u65b9\u5dee\u800c\u4e0d\u662f\u76f4\u63a5\u8f93\u51fa\u65b9\u5dee\u03c3\u00b2&#xff0c;\u662f\u4e3a\u4e86\u4fdd\u8bc1\u65b9\u5dee\u503c\u4e3a\u6b63&#xff0c;\u5e76\u589e\u52a0\u8bad\u7ec3\u7684\u6570\u503c\u7a33\u5b9a\u6027\u3002<\/p>\n<\/li>\n<li>\n<p>\u91c7\u6837&#xff08;Sampling&#xff09;\u4e0e\u91cd\u53c2\u6570\u5316\u6280\u5de7 \u63a5\u4e0b\u6765&#xff0c;\u6211\u4eec\u9700\u8981\u4ece\u7f16\u7801\u5668\u5b9a\u4e49\u7684\u5206\u5e03N(\u03bc, \u03c3\u00b2)\u4e2d\u91c7\u6837\u4e00\u4e2az\u3002\u7136\u800c&#xff0c;\u201c\u91c7\u6837\u201d\u8fd9\u4e2a\u64cd\u4f5c\u672c\u8eab\u662f\u968f\u673a\u7684&#xff0c;\u4e0d\u53ef\u5fae\u5206\u7684&#xff0c;\u8fd9\u610f\u5473\u7740\u68af\u5ea6\u65e0\u6cd5\u4ece\u89e3\u7801\u5668\u56de\u4f20\u5230\u7f16\u7801\u5668\u3002\u8fd9\u662f\u4e00\u4e2a\u81f4\u547d\u7684\u95ee\u9898\u3002 VAE\u4f7f\u7528\u4e86\u4e00\u4e2a\u88ab\u79f0\u4e3a**\u91cd\u53c2\u6570\u5316\u6280\u5de7&#xff08;Reparameterization Trick&#xff09;**\u7684\u5929\u624d\u65b9\u6cd5\u6765\u89e3\u51b3\u5b83\u3002\u6211\u4eec\u4e0d\u76f4\u63a5\u4eceN(\u03bc, \u03c3\u00b2)\u4e2d\u91c7\u6837&#xff0c;\u800c\u662f\u6362\u4e00\u79cd\u7b49\u4ef7\u7684\u65b9\u5f0f&#xff1a;<\/p>\n<li>\u9996\u5148&#xff0c;\u4ece\u4e00\u4e2a\u56fa\u5b9a\u7684\u3001\u4e0e\u6a21\u578b\u53c2\u6570\u65e0\u5173\u7684\u6807\u51c6\u6b63\u6001\u5206\u5e03N(0, 1)\u4e2d\u91c7\u6837\u4e00\u4e2a\u968f\u673a\u566a\u58f0\u03b5\u3002<\/li>\n<li>\u7136\u540e&#xff0c;\u901a\u8fc7z &#061; \u03bc &#043; \u03c3 * \u03b5\u00a0&#xff08;\u5176\u4e2d\u00a0\u03c3 &#061; exp(0.5 * log(\u03c3\u00b2))&#xff09;\u6765\u8ba1\u7b97z\u3002 \u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u968f\u673a\u6027\u88ab\u8f6c\u79fb\u5230\u4e86\u03b5\u8fd9\u4e2a\u5916\u90e8\u8f93\u5165\u4e0a&#xff0c;\u800cz\u4e0e\u6a21\u578b\u53c2\u6570\u03bc\u548c\u03c3\u4e4b\u95f4\u7684\u8ba1\u7b97\u8def\u5f84\u662f\u5b8c\u5168\u786e\u5b9a\u7684\u3001\u53ef\u5fae\u5206\u7684\u3002\u68af\u5ea6\u5f97\u4ee5\u987a\u5229\u5730\u53cd\u5411\u4f20\u64ad\u3002<\/li>\n<\/li>\n<li>\n<p>\u89e3\u7801\u5668&#xff08;Decoder&#xff09; \u5176\u7ed3\u6784\u4e0eGAN\u7684\u751f\u6210\u5668\u7c7b\u4f3c&#xff08;\u5982\u4e00\u7cfb\u5217\u8f6c\u7f6e\u5377\u79ef\u5c42&#xff09;&#xff0c;\u5b83\u63a5\u6536\u4e00\u4e2a\u4ece\u6f5c\u5728\u7a7a\u95f4\u91c7\u6837\u51fa\u7684\u5411\u91cfz&#xff0c;\u5e76\u5c3d\u529b\u5c06\u5176\u91cd\u6784\u4e3a\u539f\u59cb\u7684\u8f93\u5165\u56fe\u50cfx\u3002<\/p>\n<\/li>\n<li>\n<p>\u53cc\u91cd\u635f\u5931\u51fd\u6570&#xff1a;\u5728\u201c\u590d\u523b\u201d\u4e0e\u201c\u6cdb\u5316\u201d\u95f4\u5bfb\u6c42\u5e73\u8861 VAE\u7684\u635f\u5931\u51fd\u6570&#xff0c;\u5b8c\u7f8e\u5730\u4f53\u73b0\u4e86\u5176\u8bbe\u8ba1\u54f2\u5b66\u3002\u5b83\u7531\u4e24\u4e2a\u90e8\u5206\u76f8\u52a0\u800c\u6210&#xff1a;<\/p>\n<li>\u91cd\u6784\u635f\u5931&#xff08;Reconstruction Loss&#xff09;&#xff1a;\u8fd9\u90e8\u5206\u4e0e\u6807\u51c6AE\u7684\u76ee\u6807\u4e00\u81f4&#xff0c;\u5373\u8861\u91cf\u89e3\u7801\u5668\u91cd\u6784\u51fa\u7684\u56fe\u50cfx&#039;\u4e0e\u539f\u59cb\u56fe\u50cfx\u4e4b\u95f4\u7684\u5dee\u8ddd\u3002\u5bf9\u4e8e\u56fe\u50cf&#xff0c;\u901a\u5e38\u4f7f\u7528\u5747\u65b9\u8bef\u5dee&#xff08;MSE&#xff09;\u6216\u4e8c\u5143\u4ea4\u53c9\u71b5&#xff08;Binary Cross-Entropy&#xff09;\u3002\u8fd9\u4e00\u9879\u635f\u5931&#xff0c;\u9a71\u52a8\u6a21\u578b\u53bb\u5b66\u4e60\u6709\u610f\u4e49\u7684\u7f16\u7801&#xff0c;\u4fdd\u8bc1\u201c\u590d\u523b\u201d\u5f97\u8db3\u591f\u50cf\u3002<\/li>\n<li>KL\u6563\u5ea6\u635f\u5931&#xff08;KL Divergence Loss&#xff09;&#xff1a;\u8fd9\u662fVAE\u7684\u7075\u9b42\u3002\u5b83\u4f7f\u7528KL\u6563\u5ea6\u6765\u8861\u91cf\u7f16\u7801\u5668\u8f93\u51fa\u7684\u5206\u5e03N(\u03bc, \u03c3\u00b2)\u4e0e\u6211\u4eec\u671f\u671b\u7684\u6807\u51c6\u6b63\u6001\u5206\u5e03N(0, 1)\u4e4b\u95f4\u7684\u201c\u8ddd\u79bb\u201d\u3002\u8fd9\u4e00\u9879\u635f\u5931&#xff0c;\u626e\u6f14\u7740\u4e00\u4e2a\u6b63\u5219\u5316\u9879\u7684\u89d2\u8272&#xff0c;\u5b83\u60e9\u7f5a\u90a3\u4e9b\u504f\u79bb\u6807\u51c6\u6b63\u6001\u5206\u5e03\u592a\u8fdc\u7684\u7f16\u7801\u5206\u5e03&#xff0c;\u8feb\u4f7f\u6f5c\u5728\u7a7a\u95f4\u53d8\u5f97\u89c4\u6574\u3001\u8fde\u7eed&#xff0c;\u4fdd\u8bc1\u4e86\u6a21\u578b\u7684\u201c\u6cdb\u5316\u201d\u4e0e\u751f\u6210\u80fd\u529b\u3002<\/li>\n<\/li>\n<\/ul>\n<p>VAE\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u5728\u201c\u5c3d\u53ef\u80fd\u5730\u5b8c\u7f8e\u91cd\u6784&#xff08;\u964d\u4f4e\u91cd\u6784\u635f\u5931&#xff09;\u201d\u548c\u201c\u5c3d\u53ef\u80fd\u5730\u8ba9\u6f5c\u5728\u7a7a\u95f4\u89c4\u6574&#xff08;\u964d\u4f4eKL\u6563\u5ea6\u635f\u5931&#xff09;\u201d\u8fd9\u4e24\u4e2a\u76ee\u6807\u4e4b\u95f4&#xff0c;\u5bfb\u627e\u4e00\u4e2a\u6700\u4f73\u7684\u5e73\u8861\u70b9\u3002<\/p>\n<h5>10.2.3 VAE vs. GAN&#xff1a;\u4e24\u79cd\u54f2\u5b66\u7684\u5bf9\u6bd4<\/h5>\n<p>VAE\u548cGAN\u662f\u751f\u6210\u6a21\u578b\u9886\u57df\u6700\u7ecf\u5178\u7684\u4e24\u6761\u8def\u7ebf&#xff0c;\u5b83\u4eec\u7684\u5bf9\u6bd4&#xff0c;\u80fd\u8ba9\u6211\u4eec\u66f4\u6df1\u523b\u5730\u7406\u89e3\u5404\u81ea\u7684\u7279\u70b9&#xff1a;<\/p>\n<ul>\n<li>\u751f\u6210\u8d28\u91cf&#xff1a;GAN\u7531\u4e8e\u5176\u5bf9\u6297\u6027\u8bad\u7ec3\u673a\u5236&#xff0c;\u4f1a\u65e0\u60c5\u5730\u60e9\u7f5a\u4efb\u4f55\u6a21\u7cca\u3001\u4e0d\u771f\u5b9e\u7684\u7ed3\u679c&#xff0c;\u56e0\u6b64\u901a\u5e38\u80fd\u751f\u6210\u66f4\u6e05\u6670\u3001\u66f4\u9510\u5229\u3001\u7ec6\u8282\u66f4\u4e30\u5bcc\u7684\u56fe\u50cf\u3002\u800cVAE\u7684\u76ee\u6807\u662f\u6700\u5927\u5316\u6570\u636e\u7684\u5bf9\u6570\u4f3c\u7136\u4e0b\u754c&#xff0c;\u8fd9\u4f7f\u5176\u503e\u5411\u4e8e\u751f\u6210\u66f4\u201c\u5b89\u5168\u201d\u3001\u66f4\u201c\u5e73\u5747\u201d\u7684\u7ed3\u679c&#xff0c;\u56fe\u50cf\u5f80\u5f80\u66f4\u5e73\u6ed1&#xff0c;\u4f46\u6709\u65f6\u4f1a\u663e\u5f97\u6709\u4e9b\u6a21\u7cca\u3002<\/li>\n<li>\u8bad\u7ec3\u7a33\u5b9a\u6027&#xff1a;VAE\u7684\u8bad\u7ec3\u8fc7\u7a0b\u662f\u975e\u5e38\u7a33\u5b9a\u7684&#xff0c;\u56e0\u4e3a\u5b83\u672c\u8d28\u4e0a\u53ea\u662f\u5728\u4f18\u5316\u4e00\u4e2a\u56fa\u5b9a\u7684\u3001\u5b9a\u4e49\u660e\u786e\u7684\u635f\u5931\u51fd\u6570\u3002\u800cGAN\u7684\u8bad\u7ec3\u662f\u5bfb\u627e\u7eb3\u4ec0\u5747\u8861\u7684\u8fc7\u7a0b&#xff0c;\u5145\u6ee1\u4e86\u52a8\u6001\u535a\u5f08&#xff0c;\u6781\u6613\u51fa\u73b0\u6a21\u5f0f\u5d29\u6e83\u7b49\u4e0d\u7a33\u5b9a\u7684\u95ee\u9898\u3002<\/li>\n<li>\u6f5c\u5728\u7a7a\u95f4\u7684\u8d28\u91cf&#xff1a;\u8fd9\u662fVAE\u6700\u5927\u7684\u4f18\u52bf\u3002\u5b83\u663e\u5f0f\u5730\u5b66\u4e60\u4e86\u4e00\u4e2a\u7ed3\u6784\u5316\u3001\u5e73\u6ed1\u4e14\u6709\u610f\u4e49\u7684\u6f5c\u5728\u7a7a\u95f4\u3002\u6211\u4eec\u53ef\u4ee5\u5728\u8fd9\u4e2a\u7a7a\u95f4\u4e2d\u8fdb\u884c\u63d2\u503c\u64cd\u4f5c&#xff08;\u4f8b\u5982&#xff0c;\u5c06\u4e00\u5f20\u7b11\u8138\u7684\u7f16\u7801\u548c\u4e00\u5f20\u4e0d\u7b11\u7684\u4eba\u8138\u7f16\u7801\u8fdb\u884c\u7ebf\u6027\u63d2\u503c&#xff0c;\u53ef\u4ee5\u5f97\u5230\u4e00\u7cfb\u5217\u8868\u60c5\u9010\u6e10\u53d8\u5316\u7684\u5e73\u6ed1\u8fc7\u6e21\u56fe\u50cf&#xff09;&#xff0c;\u6216\u8005\u63a2\u7d22\u4e0d\u540c\u7ef4\u5ea6\u5bf9\u751f\u6210\u7ed3\u679c\u7684\u5f71\u54cd&#xff08;\u5982\u63a7\u5236\u53d1\u8272\u3001\u5e74\u9f84\u7b49&#xff09;\u3002\u800cGAN\u7684\u6f5c\u5728\u7a7a\u95f4\u901a\u5e38\u662f\u201c\u7ea0\u7f20\u201d\u7684&#xff0c;\u5176\u7ed3\u6784\u4e0d\u660e\u786e&#xff0c;\u4e0d\u6613\u8fdb\u884c\u6709\u610f\u4e49\u7684\u64cd\u7eb5\u3002<\/li>\n<\/ul>\n<p>\u603b\u800c\u8a00\u4e4b&#xff0c;\u5982\u679c\u8ffd\u6c42\u6781\u81f4\u7684\u751f\u6210\u903c\u771f\u5ea6&#xff0c;GAN\u53ef\u80fd\u662f\u66f4\u597d\u7684\u9009\u62e9&#xff1b;\u800c\u5982\u679c\u60a8\u66f4\u770b\u91cd\u8bad\u7ec3\u7684\u7a33\u5b9a\u3001\u53ef\u63a7\u6027&#xff0c;\u4ee5\u53ca\u4e00\u4e2a\u5bcc\u6709\u89e3\u91ca\u6027\u7684\u6f5c\u5728\u7a7a\u95f4&#xff0c;VAE\u5219\u5c55\u73b0\u51fa\u5176\u72ec\u7279\u7684\u6982\u7387\u4e4b\u7f8e\u3002<\/p>\n<hr \/>\n<h4>10.3 \u6269\u6563\u6a21\u578b&#xff08;Diffusion Models&#xff09;&#xff1a;\u4ece\u566a\u58f0\u4e2d\u96d5\u523b\u6770\u4f5c<\/h4>\n<p>\u8bfb\u8005\u670b\u53cb\u4eec\u3002\u6211\u4eec\u5df2\u7ecf\u63a2\u7d22\u4e86GAN\u7684\u5bf9\u6297\u4e16\u754c\u548cVAE\u7684\u6982\u7387\u7a7a\u95f4\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u5c06\u8981\u8fdb\u5165\u4e00\u7247\u8fd1\u5e74\u6765\u5f02\u519b\u7a81\u8d77\u3001\u5e76\u8fc5\u901f\u5e2d\u5377\u6574\u4e2a\u751f\u6210\u9886\u57df\u7684\u5168\u65b0\u5927\u9646\u3002\u8fd9\u91cc\u7684\u521b\u9020\u6cd5\u5219\u65e2\u4e0d\u540c\u4e8e\u535a\u5f08&#xff0c;\u4e5f\u4e0d\u540c\u4e8e\u7f16\u7801&#xff0c;\u5b83\u6e90\u4e8e\u4e00\u79cd\u66f4\u7269\u7406\u3001\u66f4\u76f4\u89c2\u7684\u8fc7\u7a0b\u2014\u2014\u6709\u5e8f\u5730\u7834\u574f&#xff0c;\u518d\u9006\u5411\u5730\u91cd\u6784\u3002\u8fd9&#xff0c;\u5c31\u662f\u6269\u6563\u6a21\u578b&#xff08;Diffusion Models&#xff09;\u7684\u827a\u672f\u3002<\/p>\n<p>\u57282020\u5e74\u4e4b\u540e&#xff0c;\u4e00\u7c7b\u88ab\u79f0\u4e3a\u6269\u6563\u6a21\u578b&#xff08;Denoising Diffusion Probabilistic Models, DDPMs&#xff09;\u7684\u751f\u6210\u6a21\u578b&#xff0c;\u4ee5\u5176\u65e0\u4e0e\u4f26\u6bd4\u7684\u751f\u6210\u8d28\u91cf\u548c\u7a33\u5b9a\u7684\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u8fc5\u901f\u4ece\u5b66\u672f\u754c\u7684\u5ba0\u513f&#xff0c;\u6210\u4e3a\u4e86\u9a71\u52a8DALL-E 2\u3001Midjourney\u3001Stable Diffusion\u7b49\u9876\u7ea7\u6587\u751f\u56fe\u5e94\u7528\u7684\u5e55 \u7075\u611f\u6765\u6e90&#xff1a;\u975e\u5e73\u8861\u70ed\u529b\u5b66 \u6269\u6563\u6a21\u578b\u7684\u7075\u611f&#xff0c;\u6e90\u4e8e\u7269\u7406\u5b66\u4e2d\u7684\u975e\u5e73\u8861\u70ed\u529b\u5b66\u3002\u60f3\u8c61\u4e00\u6ef4\u58a8\u6c34\u6ef4\u5165\u4e00\u676f\u6e05\u6c34\u4e2d&#xff0c;\u58a8\u6c34\u5206\u5b50\u4f1a\u81ea\u53d1\u5730\u3001\u968f\u673a\u5730\u5411\u56db\u5468\u6269\u6563&#xff0c;\u76f4\u5230\u6700\u7ec8\u5747\u5300\u5730\u5206\u5e03\u5728\u6574\u676f\u6c34\u4e2d&#xff0c;\u7cfb\u7edf\u8fbe\u5230\u6700\u5927\u71b5\u7684\u65e0\u5e8f\u72b6\u6001\u3002\u6269\u6563\u6a21\u578b\u505a\u7684&#xff0c;\u5c31\u662f\u6a21\u4eff\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u7136\u540e\u518d\u5b66\u4e60\u5982\u4f55\u5c06\u8fd9\u4e2a\u8fc7\u7a0b\u201c\u5012\u5e26*\u3002<\/p>\n<ul>\n<li>\n<p>\u524d\u5411\u8fc7\u7a0b&#xff08;Forward Process \/ Diffusion Process&#xff09;&#xff1a;\u8d70\u5411\u6df7\u6c8c \u8fd9\u662f\u6269\u6563\u6a21\u578b\u4e2d\u4e00\u4e2a\u56fa\u5b9a\u7684\u3001\u65e0\u9700\u5b66\u4e60\u7684\u3001\u7eaf\u7cb9\u7684\u6570\u5b66\u8fc7\u7a0b\u3002<\/p>\n<li>\u6211\u4eec\u4ece\u4e00\u5f20\u771f\u5b9e\u7684\u3001\u6e05\u6670\u7684\u56fe\u50cfx_0\u5f00\u59cb\u3002<\/li>\n<li>\u6211\u4eec\u5b9a\u4e49\u4e00\u4e2a\u5f88\u957f\u7684\u65f6\u95f4\u6b65\u957fT&#xff08;\u4f8b\u5982&#xff0c;T&#061;1000&#xff09;\u3002<\/li>\n<li>\u5728\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65t&#xff08;\u4ece1\u5230T&#xff09;&#xff0c;\u6211\u4eec\u90fd\u5411\u4e0a\u4e00\u65f6\u523b\u7684\u56fe\u50cfx_{t-1}\u4e2d&#xff0c;\u6dfb\u52a0\u5c11\u91cf\u3001\u53ef\u63a7\u7684\u9ad8\u65af\u566a\u58f0&#xff0c;\u5f97\u5230x_t\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u7531\u4e00\u4e2a\u56fa\u5b9a\u7684\u65b9\u5dee\u8c03\u5ea6\u8868\u03b2_t\u6765\u63a7\u5236&#xff0c;\u4fdd\u8bc1\u4e86\u6bcf\u4e00\u6b65\u6dfb\u52a0\u7684\u566a\u58f0\u91cf\u662f\u9884\u5148\u5b9a\u4e49\u597d\u7684\u3002<\/li>\n<li>\u8fd9\u4e2a\u8fc7\u7a0b\u91cd\u590dT\u6b21\u4e4b\u540e&#xff0c;\u539f\u59cb\u7684\u3001\u4fe1\u606f\u4e30\u5bcc\u7684\u56fe\u50cfx_0&#xff0c;\u4f1a\u9010\u6e10\u53d8\u5f97\u6a21\u7cca&#xff0c;\u6700\u7ec8\u5728t&#061;T\u65f6&#xff0c;\u5b8c\u5168\u53d8\u6210\u4e00\u5f20\u7eaf\u7cb9\u7684\u3001\u65e0\u610f\u4e49\u7684\u3001\u4e0e\u539f\u59cb\u56fe\u50cf\u65e0\u5173\u7684\u6807\u51c6\u9ad8\u65af\u566a\u58f0\u56fe\u50cfx_T\u3002<\/li>\n<p>\u8fd9\u4e2a\u524d\u5411\u8fc7\u7a0b\u7684\u7f8e\u5999\u4e4b\u5904\u5728\u4e8e&#xff0c;\u7531\u4e8e\u6bcf\u4e00\u6b65\u6dfb\u52a0\u7684\u90fd\u662f\u9ad8\u65af\u566a\u58f0&#xff0c;\u6211\u4eec\u53ef\u4ee5\u63a8\u5bfc\u51fa\u4e00\u4e2a\u5c01\u95ed\u89e3&#xff08;Closed-form Solution&#xff09;&#xff1a;\u5bf9\u4e8e\u4efb\u610f\u4e00\u4e2a\u65f6\u95f4\u6b65t&#xff0c;\u6211\u4eec\u90fd\u53ef\u4ee5\u76f4\u63a5\u901a\u8fc7x_0\u548ct&#xff0c;\u4e00\u6b65\u8ba1\u7b97\u51fax_t\u7684\u6837\u5b50&#xff0c;\u800c\u65e0\u9700\u4ece\u5934\u6a21\u62dft\u6b21\u3002\u8fd9\u4e3a\u540e\u7eed\u9ad8\u6548\u7684\u8bad\u7ec3\u5960\u5b9a\u4e86\u57fa\u7840\u3002<\/p>\n<\/li>\n<li>\n<p>\u53cd\u5411\u8fc7\u7a0b&#xff08;Reverse Process \/ Denoising Process&#xff09;&#xff1a;\u91cd\u5851\u79e9\u5e8f \u8fd9\u624d\u662f\u6a21\u578b\u771f\u6b63\u9700\u8981\u5b66\u4e60\u7684\u6838\u5fc3\u90e8\u5206&#xff0c;\u6df7\u6c8c\u7684\u566a\u58f0\u4e2d&#xff0c;\u6062\u590d\u51fa\u4e00\u5f20\u6e05\u6670\u3001\u903c\u771f\u3001\u4e14\u4ece\u672a\u89c1\u8fc7\u7684\u56fe\u50cfx_0\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b66\u4e60\u7684\u76ee\u6807&#xff1a;\u9884\u6d4b\u566a\u58f0&#xff0c;\u800c\u975e\u56fe\u50cf \u76f4\u63a5\u8ba9\u6a21\u578b\u4ece\u5e26\u566a\u56fe\u50cfx_t\u9884\u6d4b\u51fa\u66f4\u5e72\u51c0\u7684x_{t-1}\u662f\u56f0\u96be\u7684\u3002\u4e00\u4e2a\u66f4\u5de7\u5999\u3001\u4e5f\u66f4\u6709\u6548\u7684\u5b66\u4e60\u76ee\u6807\u662f&#xff1a;\u5728\u53cd\u5411\u8fc7\u7a0b\u7684\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65t&#xff0c;\u8ba9\u6a21\u578b\u53bb\u9884\u6d4b\u51fa\u5728\u5f53\u524d\u8fd9\u5f20\u5e26\u566a\u56fe\u50cfx_t\u4e2d&#xff0c;\u6240\u5305\u542b\u7684\u566a\u58f0\u03b5\u662f\u4ec0\u4e48\u3002 \u8fd9\u4e2a\u4efb\u52a1\u76f8\u5bf9\u66f4\u7b80\u5355\u3001\u66f4\u660e\u786e\u3002\u53ea\u8981\u6a21\u578b\u80fd\u591f\u7cbe\u51c6\u5730\u9884\u6d4b\u51fa\u6211\u4eec\u5f53\u521d\u5728\u524d\u5411\u8fc7\u7a0b\u4e2d\u6dfb\u52a0\u7684\u566a\u58f0\u03b5&#xff0c;\u6211\u4eec\u5c31\u53ef\u4ee5\u8f7b\u6613\u5730\u4ecex_t\u4e2d\u201c\u51cf\u53bb\u201d\u8fd9\u4e2a\u566a\u58f0\u7684\u5f71\u54cd&#xff0c;\u4ece\u800c\u5f97\u5230\u4e00\u4e2a\u5bf9x_{t-1}\u7684\u826f\u597d\u4f30\u8ba1\u3002\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u50cf\u4e00\u4f4d\u6280\u827a\u9ad8\u8d85\u7684\u96d5\u5851\u5bb6&#xff0c;\u4ed6\u4e0d\u662f\u5728\u51ed\u7a7a\u521b\u9020&#xff0c;\u800c\u662f\u5728\u4e00\u5757\u6df7\u6c8c\u7684\u77f3\u6599&#xff08;\u5e26\u566a\u56fe\u50cf&#xff09;\u4e2d&#xff0c;\u7cbe\u51c6\u5730\u201c\u51ff\u53bb\u201d\u90a3\u4e9b\u591a\u4f59\u7684\u90e8\u5206&#xff08;\u566a\u58f0&#xff09;&#xff0c;\u4ece\u800c\u8ba9\u5185\u85cf\u7684\u6770\u4f5c&#xff08;\u5e72\u51c0\u56fe\u50cf&#xff09;\u663e\u73b0\u51fa\u6765\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>10.3.2 \u8bad\u7ec3\u4e0e\u751f\u6210<\/h5>\n<ul>\n<li>\n<p>\u6a21\u578b\u67b6\u6784&#xff1a;U-Net\u7684\u56de\u5f52 \u5728\u6269\u6563\u6a21\u578b\u4e2d&#xff0c;\u7528\u4e8e\u9884\u6d4b\u566a\u58f0\u7684\u795e\u7ecf\u7f51\u7edc&#xff0c;\u6700\u7ecf\u5178\u3001\u6700\u4e3b\u6d41\u7684\u67b6\u6784\u662fU-Net\u3002U-Net\u6700\u521d\u662f\u4e3a\u533b\u5b66\u56fe\u50cf\u5206\u5272\u8bbe\u8ba1\u7684&#xff0c;\u5176\u7279\u70b9\u662f\u62e5\u6709\u4e00\u4e2a\u5bf9\u79f0\u7684\u201c\u7f16\u7801\u5668-\u89e3\u7801\u5668\u201d\u7ed3\u6784&#xff0c;\u5e76\u4e14\u5728\u7f16\u7801\u5668\u7684\u4e0b\u91c7\u6837\u8def\u5f84\u548c\u89e3\u7801\u5668\u7684\u4e0a\u91c7\u6837\u8def\u5f84\u4e4b\u95f4&#xff0c;\u5b58\u5728\u7740\u8df3\u8dc3\u8fde\u63a5&#xff08;Skip Connections&#xff09;\u3002\u8fd9\u79cd\u7ed3\u6784&#xff0c;\u4f7f\u5f97\u6a21\u578b\u65e2\u80fd\u6355\u6349\u5230\u56fe\u50cf\u7684\u5168\u5c40\u4e0a\u4e0b\u6587\u4fe1\u606f&#xff0c;\u53c8\u80fd\u4fdd\u7559\u7cbe\u7ec6\u7684\u5c40\u90e8\u7ec6\u8282&#xff0c;\u975e\u5e38\u9002\u5408\u4e8e\u8fdb\u884c\u50cf\u7d20\u7ea7\u522b\u7684\u566a\u58f0\u9884\u6d4b\u3002\u5728\u73b0\u4ee3\u6269\u6563\u6a21\u578b\u4e2d&#xff0c;U-Net\u7684\u5185\u90e8\u901a\u5e38\u8fd8\u4f1a\u878d\u5165\u81ea\u6ce8\u610f\u529b\u6a21\u5757&#xff0c;\u4ee5\u589e\u5f3a\u5176\u5bf9\u957f\u8ddd\u79bb\u4f9d\u8d56\u7684\u5efa\u6a21\u80fd\u529b\u3002<\/p>\n<\/li>\n<li>\n<p>\u8bad\u7ec3\u8fc7\u7a0b \u6269\u6563\u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b\u975e\u5e38\u7a33\u5b9a\u4e14\u9ad8\u6548&#xff1a;<\/p>\n<li>\u4ece\u8bad\u7ec3\u96c6\u4e2d\u968f\u673a\u62bd\u53d6\u4e00\u5f20\u771f\u5b9e\u56fe\u50cfx_0\u3002<\/li>\n<li>\u968f\u673a\u9009\u62e9\u4e00\u4e2a\u65f6\u95f4\u6b65t&#xff08;\u57281\u5230T\u4e4b\u95f4&#xff09;\u3002<\/li>\n<li>\u4ece\u6807\u51c6\u9ad8\u65af\u5206\u5e03\u4e2d\u91c7\u6837\u4e00\u4e2a\u566a\u58f0\u03b5\u3002<\/li>\n<li>\u5229\u7528\u524d\u5411\u8fc7\u7a0b\u7684\u5c01\u95ed\u89e3&#xff0c;\u76f4\u63a5\u8ba1\u7b97\u51fa\u5728t\u65f6\u523b\u7684\u5e26\u566a\u56fe\u50cfx_t\u3002<\/li>\n<li>\u5c06x_t\u548c\u65f6\u95f4\u6b65t&#xff08;t\u901a\u5e38\u4f1a\u88ab\u7f16\u7801\u6210\u4e00\u4e2a\u5411\u91cf&#xff09;\u4e00\u8d77\u8f93\u5165\u5230U-Net\u6a21\u578b\u4e2d\u3002<\/li>\n<li>\u6a21\u578b\u7684\u635f\u5931\u51fd\u6570&#xff0c;\u5c31\u662fU-Net\u9884\u6d4b\u51fa\u7684\u566a\u58f0\u03b5_\u03b8(x_t, t)\u4e0e\u6211\u4eec\u771f\u5b9e\u6dfb\u52a0\u7684\u566a\u58f0\u03b5\u4e4b\u95f4\u7684\u5747\u65b9\u8bef\u5dee&#xff08;MSE&#xff09;\u3002 \u6211\u4eec\u53ea\u9700\u8981\u4f18\u5316\u8fd9\u4e2a\u7b80\u5355\u7684MSE\u635f\u5931\u5373\u53ef&#xff0c;\u6574\u4e2a\u8fc7\u7a0b\u6ca1\u6709\u5bf9\u6297&#xff0c;\u6ca1\u6709\u590d\u6742\u7684KL\u6563\u5ea6&#xff0c;\u975e\u5e38\u7a33\u5b9a\u3002<\/li>\n<\/li>\n<li>\n<p>\u751f\u6210&#xff08;\u91c7\u6837&#xff09;\u8fc7\u7a0b \u5f53\u6a21\u578b\u8bad\u7ec3\u597d\u540e&#xff0c;\u751f\u6210\u65b0\u56fe\u50cf\u7684\u8fc7\u7a0b\u5c31\u662f\u4e00\u4e2a**\u8fed\u4ee3\u5f0f\u7684\u201c\u53bb\u566a\u201d**\u8fc7\u7a0b&#xff1a;<\/p>\n<li>\u9996\u5148&#xff0c;\u4ece\u6807\u51c6\u9ad8\u65af\u5206\u5e03\u4e2d\u968f\u673a\u91c7\u6837\u4e00\u5f20\u4e0e\u76ee\u6807\u5c3a\u5bf8\u76f8\u540c\u7684\u7eaf\u566a\u58f0\u56fe\u50cf&#xff0c;\u4f5c\u4e3ax_T\u3002<\/li>\n<li>\u7136\u540e&#xff0c;\u8fdb\u884c\u4e00\u4e2a\u4ecet&#061;T\u5230t&#061;1\u7684\u5faa\u73af&#xff1a;\n<ul>\n<li>\u5728\u6bcf\u4e00\u4e2a\u65f6\u95f4\u6b65t&#xff0c;\u5c06\u5f53\u524d\u7684\u56fe\u50cfx_t\u548c\u65f6\u95f4\u6b65t\u8f93\u5165\u5230\u8bad\u7ec3\u597d\u7684U-Net\u6a21\u578b\u4e2d&#xff0c;\u5f97\u5230\u4e00\u4e2a\u5bf9\u566a\u58f0\u7684\u9884\u6d4b\u03b5_\u03b8(x_t, t)\u3002<\/li>\n<li>\u4f7f\u7528\u8fd9\u4e2a\u9884\u6d4b\u51fa\u7684\u566a\u58f0&#xff0c;\u7ed3\u5408\u53cd\u5411\u8fc7\u7a0b\u7684\u6570\u5b66\u516c\u5f0f&#xff0c;\u8ba1\u7b97\u51fa\u4e0a\u4e00\u4e2a\u65f6\u523b\u7684\u3001\u66f4\u5e72\u51c0\u7684\u56fe\u50cfx_{t-1}\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u5faa\u73af\u7ed3\u675f\u540e&#xff0c;x_0\u5c31\u662f\u6211\u4eec\u6700\u7ec8\u751f\u6210\u7684\u6e05\u6670\u56fe\u50cf\u3002<\/li>\n<\/li>\n<\/ul>\n<p>\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u50cf\u770b\u7740\u4e00\u4f4d\u9690\u5f62\u7684\u96d5\u5851\u5bb6&#xff0c;\u5728\u4e00\u5757\u5145\u6ee1\u96ea\u82b1\u70b9\u7684\u5c4f\u5e55\u4e0a&#xff0c;\u4e00\u7b14\u4e00\u7b14\u5730\u64e6\u9664\u566a\u58f0&#xff0c;\u6700\u7ec8\u8ba9\u4e00\u5e45\u7cbe\u7f8e\u7684\u753b\u4f5c\u6d6e\u73b0\u51fa\u6765&#xff0c;\u5145\u6ee1\u4e86\u4eea\u5f0f\u611f\u548c\u827a\u672f\u611f\u3002<\/p>\n<h5>10.3.3 \u6269\u6563\u6a21\u578b\u7684\u5d1b\u8d77<\/h5>\n<ul>\n<li>\n<p>\u65e0\u4e0e\u4f26\u6bd4\u7684\u751f\u6210\u8d28\u91cf\u4e0e\u591a\u6837\u6027 \u6269\u6563\u6a21\u578b\u6700\u4ee4\u4eba\u79f0\u9053\u7684&#xff0c;\u5c31\u662f\u5176\u5353\u8d8a\u7684\u751f\u6210\u8d28\u91cf\u548c\u591a\u6837\u6027\u3002\u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d&#xff0c;\u5b83\u751f\u6210\u7684\u56fe\u50cf\u5728\u4fdd\u771f\u5ea6&#xff08;Fidelity&#xff09;\u548c\u591a\u6837\u6027&#xff08;Diversity&#xff09;\u4e0a&#xff0c;\u90fd\u5168\u9762\u8d85\u8d8a\u4e86\u6700\u9876\u5c16\u7684GAN\u6a21\u578b\u3002\u5b83\u5f88\u597d\u5730\u907f\u514d\u4e86GAN\u7684\u6a21\u5f0f\u5d29\u6e83\u95ee\u9898&#xff0c;\u80fd\u591f\u751f\u6210\u8986\u76d6\u6574\u4e2a\u6570\u636e\u5206\u5e03\u7684\u3001\u4e30\u5bcc\u591a\u5f69\u7684\u6837\u672c\u3002<\/p>\n<\/li>\n<li>\n<p>\u7a33\u5b9a\u7684\u8bad\u7ec3\u4e0e\u5f3a\u5927\u7684\u53ef\u63a7\u6027 \u5982\u524d\u6240\u8ff0&#xff0c;\u6269\u6563\u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b\u975e\u5e38\u7a33\u5b9a\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u5b83\u975e\u5e38\u5bb9\u6613\u4e0e\u6761\u4ef6\u4fe1\u606f\u76f8\u7ed3\u5408&#xff0c;\u5b9e\u73b0\u5f3a\u5927\u7684\u53ef\u63a7\u751f\u6210\u3002\u901a\u8fc7\u5c06\u989d\u5916\u7684\u6761\u4ef6&#xff08;\u5982\u7c7b\u522b\u6807\u7b7e&#xff0c;\u6216\u66f4\u5f3a\u5927\u7684CLIP\u6a21\u578b\u7f16\u7801\u51fa\u7684\u6587\u672c\u7279\u5f81&#xff09;\u4e00\u540c\u8f93\u5165U-Net&#xff0c;\u6a21\u578b\u5c31\u80fd\u5b66\u4f1a\u5728\u53bb\u566a\u7684\u8fc7\u7a0b\u4e2d&#xff0c;\u59cb\u7ec8\u671d\u7740\u7b26\u5408\u6761\u4ef6\u63cf\u8ff0\u7684\u65b9\u5411\u8fdb\u884c\u201c\u96d5\u523b\u201d\u3002\u8fd9\u6b63\u662fStable Diffusion\u7b49\u6587\u751f\u56fe\u6a21\u578b\u80fd\u591f\u6839\u636e\u7528\u6237\u8f93\u5165\u7684\u4efb\u610f\u6587\u672c&#xff0c;\u751f\u6210\u9ad8\u8d28\u91cf\u56fe\u50cf\u7684\u5e95\u5c42\u6838\u5fc3\u6280\u672f\u3002\u8fd9\u79cd\u5f15\u5bfc&#xff08;Guidance&#xff09;\u6280\u672f&#xff0c;\u4f7f\u5f97\u6269\u6563\u6a21\u578b\u7684\u53ef\u63a7\u6027\u8fbe\u5230\u4e86\u524d\u6240\u672a\u6709\u7684\u9ad8\u5ea6\u3002<\/p>\n<\/li>\n<li>\n<p>\u7f3a\u70b9&#xff1a;\u6162\u901f\u7684\u201c\u827a\u672f\u521b\u4f5c\u201d \u6269\u6563\u6a21\u578b\u76ee\u524d\u6700\u4e3b\u8981\u7684\u7f3a\u70b9&#xff0c;\u5c31\u662f\u5176\u91c7\u6837\u901f\u5ea6\u76f8\u5bf9\u8f83\u6162\u3002\u56e0\u4e3a\u751f\u6210\u4e00\u5f20\u56fe\u50cf&#xff0c;\u9700\u8981\u5b8c\u6574\u5730\u6267\u884c\u4e0a\u767e\u751a\u81f3\u4e0a\u5343\u6b21\u7684\u8fed\u4ee3\u53bb\u566a\u6b65\u9aa4&#xff0c;\u6bcf\u4e00\u6b65\u90fd\u9700\u8981\u4e00\u6b21\u5b8c\u6574\u7684\u795e\u7ecf\u7f51\u7edc\u524d\u5411\u4f20\u64ad\u3002\u8fd9\u4e0eGAN\u53ea\u9700\u4e00\u6b21\u524d\u5411\u4f20\u64ad\u5c31\u80fd\u751f\u6210\u56fe\u50cf\u5f62\u6210\u4e86\u9c9c\u660e\u5bf9\u6bd4\u3002\u4e0d\u8fc7&#xff0c;\u5b66\u672f\u754c\u548c\u5de5\u4e1a\u754c\u6b63\u5728\u79ef\u6781\u7814\u7a76\u5404\u79cd\u52a0\u901f\u91c7\u6837\u7684\u65b9\u6cd5&#xff08;\u5982DDIM\u3001\u51cf\u5c11\u91c7\u6837\u6b65\u6570\u3001\u77e5\u8bc6\u84b8\u998f\u7b49&#xff09;&#xff0c;\u5e76\u5df2\u7ecf\u53d6\u5f97\u4e86\u663e\u8457\u7684\u8fdb\u5c55\u3002<\/p>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>10.4 \u5e94\u7528&#xff1a;\u5f53\u673a\u5668\u5f00\u59cb\u201c\u505a\u68a6\u201d<\/h4>\n<p>\u638c\u63e1\u4e86GAN\u3001VAE\u3001\u6269\u6563\u6a21\u578b\u8fd9\u4e9b\u5f3a\u5927\u7684\u751f\u6210\u5de5\u5177\u540e&#xff0c;\u6211\u4eec\u4fbf\u5f00\u542f\u4e86\u4e00\u6247\u901a\u5f80AIGC&#xff08;\u4eba\u5de5\u667a\u80fd\u751f\u6210\u5185\u5bb9&#xff09;\u65f6\u4ee3\u7684\u5927\u95e8\u3002\u8fd9\u4e9b\u6a21\u578b\u5982\u540c\u88ab\u8d4b\u4e88\u4e86\u201c\u68a6\u5883\u201d\u80fd\u529b\u7684\u673a\u5668&#xff0c;\u5176\u5e94\u7528\u573a\u666f\u6b63\u5728\u4ee5\u524d\u6240\u672a\u6709\u7684\u901f\u5ea6\u6269\u5c55\u3002<\/p>\n<h5>10.4.1 \u56fe\u50cf\u751f\u6210\u4e0e\u7f16\u8f91<\/h5>\n<ul>\n<li>\u65e0\u6761\u4ef6\u751f\u6210&#xff1a;\u8fd9\u662f\u751f\u6210\u6a21\u578b\u6700\u57fa\u7840\u7684\u80fd\u529b\u3002\u4f8b\u5982&#xff0c;StyleGAN\u7cfb\u5217\u53ef\u4ee5\u751f\u6210\u903c\u771f\u5230\u8089\u773c\u65e0\u6cd5\u5206\u8fa8\u7684\u3001\u4f46\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u5e76\u4e0d\u5b58\u5728\u7684\u865a\u62df\u4eba\u8138\u3002\u8fd9\u5728\u865a\u62df\u5f62\u8c61\u521b\u5efa\u3001\u6e38\u620f\u89d2\u8272\u8bbe\u8ba1\u3001\u827a\u672f\u521b\u4f5c\u7b49\u9886\u57df\u5177\u6709\u5de8\u5927\u4ef7\u503c\u3002<\/li>\n<li>\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210&#xff08;Text-to-Image&#xff09;&#xff1a;\u8fd9\u662f\u5f53\u524d\u6700\u706b\u70ed\u3001\u4e5f\u6700\u5177\u53d8\u9769\u6027\u7684\u5e94\u7528\u3002\u7528\u6237\u53ea\u9700\u8f93\u5165\u4e00\u6bb5\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0&#xff08;\u4f8b\u5982&#xff0c;\u201c\u4e00\u53ea\u7a7f\u7740\u5b87\u822a\u670d\u7684\u67ef\u57fa\u72ac&#xff0c;\u5728\u6708\u7403\u4e0a\u9a91\u7740\u4e00\u5339\u9a6c&#xff0c;\u6570\u5b57\u827a\u672f\u98ce\u683c\u201d&#xff09;&#xff0c;\u6a21\u578b&#xff08;\u5982Stable Diffusion, Midjourney&#xff09;\u5c31\u80fd\u751f\u6210\u7b26\u5408\u8fd9\u6bb5\u5929\u9a6c\u884c\u7a7a\u63cf\u8ff0\u7684\u3001\u9ad8\u8d28\u91cf\u7684\u56fe\u50cf\u3002\u8fd9\u6781\u5927\u5730\u964d\u4f4e\u4e86\u89c6\u89c9\u5185\u5bb9\u521b\u4f5c\u7684\u95e8\u69db&#xff0c;\u4e3a\u5e7f\u544a\u3001\u8bbe\u8ba1\u3001\u5a31\u4e50\u7b49\u884c\u4e1a\u5e26\u6765\u4e86\u98a0\u8986\u6027\u7684\u53d8\u9769\u3002<\/li>\n<li>\u56fe\u50cf\u4fee\u590d&#xff08;Inpainting&#xff09;\u4e0e\u6269\u5c55&#xff08;Outpainting&#xff09;&#xff1a;\u5229\u7528\u751f\u6210\u6a21\u578b\u5bf9\u56fe\u50cf\u4e0a\u4e0b\u6587\u7684\u6df1\u523b\u7406\u89e3&#xff0c;\u6211\u4eec\u53ef\u4ee5\u667a\u80fd\u5730\u586b\u5145\u56fe\u50cf\u4e2d\u88ab\u906e\u6321\u6216\u635f\u574f\u7684\u90e8\u5206&#xff08;\u4fee\u590d&#xff09;&#xff0c;\u6216\u8005\u5728\u73b0\u6709\u56fe\u50cf\u7684\u753b\u6846\u4e4b\u5916&#xff0c;\u8fdb\u884c\u5bcc\u6709\u60f3\u8c61\u529b\u7684\u3001\u7b26\u5408\u539f\u4f5c\u98ce\u683c\u7684\u6269\u5c55&#xff08;\u6269\u5c55&#xff09;&#xff0c;\u8fd9\u5728\u7167\u7247\u4fee\u590d\u3001\u5f71\u89c6\u540e\u671f\u5236\u4f5c\u7b49\u9886\u57df\u975e\u5e38\u6709\u7528\u3002<\/li>\n<\/ul>\n<h5>10.4.2 \u98ce\u683c\u8fc1\u79fb\u4e0e\u53d8\u6362<\/h5>\n<ul>\n<li>\u827a\u672f\u98ce\u683c\u8fc1\u79fb&#xff1a;\u8fd9\u662f\u751f\u6210\u6a21\u578b\u4e00\u4e2a\u7ecf\u5178\u4e14\u5bcc\u6709\u9b45\u529b\u7684\u5e94\u7528\u3002\u6211\u4eec\u53ef\u4ee5\u63d0\u53d6\u4e00\u5f20\u5185\u5bb9\u56fe\u7247&#xff08;\u5982\u4e00\u5f20\u4e2a\u4eba\u7167\u7247&#xff09;\u7684\u5185\u5bb9\u7279\u5f81&#xff0c;\u548c\u4e00\u5f20\u827a\u672f\u4f5c\u54c1&#xff08;\u5982\u68b5\u9ad8\u7684\u300a\u661f\u7a7a\u300b&#xff09;\u7684\u98ce\u683c\u7279\u5f81&#xff0c;\u7136\u540e\u8ba9\u751f\u6210\u6a21\u578b\u5c06\u4e24\u8005\u7ed3\u5408&#xff0c;\u521b\u9020\u51fa\u4e00\u5f20\u65e2\u4fdd\u7559\u4e86\u7167\u7247\u5185\u5bb9\u3001\u53c8\u5145\u6ee1\u4e86\u300a\u661f\u7a7a\u300b\u7b14\u89e6\u548c\u8272\u5f69\u98ce\u683c\u7684\u5168\u65b0\u827a\u672f\u54c1\u3002<\/li>\n<li>\u57df\u9002\u5e94&#xff08;Domain Adaptation&#xff09;&#xff1a;\u751f\u6210\u6a21\u578b\u53ef\u4ee5\u5728\u4e0d\u540c\u7684\u6570\u636e\u201c\u57df\u201d\u4e4b\u95f4\u5efa\u7acb\u6865\u6881\u3002\u4f8b\u5982&#xff0c;\u5c06\u4e00\u4e2a\u9886\u57df\u7684\u6570\u636e&#xff08;\u5982\u590f\u5929\u7684\u98ce\u666f\u7167&#xff09;&#xff0c;\u5728\u4fdd\u6301\u5176\u6838\u5fc3\u5185\u5bb9\u4e0d\u53d8\u7684\u524d\u63d0\u4e0b&#xff0c;\u8f6c\u6362\u6210\u53e6\u4e00\u4e2a\u9886\u57df&#xff08;\u5982\u51ac\u5929\u7684\u98ce\u666f\u7167&#xff09;&#xff1b;\u6216\u8005\u5c06\u6e38\u620f\u5f15\u64ce\u6e32\u67d3\u51fa\u7684\u865a\u62df\u57ce\u5e02\u573a\u666f&#xff0c;\u8f6c\u6362\u6210\u903c\u771f\u7684\u3001\u7167\u7247\u7ea7\u7684\u73b0\u5b9e\u573a\u666f&#xff0c;\u8fd9\u5bf9\u4e8e\u81ea\u52a8\u9a7e\u9a76\u7b49\u9886\u57df\u7684\u6570\u636e\u96c6\u6269\u5145\u81f3\u5173\u91cd\u8981\u3002<\/li>\n<\/ul>\n<h5>10.4.3 \u6570\u636e\u589e\u5f3a\u4e0e\u534a\u76d1\u7763\u5b66\u4e60<\/h5>\n<ul>\n<li>\u6570\u636e\u589e\u5f3a&#xff08;Data Augmentation&#xff09;&#xff1a;\u5728\u8bb8\u591a\u73b0\u5b9e\u573a\u666f\u4e2d&#xff0c;\u83b7\u53d6\u5927\u91cf\u7684\u6807\u6ce8\u6570\u636e\u662f\u6602\u8d35\u7684\u3002\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u751f\u6210\u6a21\u578b&#xff0c;\u5728\u73b0\u6709\u7684\u5c11\u91cf\u8bad\u7ec3\u6570\u636e\u57fa\u7840\u4e0a&#xff0c;\u521b\u9020\u51fa\u5927\u91cf\u65b0\u7684\u3001\u903c\u771f\u7684\u3001\u4e14\u5e26\u6709\u6b63\u786e\u6807\u7b7e\u7684\u8bad\u7ec3\u6837\u672c\u3002\u8fd9\u4e9b\u751f\u6210\u51fa\u7684\u201c\u5047\u201d\u6570\u636e&#xff0c;\u53ef\u4ee5\u4e0e\u771f\u5b9e\u6570\u636e\u6df7\u5408\u5728\u4e00\u8d77&#xff0c;\u7528\u4e8e\u8bad\u7ec3\u4e0b\u6e38\u7684\u5224\u522b\u5f0f\u6a21\u578b&#xff08;\u5982\u5206\u7c7b\u5668&#xff09;&#xff0c;\u4ece\u800c\u6709\u6548\u63d0\u5347\u5176\u6027\u80fd\u548c\u9c81\u68d2\u6027\u3002<\/li>\n<li>\u5f02\u5e38\u68c0\u6d4b&#xff1a;\u751f\u6210\u6a21\u578b\u5bf9\u4e8e\u5b66\u4e60\u6b63\u5e38\u6570\u636e\u7684\u5206\u5e03\u975e\u5e38\u5728\u884c\u3002\u6211\u4eec\u53ef\u4ee5\u8bad\u7ec3\u4e00\u4e2a\u53ea\u89c1\u8fc7\u201c\u6b63\u5e38\u201d\u6837\u672c&#xff08;\u5982\u5065\u5eb7\u7684\u7ec6\u80de\u56fe\u50cf\u3001\u6b63\u5e38\u7684\u5de5\u4e1a\u96f6\u4ef6&#xff09;\u7684\u751f\u6210\u6a21\u578b&#xff08;\u5982VAE\u6216GAN&#xff09;\u3002\u5f53\u4e00\u4e2a\u201c\u5f02\u5e38\u201d\u6837\u672c&#xff08;\u5982\u764c\u7ec6\u80de\u3001\u6709\u7f3a\u9677\u7684\u96f6\u4ef6&#xff09;\u8f93\u5165\u65f6&#xff0c;\u6a21\u578b\u5c06\u65e0\u6cd5\u5f88\u597d\u5730\u5bf9\u5176\u8fdb\u884c\u91cd\u6784\u6216\u5224\u522b&#xff0c;\u4ece\u800c\u4ea7\u751f\u5de8\u5927\u7684\u91cd\u6784\u8bef\u5dee\u6216\u5f02\u5e38\u7684\u5224\u522b\u5206\u6570\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u6211\u4eec\u53ef\u4ee5\u6709\u6548\u5730\u8bc6\u522b\u51fa\u7f55\u89c1\u7684\u5f02\u5e38\u4e8b\u4ef6\u3002<\/li>\n<\/ul>\n<p>\u5c0f\u7ed3<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u8e0f\u4e0a\u4e86\u4e00\u6bb5\u4ece\u201c\u7406\u89e3\u201d\u5230\u201c\u521b\u9020\u201d\u7684\u5947\u5999\u65c5\u7a0b\u3002\u6211\u4eec\u63a2\u7d22\u4e86\u751f\u6210\u5f0f\u6a21\u578b\u7684\u4e09\u5927\u4e3b\u6d41\u8303\u5f0f&#xff1a;<\/p>\n<ul>\n<li>\u6211\u4eec\u89c1\u8bc1\u4e86\u751f\u6210\u5bf9\u6297\u7f51\u7edc&#xff08;GAN&#xff09;\u4e2d&#xff0c;\u201c\u4f2a\u9020\u8005\u201d\u4e0e\u201c\u9274\u8d4f\u5bb6\u201d\u4e4b\u95f4\u6c38\u6052\u7684\u3001\u76f8\u4e92\u9a71\u52a8\u7684\u8fdb\u5316\u535a\u5f08&#xff0c;\u5b83\u4ee5\u5176\u5353\u8d8a\u7684\u751f\u6210\u903c\u771f\u5ea6&#xff0c;\u5f00\u542f\u4e86\u6df1\u5ea6\u4f2a\u9020\u7684\u65f6\u4ee3\u3002<\/li>\n<li>\u6211\u4eec\u9886\u7565\u4e86\u53d8\u5206\u81ea\u7f16\u7801\u5668&#xff08;VAE&#xff09;\u90a3\u5145\u6ee1\u6982\u7387\u7f8e\u5b66\u7684\u4f18\u96c5&#xff0c;\u5b83\u81f4\u529b\u4e8e\u6784\u5efa\u4e00\u4e2a\u5e73\u6ed1\u3001\u8fde\u7eed\u7684\u6f5c\u5728\u7a7a\u95f4&#xff0c;\u4e3a\u6570\u636e\u7684\u53ef\u63a7\u751f\u6210\u4e0e\u7f16\u8f91\u63d0\u4f9b\u4e86\u575a\u5b9e\u7684\u7406\u8bba\u57fa\u7840\u3002<\/li>\n<li>\u6211\u4eec\u60ca\u53f9\u4e8e\u6269\u6563\u6a21\u578b&#xff08;Diffusion Models&#xff09;\u90a3\u5982\u540c\u9006\u8f6c\u65f6\u95f4\u822c\u7684\u827a\u672f&#xff0c;\u5b83\u901a\u8fc7\u4ece\u7eaf\u7cb9\u566a\u58f0\u4e2d\u9010\u6b65\u53bb\u566a\u3001\u96d5\u523b\u6770\u4f5c\u7684\u65b9\u5f0f&#xff0c;\u5c06\u751f\u6210\u5185\u5bb9\u7684\u8d28\u91cf\u548c\u53ef\u63a7\u6027\u63a8\u5411\u4e86\u524d\u6240\u672a\u6709\u7684\u65b0\u9ad8\u5ea6\u3002<\/li>\n<\/ul>\n<p>\u6700\u540e&#xff0c;\u6211\u4eec\u5de1\u793c\u4e86\u8fd9\u4e9b\u5f3a\u5927\u7684\u751f\u6210\u5de5\u5177\u5728\u56fe\u50cf\u751f\u6210\u3001\u98ce\u683c\u8fc1\u79fb\u3001\u6570\u636e\u589e\u5f3a\u7b49\u9886\u57df\u7684\u5e7f\u6cdb\u5e94\u7528&#xff0c;\u4eb2\u8eab\u611f\u53d7\u4e86\u5f53\u673a\u5668\u5f00\u59cb\u201c\u505a\u68a6\u201d\u65f6&#xff0c;\u6240\u91ca\u653e\u51fa\u7684\u5de8\u5927\u521b\u9020\u529b\u3002\u638c\u63e1\u4e86\u751f\u6210\u5f0f\u6a21\u578b&#xff0c;\u8bfb\u8005\u4e0d\u4ec5\u638c\u63e1\u4e86\u5f53\u524dAI\u9886\u57df\u6700\u524d\u6cbf\u7684\u6280\u672f&#xff0c;\u66f4\u83b7\u5f97\u4e86\u4e00\u628a\u5f00\u542f\u672a\u6765\u5185\u5bb9\u521b\u4f5c\u65b0\u7eaa\u5143\u7684\u94a5\u5319\u3002\u5728\u672c\u4e66\u7684\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u5c55\u671b\u8fd9\u4e00\u5207\u6280\u672f\u878d\u5408\u540e&#xff0c;\u4eba\u5de5\u667a\u80fd\u66f4\u5e7f\u9614\u7684\u672a\u6765\u3002<\/p>\n<hr \/>\n<h2>\u7b2c\u56db\u90e8\u5206&#xff1a;\u5b9e\u6218\u7bc7 \u2014\u2014 \u4ece\u7406\u8bba\u5230\u4ef7\u503c\u7684\u8f6c\u5316<\/h2>\n<hr \/>\n<h3>\u7b2c\u5341\u4e00\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u8ba1\u7b97\u673a\u89c6\u89c9<\/h3>\n<ul>\n<li>11.1 \u56fe\u50cf\u5206\u7c7b&#xff1a;\u6784\u5efa\u4e00\u4e2a\u5783\u573e\u5206\u7c7b\u7cfb\u7edf\u3002<\/li>\n<li>11.2 \u76ee\u6807\u68c0\u6d4b&#xff1a;\u5b9e\u73b0\u4e00\u4e2a\u5b9e\u65f6\u4eba\u8138\u6216\u8f66\u8f86\u68c0\u6d4b\u5668\u3002<\/li>\n<li>11.3 \u56fe\u50cf\u98ce\u683c\u8fc1\u79fb&#xff1a;\u5c06\u7167\u7247\u53d8\u6210\u68b5\u9ad8\u98ce\u683c\u7684\u6cb9\u753b\u3002<\/li>\n<\/ul>\n<p>\u4ece\u201c\u770b\u61c2\u201d\u5230\u201c\u770b\u89c1\u201d\u7684\u98de\u8dc3<\/p>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u5728\u8fc7\u53bb\u7684\u7bc7\u7ae0\u91cc&#xff0c;\u6211\u4eec\u4e00\u540c\u8dcb\u5c71\u6d89\u6c34&#xff0c;\u7a7f\u8d8a\u4e86\u6df1\u5ea6\u5b66\u4e60\u7684\u7406\u8bba\u4e1b\u6797\u3002\u6211\u4eec\u66fe\u5728\u7b2c\u4e03\u7ae0&#xff0c;\u501f\u52a9\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u7684\u201c\u6167\u773c\u201d&#xff0c;\u8ba9\u673a\u5668\u5b66\u4f1a\u4e86\u8bc6\u522b\u56fe\u50cf\u4e2d\u7684\u201c\u540d\u76f8\u201d\u2014\u2014\u8fd9\u662f\u732b&#xff0c;\u90a3\u662f\u72d7\u3002\u6211\u4eec\u4e5f\u5728\u7b2c\u5341\u7ae0&#xff0c;\u901a\u8fc7\u751f\u6210\u6a21\u578b\u7684\u201c\u795e\u7b14\u201d&#xff0c;\u8ba9\u673a\u5668\u62e5\u6709\u4e86\u65e0\u4e2d\u751f\u6709\u7684\u521b\u9020\u529b\u3002\u6211\u4eec\u5df2\u7ecf\u638c\u63e1\u4e86\u8ba9\u673a\u5668\u201c\u770b\u61c2\u201d\u50cf\u7d20\u6570\u636e\u7684\u57fa\u672c\u6cd5\u95e8\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u771f\u6b63\u7684\u667a\u6167&#xff0c;\u4e0d\u6b62\u4e8e\u201c\u770b\u61c2\u201d&#xff0c;\u66f4\u5728\u4e8e\u201c\u770b\u89c1\u201d\u3002<\/p>\n<p>\u201c\u770b\u61c2\u201d&#xff0c;\u662f\u5206\u8fa8\u4e0e\u8bc6\u522b&#xff0c;\u662f\u903b\u8f91\u7684\u5224\u65ad&#xff0c;\u662f\u5bf9\u4e16\u754c\u8868\u8c61\u7684\u6479\u5199\u3002\u800c\u201c\u770b\u89c1\u201d&#xff0c;\u662f\u6d1e\u5bdf\u4e0e\u7406\u89e3&#xff0c;\u662f\u8054\u7cfb\u7684\u5efa\u7acb&#xff0c;\u662f\u5bf9\u4e8b\u7269\u672c\u8d28\u4e0e\u5185\u5728\u89c4\u5f8b\u7684\u5f7b\u609f\u3002\u6b63\u5982\u7985\u5b97\u6240\u8a00&#xff0c;\u201c\u89c1\u5c71\u662f\u5c71&#xff0c;\u89c1\u6c34\u662f\u6c34\u201d\u662f\u521d\u5883&#xff1b;\u201c\u89c1\u5c71\u4e0d\u662f\u5c71&#xff0c;\u89c1\u6c34\u4e0d\u662f\u6c34\u201d\u662f\u8fc7\u7a0b&#xff1b;\u6700\u7ec8\u201c\u89c1\u5c71\u8fd8\u662f\u5c71&#xff0c;\u89c1\u6c34\u8fd8\u662f\u6c34\u201d&#xff0c;\u5219\u662f\u5f7b\u609f\u540e\u7684\u56de\u5f52\u3002\u6211\u4eec\u7684\u5b66\u4e60\u4e4b\u65c5&#xff0c;\u4ea6\u590d\u5982\u662f\u3002<\/p>\n<p>\u672c\u7ae0&#xff0c;\u4fbf\u662f\u6211\u4eec\u4ece\u201c\u770b\u61c2\u201d\u8fc8\u5411\u201c\u770b\u89c1\u201d\u7684\u4fee\u884c\u9053\u573a\u3002\u6211\u4eec\u5c06\u8d70\u51fa\u7406\u8bba\u7684\u9759\u4fee\u5ba4&#xff0c;\u6b65\u5165\u771f\u5b9e\u4e16\u754c\u7684\u70df\u706b\u4eba\u95f4\u3002\u9879\u76ee\u5b9e\u6218&#xff0c;\u5b83\u5e76\u975e\u4ee3\u7801\u7684\u7b80\u5355\u5806\u780c&#xff0c;\u800c\u662f\u8fde\u63a5\u62bd\u8c61\u6a21\u578b\u4e0e\u9c9c\u6d3b\u73b0\u5b9e\u7684\u6865\u6881\u3002\u5728\u8fd9\u91cc&#xff0c;\u51b0\u51b7\u7684\u7b97\u6cd5\u5c06\u62e5\u6709\u6e29\u5ea6&#xff0c;\u53bb\u89e3\u51b3\u73af\u4fdd\u7684\u96be\u9898&#xff1b;\u5728\u8fd9\u91cc&#xff0c;\u4e25\u8c28\u7684\u6570\u5b66\u5c06\u9082\u9005\u827a\u672f&#xff0c;\u53bb\u6325\u6d12\u68b5\u9ad8\u7684\u6fc0\u60c5\u3002<\/p>\n<p>\u8fd9\u4e0d\u4ec5\u662f\u6280\u672f\u7684\u5e94\u7528&#xff0c;\u66f4\u662f\u4e00\u573a\u6df1\u523b\u7684\u201c\u77e5\u884c\u5408\u4e00\u201d\u3002\u6bcf\u4e00\u6b21\u6570\u636e\u6e05\u6d17&#xff0c;\u90fd\u662f\u5728\u78e8\u783a\u6211\u4eec\u7684\u8010\u5fc3\u4e0e\u7ec6\u81f4&#xff1b;\u6bcf\u4e00\u6b21\u6a21\u578b\u8c03\u8bd5&#xff0c;\u90fd\u662f\u5728\u8003\u9a8c\u6211\u4eec\u7684\u6d1e\u5bdf\u4e0e\u53d8\u901a&#xff1b;\u6bcf\u4e00\u6b21\u6210\u529f\u8fd0\u884c&#xff0c;\u90fd\u662f\u5bf9\u6211\u4eec\u6240\u5b66\u77e5\u8bc6\u6700\u771f\u5207\u7684\u56de\u54cd\u3002\u5728\u8fd9\u91cc&#xff0c;\u7406\u8bba\u662f\u7f57\u76d8&#xff0c;\u5b9e\u8df5\u662f\u822a\u8239&#xff0c;\u800c\u6211\u4eec&#xff0c;\u662f\u9a76\u5411\u4ef7\u503c\u5f7c\u5cb8\u7684\u822a\u6d77\u5bb6\u3002<\/p>\n<p>\u51c6\u5907\u597d\u4e86\u5417&#xff1f;\u8ba9\u6211\u4eec\u626c\u5e06\u8d77\u822a&#xff0c;\u5f00\u542f\u8fd9\u573a\u4ee5\u667a\u6167\u4e4b\u773c\u6d1e\u89c1\u4e07\u7269\u7684\u5b9e\u6218\u4e4b\u65c5\u3002\u6211\u4eec\u5c06\u4eb2\u624b\u6784\u5efa\u4e09\u4e2a\u9879\u76ee&#xff0c;\u5b83\u4eec\u5206\u522b\u4ee3\u8868\u4e86\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u4e09\u4e2a\u6838\u5fc3\u4e14\u8ff7\u4eba\u7684\u65b9\u5411&#xff1a;\u56fe\u50cf\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b\u4e0e\u98ce\u683c\u8fc1\u79fb\u3002\u8ba9\u6211\u4eec\u5728\u521b\u9020\u4e2d\u5b66\u4e60&#xff0c;\u5728\u5b9e\u8df5\u4e2d\u5347\u534e\u3002<\/p>\n<hr \/>\n<h4>11.1 \u56fe\u50cf\u5206\u7c7b&#xff1a;\u6784\u5efa\u4e00\u4e2a\u5783\u573e\u5206\u7c7b\u7cfb\u7edf \u2014\u2014 \u4e3a\u4e16\u754c\u6ce8\u5165\u4e00\u4efd\u201c\u51c0\u201d<\/h4>\n<p>\u56fe\u50cf\u5206\u7c7b&#xff0c;\u53ef\u4ee5\u8bf4\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u7684\u201cHello, World!\u201d\u3002\u5b83\u662f\u6211\u4eec\u5b66\u4e60CNN\u65f6\u6700\u5148\u63a5\u89e6\u7684\u4efb\u52a1&#xff0c;\u4e5f\u662f\u65e0\u6570\u66f4\u590d\u6742\u89c6\u89c9\u4efb\u52a1&#xff08;\u5982\u76ee\u6807\u68c0\u6d4b&#xff09;\u7684\u57fa\u77f3\u3002\u4f46\u7b80\u5355&#xff0c;\u7edd\u4e0d\u610f\u5473\u7740\u4e0d\u91cd\u8981\u3002\u6070\u6070\u76f8\u53cd&#xff0c;\u5c06\u6700\u57fa\u7840\u7684\u6280\u672f\u5e94\u7528\u4e8e\u6700\u8feb\u5207\u7684\u73b0\u5b9e\u95ee\u9898&#xff0c;\u6b63\u662f\u6280\u672f\u4ef7\u503c\u7684\u6700\u597d\u4f53\u73b0\u3002<\/p>\n<h5>11.1.1 \u9879\u76ee\u7f18\u8d77\u4e0e\u4ef7\u503c&#xff1a;\u5f53AI\u9047\u89c1\u73af\u4fdd<\/h5>\n<ul>\n<li>\n<p>\u80cc\u666f \u201c\u7eff\u6c34\u9752\u5c71\u5c31\u662f\u91d1\u5c71\u94f6\u5c71\u201d\u3002\u73af\u5883\u4fdd\u62a4&#xff0c;\u5df2\u6210\u4e3a\u6211\u4eec\u8fd9\u4e2a\u65f6\u4ee3\u6700\u5b8f\u5927\u7684\u53d9\u4e8b\u4e4b\u4e00\u3002\u800c\u5783\u573e\u5206\u7c7b&#xff0c;\u662f\u8fd9\u4e2a\u5b8f\u5927\u53d9\u4e8b\u4e2d\u4e0e\u6211\u4eec\u6bcf\u4e2a\u4eba\u606f\u606f\u76f8\u5173\u7684\u4e00\u73af\u3002\u5b83\u770b\u4f3c\u7b80\u5355&#xff0c;\u5b9e\u5219\u6311\u6218\u91cd\u91cd&#xff1a;\u5206\u7c7b\u6807\u51c6\u7e41\u591a\u3001\u6c11\u4f17\u8ba4\u77e5\u4e0d\u8db3\u3001\u4eba\u5de5\u5206\u62e3\u6210\u672c\u9ad8\u6602\u4e14\u6548\u7387\u4f4e\u4e0b\u3002 \u6b64\u65f6&#xff0c;\u6211\u4eec\u4e0d\u7981\u4f1a\u60f3&#xff1a;\u80fd\u5426\u8ba9\u673a\u5668\u4ee3\u66ff\u4eba\u773c&#xff0c;\u53bb\u5b8c\u6210\u8fd9\u9879\u91cd\u590d\u800c\u53c8\u91cd\u8981\u7684\u5de5\u4f5c&#xff1f;\u80fd\u5426\u5728\u5783\u573e\u6295\u653e\u53e3\u3001\u5206\u62e3\u6d41\u6c34\u7ebf\u4e0a&#xff0c;\u5b89\u88c5\u4e00\u53cc\u4e0d\u77e5\u75b2\u5026\u3001\u51c6\u786e\u65e0\u8bef\u7684\u201c\u773c\u775b\u201d&#xff1f;\u8fd9\u4fbf\u662f\u6211\u4eec\u8fd9\u4e2a\u9879\u76ee\u7684\u521d\u5fc3\u2014\u2014\u5f53AI\u7684\u667a\u6167&#xff0c;\u9047\u89c1\u73af\u4fdd\u7684\u8feb\u5207&#xff0c;\u6211\u4eec\u6709\u673a\u4f1a\u4e3a\u8fd9\u4e2a\u4e16\u754c\u6ce8\u5165\u4e00\u4efd\u771f\u6b63\u7684\u201c\u6d01\u51c0\u201d\u4e0e\u201c\u6548\u7387\u201d\u3002<\/p>\n<\/li>\n<li>\n<p>\u76ee\u6807 \u6211\u4eec\u7684\u76ee\u6807\u6e05\u6670\u800c\u5177\u4f53&#xff1a;\u6784\u5efa\u4e00\u4e2a\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u7684\u56fe\u50cf\u5206\u7c7b\u6a21\u578b&#xff0c;\u8be5\u6a21\u578b\u80fd\u591f\u63a5\u6536\u4e00\u5f20\u5783\u573e\u56fe\u7247&#xff0c;\u5e76\u81ea\u52a8\u5224\u65ad\u5176\u6240\u5c5e\u7c7b\u522b&#xff08;\u4f8b\u5982&#xff1a;\u53a8\u4f59\u5783\u573e\u3001\u53ef\u56de\u6536\u7269\u3001\u6709\u5bb3\u5783\u573e\u3001\u5176\u4ed6\u5783\u573e&#xff09;\u3002\u6211\u4eec\u5c06\u8d70\u8fc7\u4e00\u4e2aAI\u9879\u76ee\u7684\u5b8c\u6574\u751f\u547d\u5468\u671f&#xff0c;\u4ece\u6570\u636e\u51c6\u5907\u5230\u6a21\u578b\u90e8\u7f72&#xff0c;\u4f53\u9a8c\u4e00\u6b21\u5b8c\u6574\u7684\u201c\u4ef7\u503c\u521b\u9020\u201d\u4e4b\u65c5\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>11.1.2 \u6570\u636e\u4e4b\u9053&#xff1a;\u4ece\u6536\u96c6\u5230\u51c6\u5907<\/h5>\n<p>\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u5b87\u5b99\u4e2d&#xff0c;\u6570\u636e\u662f\u5f15\u71c3\u4e00\u5207\u7684\u201c\u661f\u5c18\u201d\u3002\u6ca1\u6709\u6570\u636e&#xff0c;\u518d\u5f3a\u5927\u7684\u6a21\u578b\u4e5f\u53ea\u662f\u4e00\u4e2a\u7a7a\u58f3\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u7684\u7b2c\u4e00\u6b65&#xff0c;\u4e5f\u662f\u81f3\u5173\u91cd\u8981\u7684\u4e00\u6b65&#xff0c;\u4fbf\u662f\u6570\u636e\u7684\u83b7\u53d6\u4e0e\u5904\u7406\u3002<\/p>\n<ul>\n<li>\n<p>\u6570\u636e\u96c6\u7684\u83b7\u53d6 \u5bf9\u4e8e\u521d\u5b66\u8005\u800c\u8a00&#xff0c;\u6700\u4fbf\u6377\u7684\u65b9\u5f0f\u662f\u4f7f\u7528\u516c\u5f00\u6570\u636e\u96c6\u3002TrashNet\u662f\u4e00\u4e2a\u5e7f\u53d7\u6b22\u8fce\u7684\u5783\u573e\u5206\u7c7b\u6570\u636e\u96c6&#xff0c;\u5b83\u5305\u542b\u4e86\u73bb\u7483\u3001\u7eb8\u5f20\u3001\u7eb8\u677f\u3001\u5851\u6599\u3001\u91d1\u5c5e\u548c\u4e00\u822c\u5783\u573e\u7b49\u7c7b\u522b\u3002\u6b64\u5916&#xff0c;\u7f51\u7edc\u4e0a\u8fd8\u6709\u8bb8\u591a\u7531\u7814\u7a76\u8005\u548c\u7231\u597d\u8005\u8d21\u732e\u7684\u3001\u89c4\u6a21\u66f4\u5927\u3001\u7c7b\u522b\u66f4\u4e30\u5bcc\u7684\u5783\u573e\u56fe\u50cf\u6570\u636e\u96c6\u3002 \u5f53\u7136&#xff0c;\u66f4\u5177\u6311\u6218\u6027\u4e5f\u66f4\u6709\u8da3\u7684\u65b9\u5f0f&#xff0c;\u662f\u81ea\u5efa\u6570\u636e\u96c6\u3002\u60a8\u53ef\u4ee5\u5229\u7528Python\u7684\u7f51\u7edc\u722c\u866b\u5e93&#xff08;\u5982Scrapy\u6216BeautifulSoup&#xff09;\u4ece\u641c\u7d22\u5f15\u64ce\u6216\u56fe\u5e93\u7f51\u7ad9\u4e0a\u6293\u53d6\u7279\u5b9a\u7c7b\u522b\u7684\u5783\u573e\u56fe\u7247\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u672c\u8eab&#xff0c;\u5c31\u662f\u4e00\u6b21\u5b9d\u8d35\u7684\u5de5\u7a0b\u5b9e\u8df5\u3002<\/p>\n<\/li>\n<li>\n<p>\u6570\u636e\u63a2\u7d22\u4e0e\u53ef\u89c6\u5316&#xff08;EDA&#xff09; \u62ff\u5230\u6570\u636e\u540e&#xff0c;\u5207\u5fcc\u76f4\u63a5\u6295\u5165\u8bad\u7ec3\u3002\u6211\u4eec\u5e94\u5148\u50cf\u4e00\u4f4d\u4fa6\u63a2\u4e00\u6837&#xff0c;\u4ed4\u7ec6\u52d8\u5bdf\u8fd9\u4efd\u6570\u636e\u3002<\/p>\n<ul>\n<li>\u6211\u4eec\u6709\u591a\u5c11\u6570\u636e&#xff1f;\u00a0\u6bcf\u4e2a\u7c7b\u522b\u5404\u6709\u591a\u5c11\u5f20\u56fe\u7247&#xff1f;\u662f\u5426\u5b58\u5728\u4e25\u91cd\u7684\u7c7b\u522b\u4e0d\u5747\u8861\u95ee\u9898&#xff08;\u4f8b\u5982&#xff0c;\u67d0\u4e2a\u7c7b\u522b\u7684\u56fe\u7247\u6570\u91cf\u8fdc\u5c11\u4e8e\u5176\u4ed6\u7c7b\u522b&#xff09;&#xff1f;\u7c7b\u522b\u4e0d\u5747\u8861\u4f1a\u4f7f\u6a21\u578b\u504f\u5411\u4e8e\u5b66\u4e60\u6837\u672c\u91cf\u5927\u7684\u7c7b\u522b&#xff0c;\u5bfc\u81f4\u5bf9\u5c0f\u6837\u672c\u7c7b\u522b\u7684\u8bc6\u522b\u80fd\u529b\u5dee\u3002<\/li>\n<li>\u6570\u636e\u8d28\u91cf\u5982\u4f55&#xff1f;\u00a0\u56fe\u7247\u662f\u5426\u6e05\u6670&#xff1f;\u662f\u5426\u5b58\u5728\u5927\u91cf\u6807\u6ce8\u9519\u8bef\u6216\u65e0\u5173\u7684\u56fe\u7247&#xff1f; \u901a\u8fc7\u7f16\u5199\u7b80\u5355\u7684\u811a\u672c&#xff0c;\u7edf\u8ba1\u5404\u7c7b\u522b\u6570\u91cf\u5e76\u7ed8\u5236\u6210\u67f1\u72b6\u56fe&#xff0c;\u518d\u968f\u673a\u62bd\u53d6\u4e00\u4e9b\u6837\u672c\u8fdb\u884c\u4eba\u5de5\u67e5\u770b&#xff0c;\u662f\u8fdb\u884c\u6570\u636e\u63a2\u7d22\u7684\u5fc5\u8981\u6b65\u9aa4\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6570\u636e\u9884\u5904\u7406\u4e0e\u589e\u5f3a \u539f\u59cb\u6570\u636e\u5982\u540c\u672a\u7ecf\u96d5\u7422\u7684\u749e\u7389&#xff0c;\u9700\u8981\u7cbe\u5fc3\u6253\u78e8\u624d\u80fd\u7115\u53d1\u5149\u5f69\u3002<\/p>\n<ul>\n<li>\u56fe\u50cf\u5c3a\u5bf8\u5f52\u4e00\u5316&#xff1a;\u7531\u4e8e\u6570\u636e\u96c6\u4e2d\u7684\u56fe\u7247\u5c3a\u5bf8\u53ef\u80fd\u5343\u5dee\u4e07\u522b&#xff0c;\u800c\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u9700\u8981\u56fa\u5b9a\u7684\u5927\u5c0f&#xff08;\u4f8b\u5982224&#215;224\u50cf\u7d20&#xff09;&#xff0c;\u56e0\u6b64\u6211\u4eec\u9700\u8981\u5c06\u6240\u6709\u56fe\u7247\u7edf\u4e00\u7f29\u653e\u5230\u76ee\u6807\u5c3a\u5bf8\u3002<\/li>\n<li>\u6570\u636e\u589e\u5f3a&#xff08;Data Augmentation&#xff09;&#xff1a;\u8fd9\u662f\u5e94\u5bf9\u6570\u636e\u91cf\u4e0d\u8db3\u548c\u9632\u6b62\u8fc7\u62df\u5408\u7684\u5f3a\u5927\u6cd5\u5b9d\u3002\u6211\u4eec\u53ef\u4ee5\u5728\u5c06\u56fe\u7247\u9001\u5165\u7f51\u7edc\u8bad\u7ec3\u4e4b\u524d&#xff0c;\u5bf9\u5b83\u4eec\u8fdb\u884c\u4e00\u7cfb\u5217\u968f\u673a\u7684\u53d8\u6362&#xff0c;\u751f\u6210\u201c\u65b0\u201d\u7684\u8bad\u7ec3\u6837\u672c\u3002\u8fd9\u5c31\u50cf\u8ba9\u6a21\u578b\u4ece\u4e0d\u540c\u7684\u89d2\u5ea6\u3001\u5728\u4e0d\u540c\u7684\u5149\u7167\u4e0b\u89c2\u5bdf\u540c\u4e00\u4e2a\u7269\u4f53&#xff0c;\u4ece\u800c\u5b66\u4e60\u5230\u66f4\u672c\u8d28\u3001\u66f4\u9c81\u68d2\u7684\u7279\u5f81\u3002\n<ul>\n<li>\u5e38\u7528\u7684\u589e\u5f3a\u624b\u6bb5\u5305\u62ec&#xff1a;\n<ul>\n<li>\u968f\u673a\u6c34\u5e73\/\u5782\u76f4\u7ffb\u8f6c&#xff1a;\u4e00\u5f20\u201c\u6b63\u7740\u653e\u201d\u7684\u6613\u62c9\u7f50\u56fe\u7247\u548c\u201c\u5012\u7740\u653e\u201d\u7684\u6613- \u62c9\u7f50\u56fe\u7247&#xff0c;\u90fd\u5e94\u88ab\u8bc6\u522b\u4e3a\u53ef\u56de\u6536\u7269\u3002<\/li>\n<li>\u968f\u673a\u65cb\u8f6c&#xff1a;\u5728\u4e00\u5b9a\u89d2\u5ea6\u8303\u56f4\u5185\u968f\u673a\u65cb\u8f6c\u56fe\u7247\u3002<\/li>\n<li>\u968f\u673a\u88c1\u526a\u4e0e\u7f29\u653e&#xff1a;\u6a21\u62df\u7269\u4f53\u5728\u753b\u9762\u4e2d\u8fdc\u8fd1\u4e0d\u540c\u7684\u60c5\u51b5\u3002<\/li>\n<li>\u8272\u5f69\u6296\u52a8&#xff1a;\u968f\u673a\u6539\u53d8\u56fe\u50cf\u7684\u4eae\u5ea6\u3001\u5bf9\u6bd4\u5ea6\u3001\u9971\u548c\u5ea6&#xff0c;\u6a21\u62df\u4e0d\u540c\u5149\u7167\u6761\u4ef6\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u5e78\u8fd0\u7684\u662f&#xff0c;tf.keras.preprocessing.image.ImageDataGenerator\u6216tf.dataAPI\u7b49\u5de5\u5177&#xff0c;\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u6781\u5176\u65b9\u4fbf\u7684\u6570\u636e\u52a0\u8f7d\u4e0e\u5b9e\u65f6\u589e\u5f3a\u529f\u80fd\u3002<\/p>\n<\/li>\n<\/ul>\n<p>import tensorflow as tf<\/p>\n<p># \u5047\u8bbe\u6211\u4eec\u7684\u6570\u636e\u5df2\u7ecf\u6309\u7c7b\u522b\u5b58\u653e\u5728\u4e0d\u540c\u6587\u4ef6\u5939\u4e2d<br \/>\n# data_dir\/<br \/>\n#   \u251c\u2500\u2500 cardboard\/<br \/>\n#   \u251c\u2500\u2500 glass\/<br \/>\n#   \u2514\u2500\u2500 &#8230;<\/p>\n<p>IMG_SIZE &#061; (224, 224)<br \/>\nBATCH_SIZE &#061; 32<\/p>\n<p># \u4f7f\u7528ImageDataGenerator\u8fdb\u884c\u6570\u636e\u52a0\u8f7d\u548c\u589e\u5f3a<br \/>\ntrain_datagen &#061; tf.keras.preprocessing.image.ImageDataGenerator(<br \/>\n    rescale&#061;1.\/255,             # \u5c06\u50cf\u7d20\u503c\u5f52\u4e00\u5316\u5230[0, 1]<br \/>\n    rotation_range&#061;40,          # \u968f\u673a\u65cb\u8f6c\u89d2\u5ea6\u8303\u56f4<br \/>\n    width_shift_range&#061;0.2,      # \u968f\u673a\u6c34\u5e73\u5e73\u79fb\u8303\u56f4<br \/>\n    height_shift_range&#061;0.2,     # \u968f\u673a\u5782\u76f4\u5e73\u79fb\u8303\u56f4<br \/>\n    shear_range&#061;0.2,            # \u968f\u673a\u9519\u5207\u53d8\u6362<br \/>\n    zoom_range&#061;0.2,             # \u968f\u673a\u7f29\u653e\u8303\u56f4<br \/>\n    horizontal_flip&#061;True,       # \u968f\u673a\u6c34\u5e73\u7ffb\u8f6c<br \/>\n    fill_mode&#061;&#039;nearest&#039;,        # \u586b\u5145\u65b0\u521b\u5efa\u50cf\u7d20\u7684\u65b9\u5f0f<br \/>\n    validation_split&#061;0.2        # \u5212\u520620%\u7684\u6570\u636e\u4f5c\u4e3a\u9a8c\u8bc1\u96c6<br \/>\n)<\/p>\n<p>train_generator &#061; train_datagen.flow_from_directory(<br \/>\n    &#039;path\/to\/your\/data_dir&#039;,<br \/>\n    target_size&#061;IMG_SIZE,<br \/>\n    batch_size&#061;BATCH_SIZE,<br \/>\n    class_mode&#061;&#039;categorical&#039;,<br \/>\n    subset&#061;&#039;training&#039;           # \u8bbe\u7f6e\u4e3a\u8bad\u7ec3\u96c6<br \/>\n)<\/p>\n<p>validation_generator &#061; train_datagen.flow_from_directory(<br \/>\n    &#039;path\/to\/your\/data_dir&#039;,<br \/>\n    target_size&#061;IMG_SIZE,<br \/>\n    batch_size&#061;BATCH_SIZE,<br \/>\n    class_mode&#061;&#039;categorical&#039;,<br \/>\n    subset&#061;&#039;validation&#039;         # \u8bbe\u7f6e\u4e3a\u9a8c\u8bc1\u96c6<br \/>\n)<\/p>\n<p>\u8fd9\u6bb5\u4ee3\u7801\u4e0d\u4ec5\u4f18\u96c5\u5730\u89e3\u51b3\u4e86\u6570\u636e\u52a0\u8f7d\u95ee\u9898&#xff0c;\u66f4\u5728\u65e0\u5f62\u4e2d\u6781\u5927\u5730\u6269\u5145\u4e86\u6211\u4eec\u7684\u6570\u636e\u96c6&#xff0c;\u4e3a\u540e\u7eed\u8bad\u7ec3\u4e00\u4e2a\u5f3a\u5927\u7684\u6a21\u578b\u5960\u5b9a\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002\u6570\u636e\u4e4b\u9053&#xff0c;\u5728\u4e8e\u656c\u754f\u4e0e\u7cbe\u5fae\u3002\u73b0\u5728&#xff0c;\u661f\u5c18\u5df2\u5907&#xff0c;\u662f\u65f6\u5019\u70b9\u71c3\u6a21\u578b\u7684\u706b\u7130\u4e86\u3002<\/p>\n<h5>11.1.3 \u6a21\u578b\u6784\u5efa&#xff1a;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a<\/h5>\n<p>\u9762\u5bf9\u4e00\u4e2a\u56fe\u50cf\u5206\u7c7b\u4efb\u52a1&#xff0c;\u6211\u4eec\u6709\u4e24\u4e2a\u9009\u62e9&#xff1a;\u4e00\u662f\u201c\u767d\u624b\u8d77\u5bb6\u201d&#xff0c;\u4ece\u96f6\u5f00\u59cb\u8bbe\u8ba1\u5e76\u8bad\u7ec3\u4e00\u4e2a\u5168\u65b0\u7684CNN\u67b6\u6784&#xff1b;\u4e8c\u662f\u201c\u501f\u529b\u6253\u529b\u201d&#xff0c;\u91c7\u7528\u8fc1\u79fb\u5b66\u4e60&#xff08;Transfer Learning&#xff09;\u3002<\/p>\n<ul>\n<li>\n<p>\u4e3a\u4f55\u9009\u62e9\u8fc1\u79fb\u5b66\u4e60 \u4ece\u5934\u8bad\u7ec3\u4e00\u4e2a\u6df1\u5ea6CNN\u6a21\u578b&#xff0c;\u662f\u4e00\u9879\u5de8\u5927\u7684\u6311\u6218\u3002\u5b83\u4e0d\u4ec5\u9700\u8981\u6d77\u91cf\u7684\u6570\u636e&#xff08;\u901a\u5e38\u662f\u767e\u4e07\u7ea7\u522b&#xff09;&#xff0c;\u8fd8\u9700\u8981\u5f3a\u5927\u7684\u8ba1\u7b97\u8d44\u6e90\u548c\u6f2b\u957f\u7684\u8bad\u7ec3\u65f6\u95f4\u3002\u8fd9\u597d\u6bd4\u8981\u6211\u4eec\u81ea\u5df1\u53bb\u63a8\u6f14\u548c\u8bc1\u660e\u725b\u987f\u4e09\u5b9a\u5f8b&#xff0c;\u867d\u7136\u7406\u8bba\u4e0a\u53ef\u884c&#xff0c;\u4f46\u65e2\u4e0d\u9ad8\u6548&#xff0c;\u4e5f\u65e0\u5fc5\u8981\u3002<\/p>\n<p>\u8fc1\u79fb\u5b66\u4e60\u5219\u63d0\u4f9b\u4e86\u4e00\u6761\u5145\u6ee1\u667a\u6167\u7684\u6377\u5f84\u3002\u5b83\u7684\u6838\u5fc3\u601d\u60f3\u662f&#xff1a;\u4e00\u4e2a\u5728\u8d85\u5927\u89c4\u6a21\u6570\u636e\u96c6&#xff08;\u5982ImageNet&#xff0c;\u5305\u542b1400\u591a\u4e07\u5f20\u56fe\u7247&#xff0c;1000\u591a\u4e2a\u7c7b\u522b&#xff09;\u4e0a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5df2\u7ecf\u5b66\u4e60\u5230\u4e86\u5173\u4e8e\u8fd9\u4e2a\u4e16\u754c\u975e\u5e38\u901a\u7528\u548c\u5e95\u5c42\u7684\u89c6\u89c9\u77e5\u8bc6\u2014\u2014\u6bd4\u5982\u8fb9\u7f18\u3001\u7eb9\u7406\u3001\u5f62\u72b6\u3001\u989c\u8272\u7b49\u3002\u8fd9\u4e9b\u77e5\u8bc6&#xff0c;\u5bf9\u4e8e\u6211\u4eec\u8bc6\u522b\u5783\u573e\u540c\u6837\u662f\u9002\u7528\u7684\u3002<\/p>\n<p>\u56e0\u6b64&#xff0c;\u6211\u4eec\u53ef\u4ee5\u201c\u501f\u7528\u201d\u8fd9\u4e2a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5c06\u5176\u5f3a\u5927\u7684\u7279\u5f81\u63d0\u53d6\u80fd\u529b\u201c\u8fc1\u79fb\u201d\u5230\u6211\u4eec\u7684\u65b0\u4efb\u52a1\u4e0a\u3002\u6211\u4eec\u7ad9\u5728\u4e86\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u53ef\u4ee5\u7528\u66f4\u5c11\u7684\u6570\u636e\u3001\u66f4\u77ed\u7684\u65f6\u95f4&#xff0c;\u8fbe\u5230\u6bd4\u4ece\u96f6\u8bad\u7ec3\u597d\u5f97\u591a\u7684\u6548\u679c\u3002\u8fd9\u662f\u4e00\u79cd\u201c\u4ed6\u5c71\u4e4b\u77f3&#xff0c;\u53ef\u4ee5\u653b\u7389\u201d\u7684\u667a\u6167\u3002<\/p>\n<\/li>\n<li>\n<p>\u9009\u62e9\u9884\u8bad\u7ec3\u6a21\u578b Keras\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u5e93\u3002\u9009\u62e9\u54ea\u4e00\u4e2a&#xff0c;\u901a\u5e38\u662f\u5728\u6a21\u578b\u5927\u5c0f&#xff08;\u53c2\u6570\u91cf&#xff09;\u3001\u8ba1\u7b97\u901f\u5ea6\u548c\u6027\u80fd\u4e4b\u95f4\u505a\u6743\u8861\u3002<\/p>\n<ul>\n<li>ResNet (Residual Network)&#xff1a;\u7ecf\u5178\u4e2d\u7684\u7ecf\u5178&#xff0c;\u901a\u8fc7\u201c\u6b8b\u5dee\u8fde\u63a5\u201d\u89e3\u51b3\u4e86\u6df1\u5ea6\u7f51\u7edc\u7684\u9000\u5316\u95ee\u9898&#xff0c;\u6027\u80fd\u5f3a\u5927\u4e14\u7a33\u5b9a\u3002ResNet50\u662f\u4e00\u4e2a\u975e\u5e38\u5747\u8861\u7684\u9009\u62e9\u3002<\/li>\n<li>MobileNet&#xff1a;\u4e13\u4e3a\u79fb\u52a8\u548c\u5d4c\u5165\u5f0f\u8bbe\u5907\u8bbe\u8ba1&#xff0c;\u6a21\u578b\u5c0f&#xff0c;\u901f\u5ea6\u5feb&#xff0c;\u867d\u7136\u7cbe\u5ea6\u7565\u6709\u727a\u7272&#xff0c;\u4f46\u5728\u8d44\u6e90\u53d7\u9650\u7684\u573a\u666f\u4e0b\u662f\u9996\u9009\u3002<\/li>\n<li>EfficientNet&#xff1a;\u901a\u8fc7\u4e00\u79cd\u590d\u5408\u7f29\u653e\u65b9\u6cd5&#xff0c;\u7cfb\u7edf\u6027\u5730\u5e73\u8861\u4e86\u7f51\u7edc\u7684\u6df1\u5ea6\u3001\u5bbd\u5ea6\u548c\u5206\u8fa8\u7387&#xff0c;\u5728\u6027\u80fd\u548c\u6548\u7387\u4e0a\u90fd\u8fbe\u5230\u4e86\u5f53\u65f6\u7684\u9876\u5c16\u6c34\u5e73\u3002EfficientNetB0\u5230B7\u63d0\u4f9b\u4e86\u4e00\u7cfb\u5217\u4ece\u8f7b\u91cf\u5230\u5e9e\u5927\u7684\u9009\u62e9\u3002<\/li>\n<\/ul>\n<p>\u5bf9\u4e8e\u6211\u4eec\u7684\u5783\u573e\u5206\u7c7b\u4efb\u52a1&#xff0c;MobileNetV2\u6216EfficientNetB0\u4f1a\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u8d77\u70b9&#xff0c;\u5b83\u4eec\u5728\u4fdd\u6301\u8f83\u9ad8\u7cbe\u5ea6\u7684\u540c\u65f6&#xff0c;\u5bf9\u8ba1\u7b97\u8d44\u6e90\u7684\u8981\u6c42\u4e5f\u76f8\u5bf9\u53cb\u597d\u3002<\/p>\n<\/li>\n<li>\n<p>\u5fae\u8c03&#xff08;Fine-tuning&#xff09;\u7b56\u7565 \u62ff\u5230\u9884\u8bad\u7ec3\u6a21\u578b\u540e&#xff0c;\u6211\u4eec\u901a\u5e38\u4f1a\u53bb\u6389\u5b83\u539f\u672c\u7528\u4e8eImageNet\u5206\u7c7b\u7684\u9876\u5c42&#xff08;\u5168\u8fde\u63a5\u5c42&#xff09;&#xff0c;\u7136\u540e\u63a5\u4e0a\u6211\u4eec\u81ea\u5df1\u4e3a\u5783\u573e\u5206\u7c7b\u4efb\u52a1\u8bbe\u8ba1\u7684\u65b0\u7684\u5206\u7c7b\u5934\u3002<\/p>\n<li>\u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b&#xff1a;\u52a0\u8f7d\u6a21\u578b&#xff0c;\u5e76\u6307\u5b9ainclude_top&#061;False\u6765\u79fb\u9664\u9876\u5c42\u3002<\/li>\n<li>\u51bb\u7ed3\u57fa\u7840\u5c42&#xff1a;\u5c06\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u4e3b\u4f53\u90e8\u5206&#xff08;\u6211\u4eec\u79f0\u4e4b\u4e3a\u201c\u57fa\u7840\u6a21\u578b\u201d&#xff09;\u7684\u6743\u91cd\u8bbe\u7f6e\u4e3a\u4e0d\u53ef\u8bad\u7ec3&#xff08;base_model.trainable &#061; False&#xff09;\u3002\u8fd9\u662f\u8fc1\u79fb\u5b66\u4e60\u7684\u7b2c\u4e00\u6b65&#xff0c;\u6211\u4eec\u5148\u7528\u5b83\u5f3a\u5927\u7684\u3001\u56fa\u5b9a\u7684\u7279\u5f81\u63d0\u53d6\u80fd\u529b&#xff0c;\u6765\u8bad\u7ec3\u6211\u4eec\u65b0\u5efa\u7684\u3001\u968f\u673a\u521d\u59cb\u5316\u7684\u5206\u7c7b\u5934\u3002<\/li>\n<li>\u6dfb\u52a0\u81ea\u5b9a\u4e49\u5206\u7c7b\u5934&#xff1a;\u5728\u57fa\u7840\u6a21\u578b\u4e4b\u4e0a&#xff0c;\u6dfb\u52a0\u4e00\u4e2aGlobalAveragePooling2D\u5c42&#xff08;\u7528\u4e8e\u5c06\u7279\u5f81\u56fe\u5c55\u5e73&#xff09;&#xff0c;\u518d\u63a5\u4e0a\u4e00\u4e2a\u6216\u591a\u4e2aDense\u5c42&#xff0c;\u6700\u540e\u662f\u4e00\u4e2a\u8f93\u51fa\u5c42&#xff0c;\u5176\u795e\u7ecf\u5143\u6570\u91cf\u7b49\u4e8e\u6211\u4eec\u7684\u5783\u573e\u7c7b\u522b\u6570&#xff0c;\u6fc0\u6d3b\u51fd\u6570\u4e3asoftmax\u3002<\/li>\n<li>\u7f16\u8bd1\u4e0e\u521d\u6b65\u8bad\u7ec3&#xff1a;\u7528\u8f83\u5c0f\u7684\u5b66\u4e60\u7387\u7f16\u8bd1\u6a21\u578b&#xff0c;\u5e76\u8fdb\u884c\u521d\u6b65\u8bad\u7ec3\u3002<\/li>\n<li>\u89e3\u51bb\u4e0e\u5fae\u8c03&#xff08;\u53ef\u9009\u4f46\u63a8\u8350&#xff09;&#xff1a;\u5f53\u65b0\u7684\u5206\u7c7b\u5934\u8bad\u7ec3\u7a33\u5b9a\u540e&#xff0c;\u6211\u4eec\u53ef\u4ee5\u9009\u62e9\u6027\u5730\u89e3\u51bb\u57fa\u7840\u6a21\u578b\u7684\u9876\u90e8\u51e0\u5c42&#xff08;\u6216\u8005\u5168\u90e8\u5c42&#xff09;&#xff0c;\u5e76\u7528\u4e00\u4e2a\u6781\u4f4e\u7684\u5b66\u4e60\u7387\u7ee7\u7eed\u8fdb\u884c\u7aef\u5230\u7aef\u7684\u8bad\u7ec3\u3002\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u662f\u5fae\u8c03\u3002\u5b83\u5141\u8bb8\u9884\u8bad\u7ec3\u597d\u7684\u5e95\u5c42\u7279\u5f81&#xff0c;\u6839\u636e\u6211\u4eec\u7279\u5b9a\u4efb\u52a1\u7684\u6570\u636e\u8fdb\u884c\u5fae\u5c0f\u7684\u8c03\u6574&#xff0c;\u4ece\u800c\u66f4\u597d\u5730\u9002\u5e94\u65b0\u4efb\u52a1&#xff0c;\u6709\u671b\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u578b\u6027\u80fd\u3002<\/li>\n<\/li>\n<\/ul>\n<p># 1. \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\u4f5c\u4e3a\u57fa\u7840\u6a21\u578b<br \/>\nbase_model &#061; tf.keras.applications.MobileNetV2(input_shape&#061;(224, 224, 3),<br \/>\n                                               include_top&#061;False, # \u4e0d\u5305\u62ec\u9876\u90e8\u7684\u5206\u7c7b\u5668<br \/>\n                                               weights&#061;&#039;imagenet&#039;)<\/p>\n<p># 2. \u51bb\u7ed3\u57fa\u7840\u6a21\u578b<br \/>\nbase_model.trainable &#061; False<\/p>\n<p># 3. \u6dfb\u52a0\u81ea\u5b9a\u4e49\u5206\u7c7b\u5934<br \/>\ninputs &#061; tf.keras.Input(shape&#061;(224, 224, 3))<br \/>\nx &#061; base_model(inputs, training&#061;False) # \u5728\u63a8\u7406\u6a21\u5f0f\u4e0b\u8fd0\u884c\u57fa\u7840\u6a21\u578b<br \/>\nx &#061; tf.keras.layers.GlobalAveragePooling2D()(x)<br \/>\nx &#061; tf.keras.layers.Dropout(0.2)(x) # \u6dfb\u52a0Dropout\u9632\u6b62\u8fc7\u62df\u5408<br \/>\noutputs &#061; tf.keras.layers.Dense(len(train_generator.class_indices), activation&#061;&#039;softmax&#039;)(x)<\/p>\n<p>model &#061; tf.keras.Model(inputs, outputs)<\/p>\n<p>model.summary()<\/p>\n<h5>11.1.4 \u8bad\u7ec3\u4e0e\u8bc4\u4f30&#xff1a;\u70bc\u4e39\u672f\u7684\u5b9e\u6218\u6f14\u7ec3<\/h5>\n<p>\u6a21\u578b\u5df2\u5efa&#xff0c;\u7089\u706b\u5df2\u751f&#xff0c;\u73b0\u5728\u8fdb\u5165\u6700\u5173\u952e\u7684\u201c\u70bc\u4e39\u201d\u73af\u8282\u3002<\/p>\n<ul>\n<li>\n<p>\u7f16\u8bd1\u6a21\u578b \u5728\u8bad\u7ec3\u4e4b\u524d&#xff0c;\u6211\u4eec\u9700\u8981\u4e3a\u6a21\u578b\u914d\u7f6e\u5b66\u4e60\u8fc7\u7a0b\u3002<\/p>\n<ul>\n<li>\u4f18\u5316\u5668&#xff1a;Adam\u4f9d\u7136\u662f\u6211\u4eec\u7684\u9996\u9009&#xff0c;\u5b83\u7a33\u5065\u4e14\u9ad8\u6548\u3002\u521d\u59cb\u5b66\u4e60\u7387\u53ef\u4ee5\u8bbe\u7f6e\u5f97\u7a0d\u4f4e&#xff0c;\u4f8b\u59820.001\u3002<\/li>\n<li>\u635f\u5931\u51fd\u6570&#xff1a;\u5bf9\u4e8e\u591a\u5206\u7c7b\u95ee\u9898&#xff0c;categorical_crossentropy\u662f\u6807\u51c6\u7684\u9009\u62e9&#xff08;\u5982\u679c\u6807\u7b7e\u662f\u6574\u6570\u7f16\u7801&#xff0c;\u5219\u4f7f\u7528sparse_categorical_crossentropy&#xff09;\u3002<\/li>\n<li>\u8bc4\u4f30\u6307\u6807&#xff1a;\u6211\u4eec\u6700\u5173\u5fc3\u7684\u662faccuracy&#xff08;\u51c6\u786e\u7387&#xff09;&#xff0c;\u4f46\u52a0\u5165\u5176\u4ed6\u6307\u6807\u5982Precision\u548cRecall\u80fd\u63d0\u4f9b\u66f4\u4e30\u5bcc\u7684\u89c6\u89d2\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u914d\u7f6e\u56de\u8c03\u51fd\u6570&#xff08;Callbacks&#xff09; \u56de\u8c03\u51fd\u6570\u662fKeras\u4e2d\u975e\u5e38\u5f3a\u5927\u7684\u5de5\u5177&#xff0c;\u5b83\u4eec\u50cf\u662f\u5728\u70bc\u4e39\u7089\u65c1\u5b88\u62a4\u7684\u7ae5\u5b50&#xff0c;\u53ef\u4ee5\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u7279\u5b9a\u8282\u70b9\u6267\u884c\u9884\u8bbe\u7684\u64cd\u4f5c\u3002<\/p>\n<ul>\n<li>ModelCheckpoint&#xff1a;\u5728\u6bcf\u4e2aepoch\u7ed3\u675f\u540e&#xff0c;\u68c0\u67e5\u9a8c\u8bc1\u96c6\u4e0a\u7684\u6027\u80fd&#xff08;\u5982val_accuracy&#xff09;&#xff0c;\u5e76\u53ea\u4fdd\u5b58\u8fc4\u4eca\u4e3a\u6b62\u6027\u80fd\u6700\u597d\u7684\u6a21\u578b\u6743\u91cd\u3002\u8fd9\u662f\u9632\u6b62\u6211\u4eec\u5fc3\u8840\u767d\u8d39\u7684\u201c\u4fdd\u9669\u4e1d\u201d\u3002<\/li>\n<li>EarlyStopping&#xff1a;\u76d1\u63a7\u9a8c\u8bc1\u96c6\u635f\u5931val_loss&#xff0c;\u5982\u679c\u5b83\u5728\u8fde\u7eed\u591a\u4e2aepoch&#xff08;\u7531patience\u53c2\u6570\u8bbe\u5b9a&#xff09;\u5185\u4e0d\u518d\u4e0b\u964d&#xff0c;\u5c31\u63d0\u524d\u7ec8\u6b62\u8bad\u7ec3&#xff0c;\u9632\u6b62\u8fc7\u62df\u5408&#xff0c;\u5e76\u8282\u7ea6\u65f6\u95f4\u3002<\/li>\n<li>ReduceLROnPlateau&#xff1a;\u76d1\u63a7\u67d0\u4e2a\u6307\u6807&#xff08;\u5982val_loss&#xff09;&#xff0c;\u5f53\u5b83\u201c\u505c\u6ede\u4e0d\u524d\u201d\u65f6&#xff0c;\u81ea\u52a8\u964d\u4f4e\u5b66\u4e60\u7387\u3002\u8fd9\u5c31\u50cf\u5728\u4e0b\u5c71\u9047\u5230\u5e73\u5730\u65f6&#xff0c;\u6211\u4eec\u653e\u6162\u811a\u6b65&#xff0c;\u66f4\u7cbe\u7ec6\u5730\u63a2\u7d22\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7ed3\u679c\u5206\u6790 \u8bad\u7ec3\u5b8c\u6210\u540e&#xff0c;model.fit()\u8fd4\u56de\u7684history\u5bf9\u8c61\u8bb0\u5f55\u4e86\u6bcf\u4e2aepoch\u7684\u8bad\u7ec3\u548c\u9a8c\u8bc1\u6307\u6807\u3002\u5c06\u5176\u53ef\u89c6\u5316\u662f\u5206\u6790\u6a21\u578b\u884c\u4e3a\u7684\u6700\u4f73\u65b9\u5f0f\u3002<\/p>\n<ul>\n<li>\u89e3\u8bfb\u8bad\u7ec3\u66f2\u7ebf&#xff1a;\u7ed8\u5236\u8bad\u7ec3\/\u9a8c\u8bc1\u7684\u635f\u5931\u548c\u51c6\u786e\u7387\u66f2\u7ebf\u3002\u7406\u60f3\u60c5\u51b5\u4e0b&#xff0c;\u4e24\u6761\u66f2\u7ebf\u90fd\u5e94\u5e73\u7a33\u6536\u655b\u3002\u5982\u679c\u8bad\u7ec3\u635f\u5931\u8fdc\u4f4e\u4e8e\u9a8c\u8bc1\u635f\u5931&#xff0c;\u4e14\u4e8c\u8005\u5dee\u8ddd\u968f\u65f6\u95f4\u62c9\u5927&#xff0c;\u5219\u662f\u660e\u663e\u7684\u8fc7\u62df\u5408\u4fe1\u53f7\u3002<\/li>\n<li>\u4f7f\u7528\u6df7\u6dc6\u77e9\u9635&#xff1a;\u6df7\u6dc6\u77e9\u9635&#xff08;Confusion Matrix&#xff09;\u80fd\u6e05\u6670\u5730\u5c55\u793a\u6a21\u578b\u5728\u54ea\u4e9b\u7c7b\u522b\u4e0a\u8868\u73b0\u826f\u597d&#xff0c;\u53c8\u5bb9\u6613\u5c06\u54ea\u4e9b\u7c7b\u522b\u6df7\u6dc6\u3002\u4f8b\u5982&#xff0c;\u6a21\u578b\u662f\u5426\u7ecf\u5e38\u628a\u201c\u7eb8\u677f\u201d\u9519\u8ba4\u4e3a\u201c\u7eb8\u5f20\u201d&#xff1f;\u8fd9\u4e3a\u6211\u4eec\u6307\u660e\u4e86\u6a21\u578b\u6539\u8fdb\u7684\u65b9\u5411&#xff08;\u6bd4\u5982&#xff0c;\u589e\u52a0\u8fd9\u4e24\u7c7b\u6613\u6df7\u6dc6\u6837\u672c\u7684\u8bad\u7ec3\u6570\u636e&#xff09;\u3002<\/li>\n<li>\u7cbe\u786e\u7387\u3001\u53ec\u56de\u7387\u3001F1\u5206\u6570&#xff1a;\u9664\u4e86\u603b\u4f53\u51c6\u786e\u7387&#xff0c;\u6211\u4eec\u8fd8\u5e94\u8ba1\u7b97\u6bcf\u4e2a\u7c7b\u522b\u7684\u8fd9\u4e9b\u6307\u6807&#xff0c;\u4ee5\u83b7\u5f97\u66f4\u5168\u9762\u7684\u6027\u80fd\u8bc4\u4f30&#xff0c;\u5c24\u5176\u662f\u5728\u7c7b\u522b\u4e0d\u5747\u8861\u7684\u60c5\u51b5\u4e0b\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p># \u7f16\u8bd1\u6a21\u578b<br \/>\nmodel.compile(optimizer&#061;tf.keras.optimizers.Adam(learning_rate&#061;0.001),<br \/>\n              loss&#061;&#039;categorical_crossentropy&#039;,<br \/>\n              metrics&#061;[&#039;accuracy&#039;])<\/p>\n<p># \u914d\u7f6e\u56de\u8c03\u51fd\u6570<br \/>\ncheckpoint_cb &#061; tf.keras.callbacks.ModelCheckpoint(&#034;best_garbage_model.h5&#034;, save_best_only&#061;True)<br \/>\nearly_stopping_cb &#061; tf.keras.callbacks.EarlyStopping(patience&#061;10, restore_best_weights&#061;True)<\/p>\n<p># \u5f00\u59cb\u8bad\u7ec3<br \/>\nhistory &#061; model.fit(train_generator,<br \/>\n                    epochs&#061;50, # \u5148\u8bad\u7ec3\u8f83\u591a\u8f6e\u6b21&#xff0c;\u8ba9EarlyStopping\u51b3\u5b9a\u4f55\u65f6\u505c\u6b62<br \/>\n                    validation_data&#061;validation_generator,<br \/>\n                    callbacks&#061;[checkpoint_cb, early_stopping_cb])<\/p>\n<h5>11.1.5 \u90e8\u7f72\u4e0e\u5e94\u7528&#xff1a;\u8ba9\u6a21\u578b\u8d70\u51fa\u5b9e\u9a8c\u5ba4<\/h5>\n<p>\u4e00\u4e2a\u6a21\u578b\u771f\u6b63\u7684\u4ef7\u503c&#xff0c;\u5728\u4e8e\u5b83\u80fd\u88ab\u5e94\u7528\u3002\u6211\u4eec\u5c06\u5b66\u4e60\u5982\u4f55\u5c06\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u4ece\u4e00\u4e2a\u7814\u7a76\u539f\u578b&#xff0c;\u53d8\u6210\u4e00\u4e2a\u53ef\u4ee5\u63d0\u4f9b\u670d\u52a1\u7684\u7b80\u5355\u5e94\u7528\u3002<\/p>\n<ul>\n<li>\n<p>\u6a21\u578b\u5bfc\u51fa\u4e0e\u5c01\u88c5 \u8bad\u7ec3\u5b8c\u6210\u540e&#xff0c;\u6211\u4eec\u5c06\u6027\u80fd\u6700\u4f73\u7684\u6a21\u578b\u4fdd\u5b58\u4e3aKeras\u7684H5\u683c\u5f0f\u6216TensorFlow\u7684SavedModel\u683c\u5f0f\u3002SavedModel\u683c\u5f0f\u66f4\u9002\u5408\u90e8\u7f72&#xff0c;\u56e0\u4e3a\u5b83\u5305\u542b\u4e86\u5b8c\u6574\u7684\u8ba1\u7b97\u56fe\u548c\u6743\u91cd&#xff0c;\u5177\u6709\u66f4\u597d\u7684\u8de8\u5e73\u53f0\u517c\u5bb9\u6027\u3002<\/p>\n<\/li>\n<li>\n<p>\u6784\u5efa\u7b80\u5355\u7684Web\u5e94\u7528 \u6211\u4eec\u53ef\u4ee5\u4f7f\u7528\u8f7b\u91cf\u7ea7\u7684Python Web\u6846\u67b6&#xff0c;\u5982Flask\u6216FastAPI&#xff0c;\u6765\u4e3a\u6211\u4eec\u7684\u6a21\u578b\u7a7f\u4e0a\u4e00\u5c42\u201c\u5916\u8863\u201d&#xff0c;\u8ba9\u5b83\u80fd\u901a\u8fc7HTTP\u534f\u8bae\u4e0e\u5916\u754c\u4ea4\u4e92\u3002<\/p>\n<ul>\n<li>\u57fa\u672c\u6d41\u7a0b&#xff1a;\n<li>\u521b\u5efa\u4e00\u4e2aWeb\u670d\u52a1\u3002<\/li>\n<li>\u5b9a\u4e49\u4e00\u4e2aAPI\u7aef\u70b9&#xff08;\u4f8b\u5982\/predict&#xff09;&#xff0c;\u8be5\u7aef\u70b9\u63a5\u53d7\u7528\u6237\u901a\u8fc7POST\u8bf7\u6c42\u4e0a\u4f20\u7684\u56fe\u7247\u3002<\/li>\n<li>\u5728\u540e\u7aef&#xff0c;\u670d\u52a1\u63a5\u6536\u56fe\u7247&#xff0c;\u8fdb\u884c\u4e0e\u8bad\u7ec3\u65f6\u76f8\u540c\u7684\u9884\u5904\u7406&#xff08;\u7f29\u653e\u5c3a\u5bf8\u3001\u5f52\u4e00\u5316&#xff09;\u3002<\/li>\n<li>\u52a0\u8f7d\u6211\u4eec\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5bf9\u9884\u5904\u7406\u540e\u7684\u56fe\u7247\u8fdb\u884c\u9884\u6d4b\u3002<\/li>\n<li>\u5c06\u9884\u6d4b\u7ed3\u679c&#xff08;\u7c7b\u522b\u540d\u79f0\u548c\u7f6e\u4fe1\u5ea6&#xff09;\u4ee5JSON\u683c\u5f0f\u8fd4\u56de\u7ed9\u7528\u6237\u3002<\/li>\n<\/li>\n<\/ul>\n<p>\u8fd9\u6837\u4e00\u4e2a\u7b80\u5355\u7684Web\u5e94\u7528&#xff0c;\u5c31\u8ba9\u6211\u4eec\u7684\u5783\u573e\u5206\u7c7b\u6a21\u578b\u201c\u6d3b\u201d\u4e86\u8d77\u6765\u3002\u4efb\u4f55\u6709\u7f51\u7edc\u8fde\u63a5\u7684\u8bbe\u5907&#xff08;\u5982\u624b\u673a\u3001\u7535\u8111&#xff09;\u90fd\u53ef\u4ee5\u901a\u8fc7\u8bbf\u95ee\u8fd9\u4e2a\u670d\u52a1&#xff0c;\u6765\u4f7f\u7528\u6211\u4eec\u7684AI\u80fd\u529b\u3002\u8fd9\u4fbf\u662f\u4ece\u7406\u8bba\u5230\u4ef7\u503c\u8f6c\u5316\u7684\u201c\u6700\u540e\u4e00\u516c\u91cc\u201d\u3002<\/p>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>11.2 \u76ee\u6807\u68c0\u6d4b&#xff1a;\u5b9e\u73b0\u4e00\u4e2a\u5b9e\u65f6\u4eba\u8138\u6216\u8f66\u8f86\u68c0\u6d4b\u5668<\/h4>\n<p>\u5b8c\u6210\u4e86\u56fe\u50cf\u5206\u7c7b&#xff0c;\u6211\u4eec\u8ba9\u673a\u5668\u5b66\u4f1a\u4e86\u56de\u7b54\u201c\u8fd9\u5f20\u56fe\u91cc\u6709\u4ec0\u4e48&#xff1f;\u201d\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u8981\u66f4\u8fdb\u4e00\u6b65&#xff0c;\u8ba9\u5b83\u56de\u7b54\u4e00\u4e2a\u66f4\u590d\u6742\u7684\u95ee\u9898&#xff1a;\u201c\u8fd9\u5f20\u56fe\u91cc\u7684\u4e1c\u897f&#xff0c;\u90fd\u5728\u54ea\u91cc&#xff1f;\u201d\u3002\u8fd9\u4fbf\u662f\u76ee\u6807\u68c0\u6d4b&#xff08;Object Detection&#xff09;\u3002<\/p>\n<h5>11.2.1 \u4ece\u5206\u7c7b\u5230\u68c0\u6d4b&#xff1a;\u4e0d\u6b62\u201c\u662f\u4ec0\u4e48\u201d&#xff0c;\u66f4\u8981\u201c\u5728\u54ea\u91cc\u201d<\/h5>\n<ul>\n<li>\n<p>\u4efb\u52a1\u5b9a\u4e49 \u76ee\u6807\u68c0\u6d4b\u4e0d\u4ec5\u8981\u8bc6\u522b\u51fa\u56fe\u50cf\u4e2d\u7684\u7269\u4f53\u7c7b\u522b&#xff08;Classification&#xff09;&#xff0c;\u8fd8\u8981\u7528\u4e00\u4e2a**\u8fb9\u754c\u6846&#xff08;Bounding Box&#xff09;**\u7cbe\u786e\u5730\u6807\u51fa\u5b83\u7684\u4f4d\u7f6e&#xff08;Localization&#xff09;\u3002\u4e00\u5f20\u56fe\u4e2d\u53ef\u80fd\u5305\u542b\u591a\u4e2a\u4e0d\u540c\u7c7b\u522b\u7684\u7269\u4f53&#xff0c;\u6a21\u578b\u9700\u8981\u5c06\u5b83\u4eec\u5168\u90e8\u627e\u51fa\u3002<\/p>\n<\/li>\n<li>\n<p>\u6838\u5fc3\u6311\u6218 \u76f8\u6bd4\u5206\u7c7b&#xff0c;\u76ee\u6807\u68c0\u6d4b\u7684\u590d\u6742\u6027\u6307\u6570\u7ea7\u589e\u957f\u3002\u6a21\u578b\u9700\u8981\u540c\u65f6\u8f93\u51fa\u4e24\u90e8\u5206\u4fe1\u606f&#xff1a;<\/p>\n<li>\u7c7b\u522b\u6982\u7387&#xff1a;\u8fd9\u4e2a\u6846\u91cc\u662f\u732b\u7684\u6982\u7387\u662f\u591a\u5c11&#xff1f;\u662f\u72d7\u7684\u6982\u7387\u662f\u591a\u5c11&#xff1f;<\/li>\n<li>\u8fb9\u754c\u6846\u5750\u6807&#xff1a;\u901a\u5e38\u7528\u6846\u7684\u4e2d\u5fc3\u70b9\u5750\u6807(x, y)\u4ee5\u53ca\u6846\u7684\u5bbd(w)\u548c\u9ad8(h)\u6765\u8868\u793a\u3002 \u6a21\u578b\u9700\u8981\u5728\u4e00\u4e2a\u753b\u9762\u4e2d&#xff0c;\u9762\u5bf9\u4e0d\u540c\u5c3a\u5bf8\u3001\u4e0d\u540c\u957f\u5bbd\u6bd4\u3001\u76f8\u4e92\u906e\u6321\u3001\u6570\u91cf\u4e0d\u5b9a\u7684\u76ee\u6807&#xff0c;\u8fdb\u884c\u9ad8\u6548\u4e14\u51c6\u786e\u7684\u9884\u6d4b\u3002<\/li>\n<\/li>\n<\/ul>\n<h5>11.2.2 \u4e3b\u6d41\u6846\u67b6\u4e0e\u6a21\u578b\u6982\u89c8<\/h5>\n<p>\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u4e3b\u8981\u5206\u4e3a\u4e24\u5927\u6d41\u6d3e&#xff1a;<\/p>\n<ul>\n<li>\n<p>\u4e24\u9636\u6bb5\u68c0\u6d4b\u5668&#xff08;Two-stage Detectors&#xff09; \u8fd9\u7c7b\u65b9\u6cd5\u7684\u601d\u60f3\u662f\u201c\u5148\u63d0\u8bae&#xff0c;\u540e\u5206\u7c7b\u201d&#xff0c;\u5206\u4e24\u6b65\u8d70\u3002<\/p>\n<li>\u533a\u57df\u63d0\u8bae&#xff08;Region Proposal&#xff09;&#xff1a;\u9996\u5148&#xff0c;\u901a\u8fc7\u4e00\u4e2a\u72ec\u7acb\u7684\u7f51\u7edc&#xff08;\u5982RPN, Region Proposal Network&#xff09;\u5feb\u901f\u5730\u5728\u56fe\u50cf\u4e0a\u626b\u63cf&#xff0c;\u627e\u51fa\u6240\u6709\u53ef\u80fd\u5305\u542b\u7269\u4f53\u7684\u5019\u9009\u533a\u57df\u3002<\/li>\n<li>\u5206\u7c7b\u4e0e\u56de\u5f52&#xff1a;\u7136\u540e&#xff0c;\u5bf9\u8fd9\u4e9b\u5019\u9009\u533a\u57df\u8fdb\u884c\u7cbe\u786e\u7684\u5206\u7c7b\u548c\u8fb9\u754c\u6846\u4f4d\u7f6e\u7684\u5fae\u8c03\u3002<\/li>\n<ul>\n<li>\u4ee3\u8868&#xff1a;R-CNN\u5bb6\u65cf&#xff0c;\u5982Fast R-CNN\u548cFaster R-CNN\u3002<\/li>\n<li>\u7279\u70b9&#xff1a;\u7cbe\u5ea6\u9ad8&#xff0c;\u4f46\u56e0\u4e3a\u662f\u4e24\u6b65\u64cd\u4f5c&#xff0c;\u901f\u5ea6\u76f8\u5bf9\u8f83\u6162\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5355\u9636\u6bb5\u68c0\u6d4b\u5668&#xff08;One-stage Detectors&#xff09; \u8fd9\u7c7b\u65b9\u6cd5\u8ffd\u6c42\u6781\u81f4\u7684\u901f\u5ea6\u4e0e\u6548\u7387&#xff0c;\u5b83\u4eec\u5c06\u533a\u57df\u63d0\u8bae\u548c\u5206\u7c7b\/\u56de\u5f52\u4e24\u6b65\u5408\u4e8c\u4e3a\u4e00\u3002<\/p>\n<ul>\n<li>\u4ee3\u8868&#xff1a;YOLO (You Only Look Once)\u00a0\u548c\u00a0SSD (Single Shot MultiBox Detector)\u3002<\/li>\n<li>\u601d\u60f3&#xff1a;\u5b83\u4eec\u5c06\u56fe\u50cf\u5212\u5206\u4e3a\u7f51\u683c&#xff08;grid&#xff09;&#xff0c;\u76f4\u63a5\u5728\u6bcf\u4e2a\u7f51\u683c\u4e0a\u9884\u6d4b\u53ef\u80fd\u5b58\u5728\u7684\u7269\u4f53\u53ca\u5176\u8fb9\u754c\u6846\u548c\u7c7b\u522b\u3002\u6574\u4e2a\u8fc7\u7a0b\u4e00\u6c14\u5475\u6210&#xff0c;\u53ea\u201c\u770b\u201d\u4e00\u6b21\u56fe\u50cf\u3002<\/li>\n<li>\u7279\u70b9&#xff1a;\u901f\u5ea6\u6781\u5feb&#xff0c;\u53ef\u4ee5\u8fbe\u5230\u5b9e\u65f6\u68c0\u6d4b\u7684\u8981\u6c42&#xff0c;\u867d\u7136\u65e9\u671f\u7248\u672c\u7684\u7cbe\u5ea6\u7565\u4f4e\u4e8e\u4e24\u9636\u6bb5\u68c0\u6d4b\u5668&#xff0c;\u4f46\u968f\u7740\u53d1\u5c55&#xff08;\u5c24\u5176\u662fYOLO\u7cfb\u5217&#xff09;&#xff0c;\u5176\u7cbe\u5ea6\u5df2\u7ecf\u975e\u5e38\u6709\u7ade\u4e89\u529b\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6a21\u578b\u9009\u62e9 \u5728\u9700\u8981\u8fdb\u884c\u5b9e\u65f6\u68c0\u6d4b\u7684\u5e94\u7528\u573a\u666f&#xff0c;\u5982\u89c6\u9891\u76d1\u63a7\u3001\u81ea\u52a8\u9a7e\u9a76\u8f85\u52a9\u3001\u52a8\u6001\u4ea4\u4e92\u7b49&#xff0c;YOLO\u7cfb\u5217\u65e0\u7591\u662f\u5f53\u4eca\u4e1a\u754c\u7684\u738b\u8005\u548c\u9996\u9009\u3002\u5b83\u7684\u8bbe\u8ba1\u54f2\u5b66\u2014\u2014\u5feb\u3001\u51c6\u3001\u72e0\u2014\u2014\u5b8c\u7f8e\u5951\u5408\u4e86\u8fd9\u4e9b\u9700\u6c42\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>11.2.3 \u5b9e\u6218YOLOv5\/v8&#xff1a;\u5feb\u3001\u51c6\u3001\u72e0\u7684\u68c0\u6d4b\u5229\u5668<\/h5>\n<p>\u6211\u4eec\u5c06\u76f4\u63a5\u4f7f\u7528\u76ee\u524d\u6700\u6d41\u884c\u3001\u751f\u6001\u6700\u5b8c\u5584\u7684YOLOv8&#xff08;\u7531Ultralytics\u516c\u53f8\u5f00\u53d1&#xff09;\u6765\u8fdb\u884c\u5b9e\u6218\u3002\u5b83\u5c06\u590d\u6742\u7684\u6a21\u578b\u8bad\u7ec3\u548c\u63a8\u7406\u8fc7\u7a0b\u5c01\u88c5\u5f97\u6781\u5176\u53cb\u597d\u3002<\/p>\n<ul>\n<li>\n<p>\u73af\u5883\u914d\u7f6e\u4e0e\u5b89\u88c5 \u5b89\u88c5YOLOv8\u975e\u5e38\u7b80\u5355&#xff0c;\u53ea\u9700\u4e00\u884cpip\u547d\u4ee4&#xff1a; pip install ultralytics<\/p>\n<\/li>\n<li>\n<p>\u4f7f\u7528\u9884\u8bad\u7ec3\u6a21\u578b\u8fdb\u884c\u63a8\u7406 YOLOv8\u63d0\u4f9b\u4e86\u5728COCO\u6570\u636e\u96c6\u4e0a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u53ef\u4ee5\u8bc6\u522b80\u4e2a\u5e38\u89c1\u7c7b\u522b\u3002\u6211\u4eec\u53ef\u4ee5\u76f4\u63a5\u8c03\u7528\u5b83\u6765\u4f53\u9a8c\u5176\u5f3a\u5927\u7684\u5a01\u529b\u3002<\/p>\n<p> from ultralytics import YOLO<\/p>\n<p># \u52a0\u8f7d\u4e00\u4e2a\u9884\u8bad\u7ec3\u7684YOLOv8n\u6a21\u578b (n\u4ee3\u8868nano&#xff0c;\u6700\u5c0f\u6700\u5feb\u7684\u7248\u672c)<br \/>\nmodel &#061; YOLO(&#039;yolov8n.pt&#039;)<\/p>\n<p># \u4ece\u56fe\u7247\u8fdb\u884c\u9884\u6d4b<br \/>\nresults &#061; model(&#039;path\/to\/your\/image.jpg&#039;)<\/p>\n<p># \u4ece\u89c6\u9891\u8fdb\u884c\u9884\u6d4b&#xff0c;\u5e76\u5b9e\u65f6\u663e\u793a\u7ed3\u679c<br \/>\n# results &#061; model(&#039;path\/to\/your\/video.mp4&#039;, show&#061;True)<\/p>\n<p># \u4fdd\u5b58\u9884\u6d4b\u7ed3\u679c<br \/>\n# model.predict(&#039;path\/to\/your\/image.jpg&#039;, save&#061;True)<\/p>\n<p>\u4ec5\u4ec5\u51e0\u884c\u4ee3\u7801&#xff0c;\u4e00\u4e2a\u5f3a\u5927\u7684\u76ee\u6807\u68c0\u6d4b\u5668\u5c31\u5f00\u59cb\u4e3a\u60a8\u5de5\u4f5c\u4e86&#xff01;<\/p>\n<\/li>\n<li>\n<p>\u5728\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u8bad\u7ec3 YOLO\u7684\u771f\u6b63\u9b45\u529b\u5728\u4e8e&#xff0c;\u6211\u4eec\u53ef\u4ee5\u5f88\u65b9\u4fbf\u5730\u7528\u5b83\u6765\u8bad\u7ec3\u81ea\u5df1\u7684\u68c0\u6d4b\u4efb\u52a1&#xff08;\u6bd4\u5982&#xff0c;\u68c0\u6d4b\u5b89\u5168\u5e3d\u4f69\u6234\u3001\u8bc6\u522b\u7279\u5b9a\u79cd\u7c7b\u7684\u91ce\u751f\u52a8\u7269\u7b49&#xff09;\u3002<\/p>\n<li>\u51c6\u5907\u6570\u636e\u96c6&#xff1a;\u8fd9\u662f\u6700\u8017\u65f6\u7684\u4e00\u6b65\u3002\u60a8\u9700\u8981\u5927\u91cf\u7684\u56fe\u7247&#xff0c;\u5e76\u4f7f\u7528\u6807\u6ce8\u5de5\u5177&#xff08;\u5982LabelImg, CVAT&#xff09;\u4e3a\u6bcf\u5f20\u56fe\u7247\u4e2d\u7684\u6bcf\u4e2a\u76ee\u6807\u7269\u4f53\u753b\u4e0a\u8fb9\u754c\u6846&#xff0c;\u5e76\u6307\u5b9a\u5176\u7c7b\u522b\u3002<\/li>\n<li>\u7ec4\u7ec7\u6570\u636e\u683c\u5f0f&#xff1a;YOLO\u9700\u8981\u7279\u5b9a\u7684\u76ee\u5f55\u7ed3\u6784\u548c\u6807\u7b7e\u6587\u4ef6\u683c\u5f0f\u3002\u901a\u5e38\u662f\u56fe\u7247\u6587\u4ef6\u548c\u5bf9\u5e94\u7684txt\u6807\u7b7e\u6587\u4ef6\u653e\u5728\u4e00\u8d77&#xff0c;txt\u6587\u4ef6\u4e2d\u6bcf\u4e00\u884c\u4ee3\u8868\u4e00\u4e2a\u7269\u4f53&#xff0c;\u683c\u5f0f\u4e3a&lt;class_id&gt; &lt;x_center&gt; &lt;y_center&gt; &lt;width&gt; &lt;height&gt;&#xff08;\u5747\u4e3a\u5f52\u4e00\u5316\u540e\u7684\u503c&#xff09;\u3002<\/li>\n<li>\u521b\u5efaYAML\u914d\u7f6e\u6587\u4ef6&#xff1a;\u60a8\u9700\u8981\u521b\u5efa\u4e00\u4e2a.yaml\u6587\u4ef6&#xff0c;\u544a\u8bc9YOLO\u60a8\u7684\u8bad\u7ec3\u6570\u636e\u5728\u54ea\u91cc\u3001\u9a8c\u8bc1\u6570\u636e\u5728\u54ea\u91cc\u3001\u6709\u591a\u5c11\u4e2a\u7c7b\u522b\u4ee5\u53ca\u7c7b\u522b\u7684\u540d\u79f0\u662f\u4ec0\u4e48\u3002<\/li>\n<li>\u5f00\u542f\u8bad\u7ec3&#xff1a;\u901a\u8fc7\u547d\u4ee4\u884c\u6216Python\u811a\u672c&#xff0c;\u8c03\u7528YOLO\u7684\u8bad\u7ec3\u529f\u80fd&#xff0c;\u6307\u5b9a\u60a8\u7684\u6a21\u578b&#xff08;\u53ef\u4ee5\u4ece\u9884\u8bad\u7ec3\u6a21\u578b\u5f00\u59cb\u5fae\u8c03&#xff09;\u3001\u6570\u636e\u914d\u7f6e\u6587\u4ef6\u3001\u8bad\u7ec3\u8f6e\u6b21\u7b49\u53c2\u6570\u3002<\/li>\n<p> # \u901a\u8fc7Python\u811a\u672c\u5f00\u59cb\u8bad\u7ec3<br \/>\nfrom ultralytics import YOLO<\/p>\n<p># \u52a0\u8f7d\u4e00\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b\u4f5c\u4e3a\u8d77\u70b9<br \/>\nmodel &#061; YOLO(&#039;yolov8n.pt&#039;)<\/p>\n<p># \u8bad\u7ec3\u6a21\u578b<br \/>\nresults &#061; model.train(data&#061;&#039;path\/to\/your\/data.yaml&#039;, epochs&#061;100, imgsz&#061;640)\n <\/li>\n<\/ul>\n<h5>11.2.4 \u6027\u80fd\u8bc4\u4f30\u4e0e\u4f18\u5316<\/h5>\n<ul>\n<li>\n<p>\u5173\u952e\u6307\u6807<\/p>\n<ul>\n<li>\u4ea4\u5e76\u6bd4&#xff08;Intersection over Union, IoU&#xff09;&#xff1a;\u8fd9\u662f\u8861\u91cf\u9884\u6d4b\u6846\u4e0e\u771f\u5b9e\u6846\u91cd\u5408\u7a0b\u5ea6\u7684\u6307\u6807\u3002\u5b83\u662f\u9884\u6d4b\u6846\u548c\u771f\u5b9e\u6846\u7684\u4ea4\u96c6\u9762\u79ef\u9664\u4ee5\u5b83\u4eec\u7684\u5e76\u96c6\u9762\u79ef\u3002IoU\u9608\u503c&#xff08;\u59820.5&#xff09;\u5e38\u88ab\u7528\u6765\u5224\u65ad\u4e00\u4e2a\u9884\u6d4b\u662f\u5426\u4e3a\u201c\u6b63\u786e\u201d\u7684\u3002<\/li>\n<li>\u5e73\u5747\u7cbe\u5ea6\u5747\u503c&#xff08;mean Average Precision, mAP&#xff09;&#xff1a;\u8fd9\u662f\u76ee\u6807\u68c0\u6d4b\u9886\u57df\u6700\u6838\u5fc3\u7684\u8bc4\u4f30\u6307\u6807\u3002\u5b83\u7efc\u5408\u4e86\u6a21\u578b\u5728\u6240\u6709\u7c7b\u522b\u3001\u4e0d\u540cIoU\u9608\u503c\u548c\u4e0d\u540c\u7f6e\u4fe1\u5ea6\u9608\u503c\u4e0b\u7684\u7cbe\u786e\u7387\u548c\u53ec\u56de\u7387\u8868\u73b0&#xff0c;\u80fd\u591f\u975e\u5e38\u5168\u9762\u5730\u8861\u91cf\u6a21\u578b\u7684\u6027\u80fd\u3002mAP&#064;.5\u8868\u793a\u5728IoU\u9608\u503c\u4e3a0.5\u65f6\u7684mAP&#xff0c;mAP&#064;.5:.95\u8868\u793a\u5728IoU\u9608\u503c\u4ece0.5\u52300.95&#xff08;\u6b65\u957f0.05&#xff09;\u533a\u95f4\u5185\u6240\u6709mAP\u7684\u5e73\u5747\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7ed3\u679c\u53ef\u89c6\u5316\u4e0e\u5206\u6790 YOLO\u5728\u8bad\u7ec3\u7ed3\u675f\u540e&#xff0c;\u4f1a\u81ea\u52a8\u751f\u6210\u4e30\u5bcc\u7684\u53ef\u89c6\u5316\u7ed3\u679c&#xff0c;\u5305\u62ec\u635f\u5931\u66f2\u7ebf\u3001\u6df7\u6dc6\u77e9\u9635\u3001\u4ee5\u53ca\u5404\u4e2a\u7c7b\u522b\u7684P-R&#xff08;\u7cbe\u786e\u7387-\u53ec\u56de\u7387&#xff09;\u66f2\u7ebf\u7b49&#xff0c;\u4fdd\u5b58\u5728runs\/detect\/train\u76ee\u5f55\u4e0b\u3002\u4ed4\u7ec6\u5206\u6790\u8fd9\u4e9b\u56fe\u8868&#xff0c;\u662f\u8bca\u65ad\u6a21\u578b\u95ee\u9898\u3001\u8fdb\u884c\u4f18\u5316\u7684\u5173\u952e\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>11.2.5 \u5e94\u7528\u573a\u666f\u63a2\u8ba8&#xff1a;\u4ece\u5b89\u9632\u5230\u81ea\u52a8\u9a7e\u9a76<\/h5>\n<p>\u76ee\u6807\u68c0\u6d4b\u6280\u672f\u662f\u5f53\u4ecaAI\u5e94\u7528\u6700\u5e7f\u6cdb\u7684\u5206\u652f\u4e4b\u4e00\u3002\u5b83\u7684\u8eab\u5f71\u65e0\u5904\u4e0d\u5728&#xff1a;<\/p>\n<ul>\n<li>\u667a\u80fd\u5b89\u9632&#xff1a;\u4eba\u8138\u8bc6\u522b\u95e8\u7981\u3001\u5f02\u5e38\u884c\u4e3a\u68c0\u6d4b\u3001\u4eba\u7fa4\u5bc6\u5ea6\u5206\u6790\u3002<\/li>\n<li>\u81ea\u52a8\u9a7e\u9a76&#xff1a;\u5b9e\u65f6\u68c0\u6d4b\u8f66\u8f86\u3001\u884c\u4eba\u3001\u4ea4\u901a\u6807\u5fd7&#xff0c;\u662f\u5b9e\u73b0\u73af\u5883\u611f\u77e5\u7684\u6838\u5fc3\u3002<\/li>\n<li>\u96f6\u552e\u884c\u4e1a&#xff1a;\u65e0\u4eba\u5546\u5e97\u7684\u5546\u54c1\u8bc6\u522b\u3001\u8d27\u67b6\u7a7a\u7f3a\u68c0\u6d4b\u3001\u5ba2\u6d41\u5206\u6790\u3002<\/li>\n<li>\u533b\u7597\u5f71\u50cf&#xff1a;\u5728CT\u6216X\u5149\u7247\u4e2d\u81ea\u52a8\u68c0\u6d4b\u80bf\u7624\u3001\u75c5\u7076\u7b49\u3002<\/li>\n<\/ul>\n<p>\u638c\u63e1\u4e86YOLO&#xff0c;\u60a8\u5c31\u62e5\u6709\u4e86\u4e00\u628a\u6253\u5f00\u65e0\u6570\u8ba1\u7b97\u673a\u89c6\u89c9\u5e94\u7528\u5927\u95e8\u7684\u94a5\u5319\u3002<\/p>\n<hr \/>\n<h4>11.3 \u56fe\u50cf\u98ce\u683c\u8fc1\u79fb&#xff1a;\u5c06\u7167\u7247\u53d8\u6210\u68b5\u9ad8\u98ce\u683c\u7684\u6cb9\u753b<\/h4>\n<p>\u5728\u5b8c\u6210\u4e86\u4e24\u4e2a\u201c\u5b9e\u7528\u4e3b\u4e49\u201d\u7684\u9879\u76ee\u540e&#xff0c;\u8ba9\u6211\u4eec\u5c06\u76ee\u5149\u6295\u5411\u4e00\u4e2a\u5145\u6ee1\u521b\u9020\u6027\u4e0e\u7f8e\u611f\u7684\u9886\u57df&#xff1a;\u795e\u7ecf\u98ce\u683c\u8fc1\u79fb&#xff08;Neural Style Transfer&#xff09;\u3002\u8fd9\u4e2a\u9879\u76ee\u5c06\u5411\u6211\u4eec\u5c55\u793a&#xff0c;\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e0d\u4ec5\u80fd\u8bc6\u522b\u4e16\u754c&#xff0c;\u8fd8\u80fd\u4ee5\u4e00\u79cd\u4ee4\u4eba\u60ca\u53f9\u7684\u65b9\u5f0f\u201c\u91cd\u65b0\u8be0\u91ca\u201d\u4e16\u754c\u3002<\/p>\n<h5>11.3.1 \u9879\u76ee\u6784\u60f3&#xff1a;\u5f53\u795e\u7ecf\u7f51\u7edc\u6210\u4e3a\u201c\u753b\u5bb6\u201d<\/h5>\n<ul>\n<li>\n<p>\u7075\u611f\u6765\u6e90 2015\u5e74&#xff0c;Gatys\u7b49\u4eba\u5728\u4e00\u7bc7\u8bba\u6587\u4e2d\u9996\u6b21\u63d0\u51fa\u4e86\u8fd9\u4e2a\u60f3\u6cd5&#xff0c;\u5e76\u5c55\u793a\u4e86\u60ca\u4eba\u7684\u6548\u679c&#xff1a;\u4ed6\u4eec\u80fd\u5c06\u4e00\u5f20\u666e\u901a\u7167\u7247\u7684\u5185\u5bb9&#xff0c;\u4e0e\u68b5\u9ad8\u7684\u300a\u661f\u591c\u300b\u3001\u8499\u514b\u7684\u300a\u5450\u558a\u300b\u7b49\u540d\u753b\u7684\u98ce\u683c\u878d\u5408\u5728\u4e00\u8d77&#xff0c;\u751f\u6210\u4e00\u5e45\u5168\u65b0\u7684\u3001\u517c\u5177\u4e8c\u8005\u795e\u97f5\u7684\u827a\u672f\u4f5c\u54c1\u3002<\/p>\n<\/li>\n<li>\n<p>\u6838\u5fc3\u601d\u60f3 \u8fd9\u4e2a\u7b97\u6cd5\u7684\u7edd\u5999\u4e4b\u5904\u5728\u4e8e&#xff0c;\u5b83\u8ba4\u4e3a\u4e00\u5f20\u56fe\u50cf\u7684\u5185\u5bb9&#xff08;Content&#xff09;\u548c\u98ce\u683c&#xff08;Style&#xff09;\u662f\u53ef\u4ee5\u88ab\u795e\u7ecf\u7f51\u7edc\u5206\u79bb\u5f00\u6765&#xff0c;\u5e76\u8fdb\u884c\u91cf\u5316\u548c\u91cd\u7ec4\u7684\u3002<\/p>\n<ul>\n<li>\u5185\u5bb9&#xff1a;\u6307\u7684\u662f\u56fe\u50cf\u4e2d\u7269\u4f53\u7684\u5b8f\u89c2\u7ed3\u6784\u548c\u6392\u5217&#xff0c;\u5373\u201c\u753b\u4e86\u4ec0\u4e48\u201d\u3002<\/li>\n<li>\u98ce\u683c&#xff1a;\u6307\u7684\u662f\u56fe\u50cf\u7684\u7eb9\u7406\u3001\u7b14\u89e6\u3001\u8272\u5f69\u642d\u914d\u7b49&#xff0c;\u5373\u201c\u662f\u600e\u4e48\u753b\u7684\u201d\u3002 \u6211\u4eec\u7684\u76ee\u6807\u662f&#xff0c;\u751f\u6210\u4e00\u5f20\u65b0\u7684\u56fe\u50cf&#xff0c;\u5b83\u5728\u201c\u5185\u5bb9\u201d\u4e0a\u63a5\u8fd1\u6211\u4eec\u7684\u5185\u5bb9\u56fe&#xff0c;\u5728\u201c\u98ce\u683c\u201d\u4e0a\u63a5\u8fd1\u6211\u4eec\u7684\u98ce\u683c\u56fe\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>11.3.2 VGG\u7f51\u7edc\u7684\u201c\u795e\u542f\u201d&#xff1a;\u4e3a\u4f55\u5b83\u80fd\u611f\u77e5\u5185\u5bb9\u4e0e\u98ce\u683c&#xff1f;<\/h5>\n<p>\u7b97\u6cd5\u7684\u5b9e\u73b0&#xff0c;\u5de7\u5999\u5730\u5229\u7528\u4e86\u4e00\u4e2a\u5728ImageNet\u4e0a\u9884\u8bad\u7ec3\u597d\u7684VGG\u7f51\u7edc&#xff08;\u901a\u5e38\u662fVGG16\u6216VGG19&#xff09;\u3002\u7814\u7a76\u8005\u53d1\u73b0&#xff0c;VGG\u7f51\u7edc\u7684\u4e0d\u540c\u5c42\u7ea7&#xff0c;\u5929\u7136\u5730\u5b66\u4e60\u5230\u4e86\u5bf9\u5185\u5bb9\u548c\u98ce\u683c\u7684\u4e0d\u540c\u654f\u611f\u5ea6\u3002<\/p>\n<ul>\n<li>\n<p>\u5185\u5bb9\u8868\u793a \u7f51\u7edc\u7684\u6df1\u5c42&#xff08;\u9760\u540e&#xff09;\u7279\u5f81\u56fe&#xff0c;\u7531\u4e8e\u611f\u53d7\u91ce\u66f4\u5927&#xff0c;\u6355\u6349\u5230\u7684\u662f\u66f4\u5168\u5c40\u3001\u66f4\u62bd\u8c61\u7684\u4fe1\u606f&#xff0c;\u6bd4\u5982\u7269\u4f53\u7684\u8f6e\u5ed3\u548c\u7ec4\u5408\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528VGG\u6df1\u5c42\u7684\u6fc0\u6d3b\u503c&#xff0c;\u6765\u4ee3\u8868\u4e00\u5f20\u56fe\u50cf\u7684\u5185\u5bb9\u3002<\/p>\n<\/li>\n<li>\n<p>\u98ce\u683c\u8868\u793a \u7f51\u7edc\u7684\u6d45\u5c42\u5230\u6df1\u5c42\u7684\u7279\u5f81\u56fe&#xff0c;\u90fd\u5305\u542b\u4e86\u4e30\u5bcc\u7684\u7eb9\u7406\u548c\u6a21\u5f0f\u4fe1\u606f\u3002\u98ce\u683c&#xff0c;\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u8fd9\u4e9b\u7279\u5f81\u5728\u7a7a\u95f4\u4e0a\u7684\u76f8\u5173\u6027\u3002\u5982\u4f55\u8861\u91cf\u8fd9\u79cd\u76f8\u5173\u6027&#xff1f;\u7b54\u6848\u662f\u683c\u62c9\u59c6\u77e9\u9635&#xff08;Gram Matrix&#xff09;\u3002<\/p>\n<ul>\n<li>\u683c\u62c9\u59c6\u77e9\u9635&#xff1a;\u5bf9\u4e8e\u4e00\u4e2a\u5377\u79ef\u5c42\u7684\u8f93\u51fa\u7279\u5f81\u56fe&#xff0c;\u6211\u4eec\u8ba1\u7b97\u5176\u4e0d\u540c\u901a\u9053&#xff08;feature map&#xff09;\u4e4b\u95f4\u7684\u5185\u79ef\u3002\u8fd9\u4e2a\u77e9\u9635\u6355\u6349\u4e86\u54ea\u4e9b\u7279\u5f81\u503e\u5411\u4e8e\u201c\u540c\u65f6\u51fa\u73b0\u201d&#xff0c;\u4ece\u800c\u91cf\u5316\u4e86\u56fe\u50cf\u7684\u7eb9\u7406\u3001\u7b14\u89e6\u548c\u8272\u5f69\u6a21\u5f0f&#xff0c;\u6210\u4e3a\u4e86\u98ce\u683c\u7684\u7edd\u4f73\u201c\u6307\u7eb9\u201d\u3002\u6211\u4eec\u53ef\u4ee5\u5229\u7528VGG\u591a\u4e2a\u5c42&#xff08;\u4ece\u6d45\u5230\u6df1&#xff09;\u7684\u683c\u62c9\u59c6\u77e9\u9635&#xff0c;\u6765\u5171\u540c\u5b9a\u4e49\u4e00\u5f20\u56fe\u50cf\u7684\u98ce\u683c\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>11.3.3 \u635f\u5931\u51fd\u6570\u7684\u8bbe\u8ba1&#xff1a;\u5185\u5bb9\u3001\u98ce\u683c\u4e0e\u6b63\u5219\u5316\u7684\u4e09\u91cd\u594f<\/h5>\n<p>\u6211\u4eec\u7684\u76ee\u6807\u662f\u751f\u6210\u4e00\u5f20\u65b0\u56fe\u50cf&#xff0c;\u8fd9\u5f20\u56fe\u50cf\u9700\u8981\u540c\u65f6\u6ee1\u8db3\u4e09\u4e2a\u6761\u4ef6&#xff0c;\u56e0\u6b64&#xff0c;\u6211\u4eec\u7684\u603b\u635f\u5931\u51fd\u6570\u4e5f\u7531\u4e09\u90e8\u5206\u6784\u6210&#xff1a;<\/p>\n<li>\n<p>\u5185\u5bb9\u635f\u5931&#xff08;Content Loss&#xff09; \u8ba1\u7b97\u751f\u6210\u56fe\u50cf\u5728VGG\u67d0\u4e2a\u6df1\u5c42&#xff08;\u5982block5_conv2&#xff09;\u7684\u7279\u5f81\u56fe&#xff0c;\u4e0e\u5185\u5bb9\u56fe\u50cf\u5728\u8be5\u5c42\u7684\u7279\u5f81\u56fe\u4e4b\u95f4\u7684\u5747\u65b9\u8bef\u5dee\u3002\u8fd9\u4e2a\u635f\u5931\u9a71\u4f7f\u751f\u6210\u56fe\u50cf\u5728\u5b8f\u89c2\u5185\u5bb9\u4e0a\u5411\u5185\u5bb9\u56fe\u770b\u9f50\u3002<\/p>\n<\/li>\n<li>\n<p>\u98ce\u683c\u635f\u5931&#xff08;Style Loss&#xff09; \u8ba1\u7b97\u751f\u6210\u56fe\u50cf\u5728VGG\u591a\u4e2a\u5c42&#xff08;\u5982block1_conv1, block2_conv1, &#8230;&#xff09;\u7684\u683c\u62c9\u59c6\u77e9\u9635&#xff0c;\u4e0e\u98ce\u683c\u56fe\u50cf\u5728\u5bf9\u5e94\u5c42\u7684\u683c\u62c9\u59c6\u77e9\u9635\u4e4b\u95f4\u7684\u5747\u65b9\u8bef\u5dee\u4e4b\u548c\u3002\u8fd9\u4e2a\u635f\u5931\u9a71\u4f7f\u751f\u6210\u56fe\u50cf\u5b66\u4e60\u98ce\u683c\u56fe\u7684\u7eb9\u7406\u548c\u7b14\u89e6\u3002<\/p>\n<\/li>\n<li>\n<p>\u603b\u53d8\u5206\u635f\u5931&#xff08;Total Variation Loss&#xff09; \u8fd9\u662f\u4e00\u4e2a\u53ef\u9009\u7684\u6b63\u5219\u5316\u9879\u3002\u5b83\u8ba1\u7b97\u751f\u6210\u56fe\u50cf\u76f8\u90bb\u50cf\u7d20\u4e4b\u95f4\u7684\u5dee\u5f02\u3002\u6700\u5c0f\u5316\u8fd9\u4e2a\u635f\u5931&#xff0c;\u53ef\u4ee5\u4f7f\u5f97\u751f\u6210\u7684\u56fe\u50cf\u66f4\u52a0\u5e73\u6ed1\u3001\u81ea\u7136&#xff0c;\u51cf\u5c11\u6742\u4e71\u7684\u566a\u70b9\u3002<\/p>\n<\/li>\n<p>\u603b\u635f\u5931 &#061; \u03b1 * \u5185\u5bb9\u635f\u5931 &#043; \u03b2 * \u98ce\u683c\u635f\u5931 &#043; \u03b3 * \u603b\u53d8\u5206\u635f\u5931 \u5176\u4e2d&#xff0c;\u03b1, \u03b2, \u03b3\u662f\u8d85\u53c2\u6570&#xff0c;\u7528\u6765\u6743\u8861\u4e09\u8005\u4e4b\u95f4\u7684\u91cd\u8981\u6027\u3002<\/p>\n<h5>11.3.4 \u7f16\u7801\u5b9e\u73b0&#xff1a;\u4e00\u6b65\u6b65\u521b\u9020\u827a\u672f<\/h5>\n<p>\u8fd9\u4e2a\u7b97\u6cd5\u7684\u5b9e\u73b0\u8fc7\u7a0b\u4e0e\u6211\u4eec\u4e4b\u524d\u7684\u9879\u76ee\u4e0d\u540c\u3002\u6211\u4eec\u4e0d\u8bad\u7ec3\u7f51\u7edc&#xff0c;VGG\u7684\u6743\u91cd\u662f\u56fa\u5b9a\u4e0d\u53d8\u7684\u3002\u6211\u4eec\u4f18\u5316\u7684\u5bf9\u8c61&#xff0c;\u662f\u751f\u6210\u56fe\u50cf\u672c\u8eab\u7684\u50cf\u7d20\u503c\u3002<\/p>\n<li>\u52a0\u8f7d\u9884\u8bad\u7ec3\u7684VGG\u6a21\u578b&#xff0c;\u5e76\u6784\u5efa\u4e00\u4e2a\u65b0\u7684\u6a21\u578b&#xff0c;\u5176\u8f93\u5165\u662f\u56fe\u50cf&#xff0c;\u8f93\u51fa\u662f\u6211\u4eec\u611f\u5174\u8da3\u7684\u90a3\u4e9b\u5185\u5bb9\u5c42\u548c\u98ce\u683c\u5c42\u7684\u6fc0\u6d3b\u503c\u3002<\/li>\n<li>\u5b9a\u4e49\u5185\u5bb9\u548c\u98ce\u683c\u635f\u5931\u51fd\u6570&#xff0c;\u4ee5\u53ca\u603b\u53d8\u5206\u635f\u5931\u51fd\u6570\u3002<\/li>\n<li>\u8bbe\u7f6e\u4f18\u5316\u5faa\u73af&#xff1a;\n<ul>\n<li>\u5c06\u751f\u6210\u56fe\u50cf\u521d\u59cb\u5316\u4e3a\u4e00\u5f20\u767d\u566a\u58f0\u56fe&#xff0c;\u6216\u8005\u76f4\u63a5\u4ece\u5185\u5bb9\u56fe\u5f00\u59cb\u3002<\/li>\n<li>\u5c06\u5176\u5b9a\u4e49\u4e3a\u4e00\u4e2a\u53ef\u8bad\u7ec3\u7684\u53d8\u91cf&#xff08;tf.Variable&#xff09;\u3002<\/li>\n<li>\u5728\u6bcf\u4e00\u6b21\u8fed\u4ee3\u4e2d&#xff1a; a. \u8ba1\u7b97\u603b\u635f\u5931\u3002 b. \u6839\u636e\u603b\u635f\u5931&#xff0c;\u8ba1\u7b97\u751f\u6210\u56fe\u50cf\u50cf\u7d20\u503c\u7684\u68af\u5ea6\u3002 c. \u4f7f\u7528\u4e00\u4e2a\u4f18\u5316\u5668&#xff08;\u5982Adam&#xff09;&#xff0c;\u6839\u636e\u68af\u5ea6\u66f4\u65b0\u751f\u6210\u56fe\u50cf\u7684\u50cf\u7d20\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u91cd\u590d\u4e0a\u4e00\u6b65\u6570\u767e\u4e0a\u5343\u6b21&#xff0c;\u76f4\u5230\u751f\u6210\u7684\u56fe\u50cf\u5728\u89c6\u89c9\u4e0a\u4ee4\u4eba\u6ee1\u610f\u4e3a\u6b62\u3002<\/li>\n<p>\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u5c31\u50cf\u4e00\u4f4d\u753b\u5bb6\u5728\u753b\u5e03\u4e0a\u53cd\u590d\u4fee\u6539&#xff0c;\u6bcf\u4e00\u6b21\u4e0b\u7b14&#xff0c;\u90fd\u8bd5\u56fe\u8ba9\u753b\u4f5c\u66f4\u63a5\u8fd1\u5185\u5bb9\u8349\u7a3f&#xff0c;\u540c\u65f6\u53c8\u66f4\u7b26\u5408\u98ce\u683c\u7684\u97f5\u5473\u3002<\/p>\n<h5>11.3.5 \u6210\u679c\u5c55\u793a\u4e0e\u5ef6\u4f38\u601d\u8003<\/h5>\n<ul>\n<li>\n<p>\u5c55\u793a\u4e0e\u63a2\u7d22&#xff1a;\u60a8\u53ef\u4ee5\u5c1d\u8bd5\u5404\u79cd\u5404\u6837\u7684\u5185\u5bb9\u56fe\u548c\u98ce\u683c\u56fe\u7ec4\u5408&#xff0c;\u6bd4\u5982\u5c06\u60a8\u7684\u81ea\u62cd\u7167\u4e0e\u6bd5\u52a0\u7d22\u7684\u7acb\u4f53\u4e3b\u4e49\u98ce\u683c\u7ed3\u5408&#xff0c;\u5c06\u57ce\u5e02\u98ce\u666f\u4e0e\u65e5\u672c\u6d6e\u4e16\u7ed8\u7684\u98ce\u683c\u7ed3\u5408&#xff0c;\u5176\u7ed3\u679c\u5f80\u5f80\u5145\u6ee1\u60ca\u559c\u3002\u8c03\u6574\u5185\u5bb9\u548c\u98ce\u683c\u635f\u5931\u7684\u6743\u91cd&#xff08;\u03b1\/\u03b2\u6bd4\u503c&#xff09;&#xff0c;\u53ef\u4ee5\u63a7\u5236\u751f\u6210\u56fe\u50cf\u66f4\u504f\u5411\u5185\u5bb9\u8fd8\u662f\u66f4\u504f\u5411\u98ce\u683c\u3002<\/p>\n<\/li>\n<li>\n<p>\u5ef6\u4f38\u601d\u8003&#xff1a;<\/p>\n<ul>\n<li>\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb&#xff1a;\u6211\u4eec\u5b9e\u73b0\u7684\u8fd9\u4e2a\u57fa\u4e8e\u4f18\u5316\u7684\u65b9\u6cd5\u901f\u5ea6\u5f88\u6162&#xff0c;\u751f\u6210\u4e00\u5f20\u56fe\u53ef\u80fd\u9700\u8981\u51e0\u5206\u949f\u3002\u4e1a\u754c\u540e\u6765\u53d1\u5c55\u51fa\u4e86\u57fa\u4e8e\u201c\u751f\u6210\u5668\u7f51\u7edc\u201d\u7684\u5feb\u901f\u98ce\u683c\u8fc1\u79fb&#xff08;Fast Style Transfer&#xff09;&#xff0c;\u5b83\u901a\u8fc7\u8bad\u7ec3\u4e00\u4e2a\u524d\u9988\u7f51\u7edc\u6765\u76f4\u63a5\u751f\u6210\u98ce\u683c\u5316\u7684\u56fe\u50cf&#xff0c;\u53ef\u4ee5\u505a\u5230\u5b9e\u65f6\u5904\u7406&#xff0c;\u4f46\u4e00\u4e2a\u6a21\u578b\u53ea\u80fd\u5b66\u4e60\u4e00\u79cd\u98ce\u683c\u3002<\/li>\n<li>AI\u521b\u9020\u529b\u7684\u8fb9\u754c&#xff1a;\u795e\u7ecf\u98ce\u683c\u8fc1\u79fb\u5f15\u53d1\u4e86\u4e00\u4e2a\u6df1\u523b\u7684\u54f2\u5b66\u95ee\u9898\u2014\u2014\u8fd9\u7a76\u7adf\u662f\u201c\u6a21\u4eff\u201d\u8fd8\u662f\u201c\u521b\u4f5c\u201d&#xff1f;\u673a\u5668\u662f\u5426\u62e5\u6709\u4e86\u201c\u5ba1\u7f8e\u201d&#xff1f;\u5b83\u751f\u6210\u7684\u4f5c\u54c1&#xff0c;\u5176\u827a\u672f\u4ef7\u503c\u4f55\u5728&#xff1f;\u8fd9\u4e9b\u95ee\u9898\u6ca1\u6709\u6807\u51c6\u7b54\u6848&#xff0c;\u4f46\u5b83\u4eec\u4fc3\u4f7f\u6211\u4eec\u53bb\u601d\u8003\u6280\u672f\u4e0e\u4eba\u6587\u3001\u667a\u80fd\u4e0e\u521b\u9020\u529b\u4e4b\u95f4\u590d\u6742\u800c\u8ff7\u4eba\u7684\u5173\u7cfb\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<p>\u5c0f\u7ed3&#xff1a;\u884c\u662f\u77e5\u4e4b\u6210<\/p>\n<p>\u672c\u7ae0\u7684\u65c5\u7a0b\u5373\u5c06\u7ed3\u675f\u3002\u56de\u9996\u671b\u53bb&#xff0c;\u6211\u4eec\u4e0d\u4ec5\u4ec5\u662f\u7f16\u5199\u4e86\u4e09\u6bb5\u4ee3\u7801&#xff0c;\u66f4\u662f\u4eb2\u5386\u4e86\u4e09\u6b21\u5b8c\u6574\u7684\u521b\u9020\u4e4b\u65c5\u3002<\/p>\n<ul>\n<li>\u6211\u4eec\u4ece\u5783\u573e\u5206\u7c7b\u5f00\u59cb&#xff0c;\u5c06\u57fa\u7840\u7684\u56fe\u50cf\u5206\u7c7b\u6280\u672f\u4e0e\u8fc1\u79fb\u5b66\u4e60\u7684\u667a\u6167\u76f8\u7ed3\u5408&#xff0c;\u4e3a\u4e00\u4e2a\u73b0\u5b9e\u7684\u73af\u4fdd\u95ee\u9898\u63d0\u4f9b\u4e86\u53ef\u884c\u7684\u89e3\u51b3\u65b9\u6848\u3002\u6211\u4eec\u5b66\u4f1a\u4e86\u6570\u636e\u5904\u7406\u3001\u6a21\u578b\u5fae\u8c03\u3001\u6027\u80fd\u8bc4\u4f30\u7684\u5168\u6d41\u7a0b\u601d\u7ef4\u3002<\/li>\n<li>\u63a5\u7740&#xff0c;\u6211\u4eec\u6311\u6218\u4e86\u76ee\u6807\u68c0\u6d4b&#xff0c;\u9a7e\u9a6d\u4e86\u5f3a\u5927\u7684YOLO\u6a21\u578b&#xff0c;\u8ba9\u673a\u5668\u4e0d\u4ec5\u80fd\u201c\u770b\u61c2\u201d&#xff0c;\u66f4\u80fd\u201c\u5b9a\u4f4d\u201d\u3002\u6211\u4eec\u4f53\u4f1a\u5230\u4e86\u4ece\u5b66\u672f\u7406\u8bba\u5230\u9ad8\u6548\u5de5\u4e1a\u7ea7\u6846\u67b6\u7684\u5de8\u5927\u98de\u8dc3&#xff0c;\u5e76\u7aa5\u89c1\u4e86\u5176\u5728\u5b89\u9632\u3001\u81ea\u52a8\u9a7e\u9a76\u7b49\u9886\u57df\u7684\u65e0\u9650\u6f5c\u529b\u3002<\/li>\n<li>\u6700\u540e&#xff0c;\u6211\u4eec\u6c89\u6d78\u5728\u795e\u7ecf\u98ce\u683c\u8fc1\u79fb\u7684\u827a\u672f\u4e16\u754c\u91cc&#xff0c;\u5229\u7528VGG\u7f51\u7edc\u5bf9\u5185\u5bb9\u548c\u98ce\u683c\u7684\u6df1\u523b\u7406\u89e3&#xff0c;\u8ba9\u6280\u672f\u4e0e\u7f8e\u5b66\u5171\u821e&#xff0c;\u521b\u9020\u51fa\u524d\u6240\u672a\u6709\u7684\u89c6\u89c9\u4f53\u9a8c\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e09\u4e2a\u9879\u76ee&#xff0c;\u5982\u4e09\u9762\u68f1\u955c&#xff0c;\u6298\u5c04\u51fa\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u4e0d\u540c\u7ef4\u5ea6\u7684\u5149\u5f69\u3002\u5b83\u4eec\u544a\u8bc9\u6211\u4eec&#xff0c;\u884c\u662f\u77e5\u4e4b\u6210\u2014\u2014\u5b9e\u8df5&#xff0c;\u662f\u77e5\u8bc6\u6700\u7ec8\u7684\u5f62\u6001&#xff0c;\u662f\u7406\u8bba\u4ef7\u503c\u7684\u6700\u7ec8\u5f52\u5bbf\u3002\u771f\u6b63\u7684\u5b66\u4e60&#xff0c;\u4ece\u6765\u4e0d\u662f\u53d1\u751f\u5728\u5b89\u9038\u7684\u9605\u8bfb\u4e2d&#xff0c;\u800c\u662f\u53d1\u751f\u5728\u90a3\u4e9b\u60a8\u4e3a\u4e86\u89e3\u51b3\u4e00\u4e2a\u771f\u5b9e\u95ee\u9898\u800c\u5f7b\u591c\u8c03\u8bd5\u3001\u82e6\u82e6\u601d\u7d22\u3001\u6700\u7ec8\u8c41\u7136\u5f00\u6717\u7684\u77ac\u95f4\u3002<\/p>\n<p>\u4ee5\u672c\u7ae0\u4e3a\u8d77\u70b9&#xff0c;\u613f\u60a8\u5e26\u7740\u8fd9\u4efd\u4eb2\u624b\u5b9e\u8df5\u8fc7\u7684\u81ea\u4fe1\u4e0e\u4f53\u609f&#xff0c;\u53bb\u63a2\u7d22\u66f4\u5e7f\u9614\u7684\u5e94\u7528\u5929\u5730\u3002\u53bb\u53d1\u73b0\u95ee\u9898&#xff0c;\u53bb\u5b9a\u4e49\u9879\u76ee&#xff0c;\u53bb\u6536\u96c6\u6570\u636e&#xff0c;\u53bb\u521b\u9020\u6a21\u578b\u3002\u7528\u60a8\u624b\u4e2d\u7684\u4ee3\u7801\u548c\u5fc3\u4e2d\u7684\u667a\u6167&#xff0c;\u53bb\u6d1e\u89c1\u4e07\u7269&#xff0c;\u53bb\u6539\u53d8\u4e16\u754c&#xff0c;\u54ea\u6015\u53ea\u662f\u6539\u53d8\u4e16\u754c\u7684\u4e00\u4e2a\u5c0f\u5c0f\u89d2\u843d&#xff0c;\u90a3\u4e5f\u662f\u60a8\u4f5c\u4e3a\u4e00\u540d\u6280\u672f\u521b\u9020\u8005&#xff0c;\u7559\u7ed9\u8fd9\u4e2a\u65f6\u4ee3\u6700\u7f8e\u7684\u5370\u8bb0\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u5341\u4e8c\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u81ea\u7136\u8bed\u8a00\u5904\u7406<\/h3>\n<p>\u4ece\u50cf\u7d20\u5230\u8bed\u7d20&#xff0c;\u5f00\u542f\u65b0\u7684\u63a2\u7d22<\/p>\n<p>\u5728\u4e0a\u4e00\u7ae0\u7684\u65c5\u7a0b\u4e2d&#xff0c;\u6211\u4eec\u8d4b\u4e88\u4e86\u673a\u5668\u4e00\u53cc\u201c\u667a\u6167\u4e4b\u773c\u201d&#xff0c;\u8ba9\u5b83\u80fd\u591f\u6d1e\u5bdf\u4e94\u5f69\u6591\u6593\u7684\u89c6\u89c9\u4e16\u754c\u3002\u6211\u4eec\u4ece\u50cf\u7d20\u7684\u6d77\u6d0b\u4e2d&#xff0c;\u8bc6\u522b\u51fa\u4e86\u7269\u4f53&#xff0c;\u5b9a\u4f4d\u4e86\u5b83\u4eec\u7684\u5b58\u5728&#xff0c;\u751a\u81f3\u6a21\u4eff\u4e86\u827a\u672f\u5bb6\u7684\u7b14\u89e6\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u5c06\u8e0f\u5165\u4e00\u4e2a\u540c\u6837\u6df1\u9083&#xff0c;\u751a\u81f3\u66f4\u4e3a\u590d\u6742\u7684\u9886\u57df\u2014\u2014\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;Natural Language Processing, NLP&#xff09;\u3002\u6211\u4eec\u5c06\u5f15\u5bfc\u673a\u5668&#xff0c;\u4ece\u201c\u770b\u201d\u4e16\u754c&#xff0c;\u8f6c\u5411\u201c\u542c\u201d\u61c2\u4e16\u754c\u3001\u201c\u8bf4\u201d\u51fa\u667a\u6167\u3002<\/p>\n<p>\u8bed\u8a00&#xff0c;\u662f\u4eba\u7c7b\u601d\u60f3\u7684\u5c45\u6240&#xff0c;\u662f\u6587\u660e\u4f20\u627f\u7684\u8f7d\u4f53\u3002\u5b83\u5e76\u975e\u50cf\u56fe\u50cf\u50cf\u7d20\u90a3\u6837\u89c4\u6574\u3001\u76f4\u767d\u3002\u8bed\u8a00\u5145\u6ee1\u4e86\u7cbe\u5999\u7684\u6b67\u4e49\u6027&#xff08;\u201c\u82f9\u679c\u201d\u53ef\u4ee5\u662f\u4e00\u79cd\u6c34\u679c&#xff0c;\u4e5f\u53ef\u4ee5\u662f\u4e00\u4e2a\u79d1\u6280\u516c\u53f8&#xff09;\u3001\u6df1\u523b\u7684\u4e0a\u4e0b\u6587\u4f9d\u8d56\u6027&#xff08;\u201c\u6211\u4e0d\u60f3\u53bb\u4e86\u201d\u7684\u771f\u6b63\u542b\u4e49&#xff0c;\u53d6\u51b3\u4e8e\u4e4b\u524d\u7684\u5bf9\u8bdd&#xff09;&#xff0c;\u4ee5\u53ca\u4e30\u5bcc\u7684\u6587\u5316\u4e0e\u60c5\u611f\u5185\u6db5\u3002\u8fd9\u4f7f\u5f97NLP\u6210\u4e3a\u4eba\u5de5\u667a\u80fd\u9886\u57df\u6700\u5177\u6311\u6218\u3001\u4e5f\u6700\u5177\u9b45\u529b\u7684\u5206\u652f\u4e4b\u4e00\u3002<\/p>\n<p>\u672c\u7ae0&#xff0c;\u6211\u4eec\u5c06\u5316\u8eab\u4e3a\u8bed\u8a00\u7684\u5de5\u7a0b\u5e08\u4e0e\u827a\u672f\u5bb6\u3002\u6211\u4eec\u5c06\u901a\u8fc7\u4e09\u4e2a\u5faa\u5e8f\u6e10\u8fdb\u7684\u6838\u5fc3\u9879\u76ee&#xff0c;\u4eb2\u624b\u5b9e\u8df5\u5982\u4f55\u5c06\u975e\u7ed3\u6784\u5316\u7684\u6587\u672c&#xff0c;\u8f6c\u5316\u4e3a\u673a\u5668\u53ef\u4ee5\u8ba1\u7b97\u548c\u7406\u89e3\u7684\u6570\u5b66\u8868\u793a\u2014\u2014\u5411\u91cf\u3002\u6211\u4eec\u5c06\u8fd0\u7528\u5728\u7b2c\u516b\u7ae0\u548c\u7b2c\u4e5d\u7ae0\u5b66\u5230\u7684\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN&#xff09;\u3001\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc&#xff08;LSTM&#xff09;\u4e43\u81f3\u66f4\u5f3a\u5927\u7684\u9884\u8bad\u7ec3\u6a21\u578b&#xff0c;\u53bb\u5b8c\u6210\u4e09\u9879\u8ff7\u4eba\u7684\u4efb\u52a1&#xff1a;<\/p>\n<li>\u6d1e\u5bdf\u60c5\u611f&#xff1a;\u6211\u4eec\u5c06\u6784\u5efa\u4e00\u4e2a\u60c5\u611f\u5206\u6790\u7cfb\u7edf&#xff0c;\u8ba9\u673a\u5668\u80fd\u591f\u8bfb\u61c2\u7535\u5f71\u8bc4\u8bba\u6587\u5b57\u80cc\u540e\u7684\u559c\u6012\u54c0\u4e50\u3002<\/li>\n<li>\u8de8\u8d8a\u969c\u788d&#xff1a;\u6211\u4eec\u5c06\u642d\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u673a\u5668\u7ffb\u8bd1\u6a21\u578b&#xff0c;\u5c1d\u8bd5\u53bb\u6253\u7834\u8bed\u8a00\u4e4b\u95f4\u7684\u58c1\u5792&#xff0c;\u6784\u5efa\u6c9f\u901a\u7684\u6865\u6881\u3002<\/li>\n<li>\u6784\u5efa\u4f19\u4f34&#xff1a;\u6211\u4eec\u5c06\u5b9e\u73b0\u4e00\u4e2a\u667a\u80fd\u95ee\u7b54\u673a\u5668\u4eba&#xff0c;\u8d4b\u4e88\u5b83\u57fa\u4e8e\u77e5\u8bc6\u8fdb\u884c\u903b\u8f91\u95ee\u7b54\u7684\u80fd\u529b\u3002<\/li>\n<p>\u8fd9\u8d9f\u65c5\u7a0b&#xff0c;\u4e0d\u4ec5\u662f\u4ee3\u7801\u4e0e\u7b97\u6cd5\u7684\u5b9e\u8df5&#xff0c;\u66f4\u662f\u5bf9\u4eba\u7c7b\u667a\u6167\u6700\u4f1f\u5927\u7ed3\u6676\u2014\u2014\u8bed\u8a00\u2014\u2014\u7684\u4e00\u6b21\u6df1\u5ea6\u89e3\u7801\u3002\u51c6\u5907\u597d&#xff0c;\u8ba9\u6211\u4eec\u4e00\u8d77\u8046\u542c\u8bed\u8a00\u7684\u8109\u640f&#xff0c;\u4e0e\u6587\u5b57\u80cc\u540e\u7684\u667a\u6167\u540c\u884c\u3002<\/p>\n<hr \/>\n<h4>12.1 \u6587\u672c\u60c5\u611f\u5206\u6790&#xff1a;\u5206\u6790\u7535\u5f71\u8bc4\u8bba\u7684\u60c5\u611f\u503e\u5411 \u2014\u2014 \u6355\u6349\u6587\u5b57\u7684\u6e29\u5ea6<\/h4>\n<p>\u8bed\u8a00\u4e0d\u4ec5\u4ec5\u662f\u4fe1\u606f\u7684\u8f7d\u4f53&#xff0c;\u66f4\u662f\u60c5\u611f\u7684\u6d41\u9732\u3002\u201c\u8fd9\u90e8\u7535\u5f71\u592a\u68d2\u4e86&#xff01;\u201d\u548c\u201c\u8fd9\u90e8\u7535\u5f71\u771f\u662f\u6d6a\u8d39\u65f6\u95f4\u201d&#xff0c;\u8fd9\u4e24\u53e5\u8bdd\u627f\u8f7d\u7684\u4fe1\u606f\u622a\u7136\u4e0d\u540c&#xff0c;\u5176\u6838\u5fc3\u5dee\u5f02\u5c31\u5728\u4e8e\u60c5\u611f\u7684\u6781\u6027\u3002\u8ba9\u673a\u5668\u5b66\u4f1a\u5206\u8fa8\u8fd9\u79cd\u60c5\u611f&#xff0c;\u5c31\u662f\u60c5\u611f\u5206\u6790&#xff08;Sentiment Analysis&#xff09;\u7684\u6838\u5fc3\u4efb\u52a1\u3002<\/p>\n<h5>12.1.1 \u9879\u76ee\u6982\u8ff0&#xff1a;\u4e3a\u4f55\u60c5\u611f\u8ba1\u7b97\u5982\u6b64\u91cd\u8981&#xff1f;<\/h5>\n<ul>\n<li>\n<p>\u5e94\u7528\u573a\u666f \u60c5\u611f\u8ba1\u7b97\u662f\u4eba\u5de5\u667a\u80fd\u5546\u4e1a\u5e94\u7528\u4e2d\u6700\u5e7f\u6cdb\u3001\u6700\u6210\u719f\u7684\u6280\u672f\u4e4b\u4e00\u3002\u5b83\u5c31\u50cf\u4e00\u628a\u80fd\u63a2\u77e5\u7fa4\u4f53\u60c5\u7eea\u7684\u201c\u8109\u640f\u8ba1\u201d&#xff0c;\u5176\u4ef7\u503c\u65e0\u5904\u4e0d\u5728&#xff1a;<\/p>\n<ul>\n<li>\u5546\u4e1a\u667a\u80fd&#xff1a;\u4f01\u4e1a\u53ef\u4ee5\u81ea\u52a8\u5206\u6790\u6d77\u91cf\u7684\u7535\u5546\u8bc4\u8bba\u3001\u793e\u4ea4\u5a92\u4f53\u5e16\u5b50&#xff0c;\u5feb\u901f\u4e86\u89e3\u7528\u6237\u5bf9\u65b0\u4ea7\u54c1\u7684\u53cd\u9988&#xff0c;\u6d1e\u5bdf\u54c1\u724c\u53e3\u7891\u3002<\/li>\n<li>\u8206\u60c5\u76d1\u63a7&#xff1a;\u653f\u5e9c\u6216\u673a\u6784\u53ef\u4ee5\u5b9e\u65f6\u8ddf\u8e2a\u516c\u4f17\u5bf9\u67d0\u4e00\u653f\u7b56\u6216\u4e8b\u4ef6\u7684\u60c5\u611f\u503e\u5411&#xff0c;\u53ca\u65f6\u505a\u51fa\u53cd\u5e94\u3002<\/li>\n<li>\u4e2a\u6027\u5316\u63a8\u8350&#xff1a;\u63a8\u8350\u7cfb\u7edf\u53ef\u4ee5\u6839\u636e\u60a8\u5bf9\u67d0\u4e9b\u7535\u5f71\u7684\u201c\u6b63\u9762\u201d\u8bc4\u4ef7&#xff0c;\u4e3a\u60a8\u63a8\u8350\u98ce\u683c\u7c7b\u4f3c\u7684\u597d\u8bc4\u7535\u5f71\u3002<\/li>\n<li>\u91d1\u878d\u5e02\u573a&#xff1a;\u901a\u8fc7\u5206\u6790\u8d22\u7ecf\u65b0\u95fb\u548c\u793e\u4ea4\u5a92\u4f53\u4e0a\u7684\u5e02\u573a\u60c5\u7eea&#xff0c;\u8f85\u52a9\u8fdb\u884c\u91cf\u5316\u4ea4\u6613\u51b3\u7b56\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4efb\u52a1\u5b9a\u4e49 \u6211\u4eec\u5c06\u8fd9\u4e2a\u4efb\u52a1\u7b80\u5316\u4e3a\u4e00\u4e2a\u7ecf\u5178\u7684\u4e8c\u5143\u5206\u7c7b\u95ee\u9898&#xff1a;\u7ed9\u5b9a\u4e00\u6bb5\u6587\u672c&#xff08;\u5728\u6211\u4eec\u7684\u9879\u76ee\u4e2d&#xff0c;\u662f\u4e00\u6bb5IMDb\u7684\u7535\u5f71\u8bc4\u8bba&#xff09;&#xff0c;\u6a21\u578b\u9700\u8981\u8f93\u51fa\u5176\u60c5\u611f\u503e\u5411\u662f\u201c\u6b63\u9762\u201d&#xff08;Positive&#xff09;\u8fd8\u662f\u201c\u8d1f\u9762\u201d&#xff08;Negative&#xff09;\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>12.1.2 \u6587\u672c\u7684\u201c\u6570\u5b57\u5316\u201d\u4e4b\u65c5&#xff1a;\u4ece\u8bcd\u8bed\u5230\u5411\u91cf<\/h5>\n<p>\u673a\u5668\u4e0d\u7406\u89e3\u6587\u5b57&#xff0c;\u53ea\u7406\u89e3\u6570\u5b57\u3002\u56e0\u6b64&#xff0c;\u5728\u6784\u5efa\u6a21\u578b\u4e4b\u524d&#xff0c;\u6211\u4eec\u5fc5\u987b\u5c06\u5343\u53d8\u4e07\u5316\u7684\u6587\u672c&#xff0c;\u8f6c\u5316\u4e3a\u89c4\u6574\u3001\u53ef\u8ba1\u7b97\u7684\u6570\u5b66\u5f62\u5f0f\u3002\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u662f\u6240\u6709NLP\u4efb\u52a1\u7684\u57fa\u77f3\u3002<\/p>\n<ul>\n<li>\n<p>\u6587\u672c\u9884\u5904\u7406 \u539f\u59cb\u7684\u6587\u672c\u662f\u201c\u7c97\u7cd9\u201d\u7684&#xff0c;\u5145\u6ee1\u4e86\u9700\u8981\u6211\u4eec\u6e05\u6d17\u548c\u6574\u7406\u7684\u201c\u6742\u8d28\u201d\u3002<\/p>\n<ul>\n<li>\u6587\u672c\u6e05\u6d17&#xff1a;\u9996\u5148&#xff0c;\u6211\u4eec\u9700\u8981\u53bb\u9664\u90a3\u4e9b\u5bf9\u60c5\u611f\u5206\u6790\u65e0\u7528\u7684\u4fe1\u606f&#xff0c;\u4f8b\u5982HTML\u6807\u7b7e&#xff08;\u00a0&#xff09;\u3001\u6807\u70b9\u7b26\u53f7\u3001\u6570\u5b57\u7b49\u3002\u901a\u5e38\u4e5f\u4f1a\u5c06\u6240\u6709\u6587\u672c\u8f6c\u6362\u4e3a\u5c0f\u5199&#xff0c;\u4ee5\u51cf\u5c11\u8bcd\u6c47\u8868\u7684\u5927\u5c0f\u3002<\/li>\n<li>\u5206\u8bcd&#xff08;Tokenization&#xff09;&#xff1a;\u5c06\u4e00\u4e2a\u5b8c\u6574\u7684\u53e5\u5b50&#xff0c;\u5207\u5206\u6210\u4e00\u4e2a\u8bcd\u8bed&#xff08;Token&#xff09;\u7684\u5217\u8868\u3002\u4f8b\u5982&#xff0c;&#034;This is great!&#034;\u00a0-&gt;\u00a0[&#039;this&#039;, &#039;is&#039;, &#039;great&#039;]\u3002<\/li>\n<li>\u53bb\u9664\u505c\u7528\u8bcd&#xff08;Stop Words&#xff09;&#xff1a;\u505c\u7528\u8bcd\u662f\u6307\u90a3\u4e9b\u975e\u5e38\u5e38\u89c1\u4f46\u901a\u5e38\u4e0d\u643a\u5e26\u592a\u591a\u5b9e\u9645\u610f\u4e49\u7684\u8bcd&#xff0c;\u5982&#039;a&#039;,\u00a0&#039;the&#039;,\u00a0&#039;is&#039;,\u00a0&#039;in&#039;\u7b49\u3002\u5728\u67d0\u4e9b\u4efb\u52a1\u4e2d&#xff0c;\u53bb\u9664\u5b83\u4eec\u53ef\u4ee5\u51cf\u5c11\u566a\u58f0&#xff0c;\u4f46\u5bf9\u4e8e\u60c5\u611f\u5206\u6790&#xff0c;\u6709\u65f6\u8fd9\u4e9b\u8bcd&#xff08;\u5982&#039;not&#039;&#xff09;\u4e5f\u53ef\u80fd\u5f88\u91cd\u8981&#xff0c;\u662f\u5426\u53bb\u9664\u9700\u8981\u89c6\u60c5\u51b5\u800c\u5b9a\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u8bcd\u5d4c\u5165&#xff08;Word Embeddings&#xff09;&#xff1a;\u8d85\u8d8aOne-Hot\u7f16\u7801\u7684\u667a\u6167 \u5c06\u8bcd\u8bed\u8f6c\u6362\u4e3a\u6570\u5b57&#xff0c;\u6700\u7b80\u5355\u7684\u65b9\u6cd5\u662fOne-Hot\u7f16\u7801\u3002\u4f46\u5b83\u7684\u5f0a\u7aef\u663e\u800c\u6613\u89c1&#xff1a;\u4f1a\u4ea7\u751f\u4e00\u4e2a\u6781\u5176\u5de8\u5927\u4e14\u7a00\u758f\u7684\u5411\u91cf&#xff0c;\u5e76\u4e14\u65e0\u6cd5\u4f53\u73b0\u8bcd\u4e0e\u8bcd\u4e4b\u95f4\u7684\u8bed\u4e49\u5173\u7cfb&#xff08;\u201cgood\u201d\u548c\u201cgreat\u201d\u7684\u5411\u91cf\u662f\u5b8c\u5168\u6b63\u4ea4\u7684&#xff09;\u3002<\/p>\n<p>\u8bcd\u5d4c\u5165&#xff08;Word Embeddings&#xff09;\u662f\u4e00\u79cd\u9769\u547d\u6027\u7684\u601d\u60f3\u3002\u5b83\u4e0d\u518d\u5c06\u8bcd\u8bed\u8868\u793a\u4e3a\u7a00\u758f\u7684\u9ad8\u7ef4\u5411\u91cf&#xff0c;\u800c\u662f\u5c06\u5176\u6620\u5c04&#xff08;embed&#xff09;\u5230\u4e00\u4e2a\u4f4e\u7ef4&#xff08;\u5982100\u7ef4\u3001300\u7ef4&#xff09;\u3001\u7a20\u5bc6\u7684\u5411\u91cf\u7a7a\u95f4\u4e2d\u3002\u8fd9\u4e2a\u6620\u5c04\u7684\u795e\u5947\u4e4b\u5904\u5728\u4e8e&#xff0c;\u5b83\u80fd\u6355\u6349\u8bcd\u8bed\u7684\u8bed\u4e49\u3002\u5728\u8bad\u7ec3\u597d\u7684\u8bcd\u5d4c\u5165\u7a7a\u95f4\u91cc&#xff1a;<\/p>\n<ul>\n<li>\u8bed\u4e49\u76f8\u8fd1\u7684\u8bcd&#xff0c;\u5176\u5411\u91cf\u5728\u7a7a\u95f4\u4e2d\u7684\u4f4d\u7f6e\u4e5f\u5f7c\u6b64\u9760\u8fd1\u3002\u4f8b\u5982&#xff0c;king\u7684\u5411\u91cf\u51cf\u53bbman\u7684\u5411\u91cf&#xff0c;\u52a0\u4e0awoman\u7684\u5411\u91cf&#xff0c;\u5176\u7ed3\u679c\u4f1a\u975e\u5e38\u63a5\u8fd1queen\u7684\u5411\u91cf\u3002<\/li>\n<li>\u5b83\u5c06\u8bcd\u8bed\u4ece\u4e00\u4e2a\u5b64\u7acb\u7684\u7b26\u53f7&#xff0c;\u53d8\u6210\u4e86\u4e00\u4e2a\u643a\u5e26\u4e30\u5bcc\u8bed\u4e49\u4fe1\u606f\u7684\u6570\u5b66\u5b9e\u4f53\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5b9e\u6218Keras\u7684Embedding\u5c42 \u5728Keras\u4e2d&#xff0c;\u5b9e\u73b0\u8bcd\u5d4c\u5165\u975e\u5e38\u7b80\u5355&#xff0c;\u6211\u4eec\u53ea\u9700\u8981\u4f7f\u7528Embedding\u5c42\u3002\u5b83\u5c31\u50cf\u4e00\u4e2a\u5927\u578b\u7684\u67e5\u8be2\u8868&#xff08;lookup table&#xff09;\u3002<\/p>\n<li>\u9996\u5148&#xff0c;\u6211\u4eec\u9700\u8981\u6784\u5efa\u4e00\u4e2a\u8bcd\u6c47\u8868&#xff08;Vocabulary&#xff09;&#xff0c;\u5c06\u6570\u636e\u96c6\u4e2d\u6240\u6709\u51fa\u73b0\u8fc7\u7684\u8bcd\u8bed&#xff0c;\u6309\u9891\u7387\u6392\u5e8f&#xff0c;\u5e76\u4e3a\u6bcf\u4e2a\u8bcd\u8bed\u5206\u914d\u4e00\u4e2a\u552f\u4e00\u7684\u6574\u6570\u7d22\u5f15\u3002<\/li>\n<li>\u7136\u540e&#xff0c;\u5c06\u6bcf\u6761\u6587\u672c\u8bc4\u8bba&#xff0c;\u4ece\u4e00\u4e2a\u8bcd\u8bed\u5e8f\u5217&#xff0c;\u8f6c\u6362\u4e3a\u4e00\u4e2a\u6574\u6570\u7d22\u5f15\u5e8f\u5217\u3002<\/li>\n<li>Embedding\u5c42\u63a5\u6536\u8fd9\u4e2a\u6574\u6570\u5e8f\u5217\u4f5c\u4e3a\u8f93\u5165\u3002\u5b83\u5185\u90e8\u7ef4\u62a4\u7740\u4e00\u4e2a\u6743\u91cd\u77e9\u9635&#xff0c;\u77e9\u9635\u7684\u6bcf\u4e00\u884c&#xff0c;\u5c31\u662f\u4e00\u4e2a\u8bcd\u8bed\u7684\u5d4c\u5165\u5411\u91cf\u3002\u5f53\u8f93\u5165\u4e00\u4e2a\u6574\u6570\u7d22\u5f15i\u65f6&#xff0c;\u5b83\u5c31\u53bb\u67e5\u8be2\u5e76\u8f93\u51fa\u6743\u91cd\u77e9\u9635\u7684\u7b2ci\u884c\u3002 \u8fd9\u4e2aEmbedding\u5c42\u7684\u6743\u91cd\u77e9\u9635&#xff0c;\u53ef\u4ee5\u968f\u673a\u521d\u59cb\u5316&#xff0c;\u7136\u540e\u5728\u8bad\u7ec3\u5206\u7c7b\u4efb\u52a1\u7684\u540c\u65f6&#xff0c;\u4e00\u5e76\u8fdb\u884c\u5b66\u4e60\u3002\u8fd9\u6837&#xff0c;\u6a21\u578b\u5c31\u80fd\u4e3a\u6211\u4eec\u7684\u7279\u5b9a\u4efb\u52a1&#xff0c;\u5b66\u4e60\u5230\u6700\u5408\u9002\u7684\u8bcd\u5d4c\u5165\u8868\u793a\u3002<\/li>\n<\/li>\n<\/ul>\n<p>from tensorflow.keras.layers import Embedding<br \/>\nfrom tensorflow.keras.preprocessing.text import Tokenizer<br \/>\nfrom tensorflow.keras.preprocessing.sequence import pad_sequences<\/p>\n<p># \u5047\u8bbe\u6211\u4eec\u6709\u8bc4\u8bba\u6587\u672c\u5217\u8868 a list of sentences<br \/>\nsentences &#061; [&#034;This movie is great&#034;, &#034;This movie is bad&#034;]<\/p>\n<p># 1. \u521b\u5efaTokenizer&#xff0c;\u6784\u5efa\u8bcd\u6c47\u8868<br \/>\ntokenizer &#061; Tokenizer(num_words&#061;10000, oov_token&#061;&#034;&lt;OOV&gt;&#034;) # \u53ea\u8003\u8651\u6700\u5e38\u89c1\u768410000\u4e2a\u8bcd<br \/>\ntokenizer.fit_on_texts(sentences)<br \/>\nword_index &#061; tokenizer.word_index<\/p>\n<p># 2. \u5c06\u6587\u672c\u8f6c\u6362\u4e3a\u6574\u6570\u5e8f\u5217<br \/>\nsequences &#061; tokenizer.texts_to_sequences(sentences)<\/p>\n<p># 3. \u586b\u5145\u5e8f\u5217&#xff0c;\u4f7f\u6240\u6709\u5e8f\u5217\u957f\u5ea6\u4e00\u81f4 (\u6a21\u578b\u9700\u8981\u5b9a\u957f\u8f93\u5165)<br \/>\npadded_sequences &#061; pad_sequences(sequences, maxlen&#061;120, padding&#061;&#039;post&#039;, truncating&#061;&#039;post&#039;)<\/p>\n<p># 4. \u5728\u6a21\u578b\u4e2d\u4f7f\u7528Embedding\u5c42<br \/>\n# vocab_size: \u8bcd\u6c47\u8868\u5927\u5c0f<br \/>\n# embedding_dim: \u5d4c\u5165\u5411\u91cf\u7684\u7ef4\u5ea6<br \/>\n# input_length: \u8f93\u5165\u5e8f\u5217\u7684\u957f\u5ea6<br \/>\nembedding_layer &#061; Embedding(input_dim&#061;10000, output_dim&#061;16, input_length&#061;120)<\/p>\n<h5>12.1.3 \u6784\u5efa\u60c5\u611f\u5206\u7c7b\u6a21\u578b&#xff1a;\u4ece\u7b80\u5355\u5230\u590d\u6742<\/h5>\n<p>\u6587\u672c\u6570\u636e\u7ecf\u8fc7\u6570\u5b57\u5316\u548c\u5d4c\u5165\u8868\u793a\u540e&#xff0c;\u5c31\u53d8\u6210\u4e86\u4e00\u4e2a(batch_size, sequence_length, embedding_dim)\u5f62\u72b6\u7684\u5f20\u91cf\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528\u5728\u524d\u9762\u7ae0\u8282\u5b66\u5230\u7684\u5404\u79cd\u7f51\u7edc\u7ed3\u6784\u6765\u5904\u7406\u5b83\u4e86\u3002<\/p>\n<ul>\n<li>\n<p>\u57fa\u4e8e\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff08;RNN\/LSTM&#xff09;\u7684\u6a21\u578b \u6587\u672c\u662f\u5178\u578b\u7684\u5e8f\u5217\u6570\u636e&#xff0c;\u8bcd\u8bed\u7684\u987a\u5e8f&#xff08;\u201c\u4e0d\u597d\u201d vs \u201c\u4e0d\u597d\u770b\u201d&#xff09;\u81f3\u5173\u91cd\u8981\u3002RNN&#xff0c;\u7279\u522b\u662fLSTM\u548cGRU&#xff0c;\u5176\u5185\u90e8\u7684\u5faa\u73af\u7ed3\u6784\u548c\u95e8\u63a7\u673a\u5236&#xff0c;\u5929\u7136\u5730\u9002\u5408\u5904\u7406\u8fd9\u79cd\u5e8f\u5217\u4f9d\u8d56\u5173\u7cfb\u3002<\/p>\n<ul>\n<li>\u5de5\u4f5c\u65b9\u5f0f&#xff1a;LSTM\u4f1a\u6309\u65f6\u95f4\u6b65\u4f9d\u6b21\u8bfb\u53d6\u8bcd\u5411\u91cf\u5e8f\u5217&#xff0c;\u5728\u6bcf\u4e2a\u65f6\u95f4\u6b65&#xff0c;\u5b83\u4f1a\u66f4\u65b0\u5176\u5185\u90e8\u7684\u9690\u85cf\u72b6\u6001&#xff0c;\u8fd9\u4e2a\u9690\u85cf\u72b6\u6001\u4f1a\u4e0d\u65ad\u7d2f\u79ef\u548c\u63d0\u70bc\u524d\u9762\u6240\u6709\u8bcd\u8bed\u7684\u4fe1\u606f\u3002<\/li>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;\u4e00\u4e2a\u5178\u578b\u7684\u7ed3\u6784\u662fEmbedding\u5c42 -&gt;\u00a0LSTM\u5c42 -&gt;\u00a0Dense\u5c42&#xff08;\u7528\u4e8e\u5206\u7c7b&#xff09;\u3002\u4e3a\u4e86\u9632\u6b62\u8fc7\u62df\u5408&#xff0c;\u53ef\u4ee5\u5728LSTM\u5c42\u4e4b\u540e\u52a0\u5165Dropout\u3002\u4f7f\u7528Bidirectional\u5305\u88c5\u5668&#xff08;\u5373\u53cc\u5411LSTM&#xff09;\u901a\u5e38\u80fd\u83b7\u5f97\u66f4\u597d\u7684\u6027\u80fd&#xff0c;\u56e0\u4e3a\u5b83\u80fd\u540c\u65f6\u4ece\u524d\u5411\u548c\u540e\u5411\u6355\u6349\u4e0a\u4e0b\u6587\u4fe1\u606f\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u57fa\u4e8e\u4e00\u7ef4\u5377\u79ef\u7f51\u7edc&#xff08;1D CNN&#xff09;\u7684\u6a21\u578b \u867d\u7136CNN\u5e38\u7528\u4e8e\u56fe\u50cf&#xff0c;\u4f46\u5b83\u540c\u6837\u53ef\u4ee5\u9ad8\u6548\u5730\u5904\u7406\u6587\u672c\u3002\u4e00\u7ef4CNN\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u6587\u672c\u7684N-gram\u7279\u5f81\u63d0\u53d6\u5668\u3002<\/p>\n<ul>\n<li>\u5de5\u4f5c\u65b9\u5f0f&#xff1a;\u4e00\u4e2a\u5927\u5c0f\u4e3ak\u7684\u4e00\u7ef4\u5377\u79ef\u6838&#xff0c;\u4f1a\u6ed1\u8fc7\u6574\u4e2a\u8bcd\u5411\u91cf\u5e8f\u5217&#xff0c;\u6bcf\u6b21\u201c\u770b\u201dk\u4e2a\u8fde\u7eed\u7684\u8bcd\u3002\u8fd9\u76f8\u5f53\u4e8e\u5728\u6355\u6349\u6587\u672c\u4e2d\u7684k-gram&#xff08;\u59822-gram, 3-gram&#xff09;\u6a21\u5f0f\u3002\u4f8b\u5982&#xff0c;\u201cnot good\u201d\u3001\u201cvery bad\u201d\u8fd9\u6837\u7684\u5173\u952e\u77ed\u8bed&#xff0c;\u5c31\u5f88\u5bb9\u6613\u88abCNN\u6355\u6349\u5230\u3002<\/li>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;\u901a\u5e38\u662fEmbedding\u5c42 -&gt; \u591a\u4e2a\u4e0d\u540c\u5927\u5c0f\u7684Conv1D\u5c42 -&gt;\u00a0GlobalMaxPooling1D\u5c42 -&gt;\u00a0Dense\u5c42\u3002\u4f7f\u7528\u4e0d\u540c\u5927\u5c0f\u7684\u5377\u79ef\u6838&#xff0c;\u53ef\u4ee5\u540c\u65f6\u6355\u6349\u4e0d\u540c\u957f\u5ea6\u7684\u77ed\u8bed\u6a21\u5f0f\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6a21\u578b\u5bf9\u6bd4\u4e0e\u9009\u62e9<\/p>\n<ul>\n<li>LSTM\u64c5\u957f\u6355\u6349\u957f\u8ddd\u79bb\u7684\u4f9d\u8d56\u5173\u7cfb&#xff0c;\u5bf9\u8bed\u5e8f\u66f4\u654f\u611f\u3002<\/li>\n<li>1D CNN\u8ba1\u7b97\u901f\u5ea6\u901a\u5e38\u6bd4LSTM\u5feb&#xff0c;\u64c5\u957f\u6355\u6349\u5173\u952e\u7684\u5c40\u90e8\u77ed\u8bed\u6a21\u5f0f\u3002<\/li>\n<li>\u5728\u5b9e\u8df5\u4e2d&#xff0c;\u4e24\u8005\u90fd\u80fd\u5728\u60c5\u611f\u5206\u6790\u4efb\u52a1\u4e0a\u53d6\u5f97\u5f88\u597d\u7684\u6548\u679c\u3002\u5bf9\u4e8e\u521d\u5b66\u8005&#xff0c;\u4ece\u4e00\u4e2a\u53cc\u5411LSTM\u6a21\u578b\u5f00\u59cb&#xff0c;\u662f\u4e00\u4e2a\u975e\u5e38\u7ecf\u5178\u4e14\u7a33\u5065\u7684\u9009\u62e9\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>import tensorflow as tf<\/p>\n<p>model &#061; tf.keras.Sequential([<br \/>\n    # 1. Embedding\u5c42<br \/>\n    tf.keras.layers.Embedding(10000, 128, input_length&#061;120),<\/p>\n<p>    # 2. \u4f7f\u7528\u53cc\u5411LSTM\u6355\u6349\u4e0a\u4e0b\u6587<br \/>\n    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(64)),<\/p>\n<p>    # 3. \u5168\u8fde\u63a5\u5c42\u8fdb\u884c\u7279\u5f81\u6574\u5408<br \/>\n    tf.keras.layers.Dense(64, activation&#061;&#039;relu&#039;),<br \/>\n    tf.keras.layers.Dropout(0.5),<\/p>\n<p>    # 4. \u8f93\u51fa\u5c42\u8fdb\u884c\u4e8c\u5143\u5206\u7c7b<br \/>\n    tf.keras.layers.Dense(1, activation&#061;&#039;sigmoid&#039;) # Sigmoid\u7528\u4e8e\u4e8c\u5143\u5206\u7c7b<br \/>\n])<\/p>\n<p>model.summary()<\/p>\n<h5>12.1.4 \u8bad\u7ec3\u3001\u8bc4\u4f30\u4e0e\u5b9e\u8df5<\/h5>\n<ul>\n<li>\n<p>\u6570\u636e\u96c6 \u6211\u4eec\u5c06\u4f7f\u7528Keras\u5185\u7f6e\u7684IMDb\u6570\u636e\u96c6&#xff0c;\u8fd9\u662f\u4e00\u4e2a\u7406\u60f3\u7684\u8d77\u70b9\u3002\u5b83\u5305\u542b\u4e86\u6765\u81ea\u4e92\u8054\u7f51\u7535\u5f71\u6570\u636e\u5e93\u768450,000\u6761\u7535\u5f71\u8bc4\u8bba&#xff0c;\u5df2\u7ecf\u88ab\u6807\u8bb0\u4e3a\u6b63\u9762&#xff08;1&#xff09;\u6216\u8d1f\u9762&#xff08;0&#xff09;\u3002\u66f4\u65b9\u4fbf\u7684\u662f&#xff0c;Keras\u5df2\u7ecf\u4e3a\u6211\u4eec\u5b8c\u6210\u4e86\u5927\u90e8\u5206\u9884\u5904\u7406\u5de5\u4f5c&#xff0c;\u6587\u672c\u88ab\u8f6c\u6362\u6210\u4e86\u6574\u6570\u5e8f\u5217&#xff0c;\u5176\u4e2d\u6bcf\u4e2a\u6574\u6570\u4ee3\u8868\u8bcd\u6c47\u8868\u4e2d\u7684\u4e00\u4e2a\u7279\u5b9a\u8bcd\u8bed\u3002<\/p>\n<\/li>\n<li>\n<p>\u8bad\u7ec3\u8fc7\u7a0b<\/p>\n<li>\u52a0\u8f7d\u6570\u636e&#xff1a;\u901a\u8fc7tf.keras.datasets.imdb.load_data()\u53ef\u4ee5\u4e00\u952e\u52a0\u8f7d\u6570\u636e\u3002\u6211\u4eec\u53ef\u4ee5\u6307\u5b9a\u8bcd\u6c47\u8868\u7684\u5927\u5c0f&#xff08;\u4f8b\u5982&#xff0c;\u53ea\u4fdd\u7559\u6700\u5e38\u89c1\u768410,000\u4e2a\u8bcd&#xff09;\u3002<\/li>\n<li>\u586b\u5145\u5e8f\u5217&#xff1a;\u7535\u5f71\u8bc4\u8bba\u7684\u957f\u5ea6\u5404\u4e0d\u76f8\u540c&#xff0c;\u4f46\u795e\u7ecf\u7f51\u7edc\u7684\u8f93\u5165\u9700\u8981\u662f\u89c4\u6574\u7684\u5f20\u91cf\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u5fc5\u987b\u4f7f\u7528pad_sequences\u51fd\u6570&#xff0c;\u5c06\u6240\u6709\u8bc4\u8bba\u5e8f\u5217\u201c\u586b\u5145\u201d\u6216\u201c\u622a\u65ad\u201d\u5230\u76f8\u540c\u7684\u957f\u5ea6\u3002\u8fd9\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u9884\u5904\u7406\u6b65\u9aa4\u3002<\/li>\n<li>\u7f16\u8bd1\u6a21\u578b&#xff1a;\n<ul>\n<li>\u635f\u5931\u51fd\u6570&#xff1a;\u5bf9\u4e8e\u4e8c\u5143\u5206\u7c7b\u95ee\u9898&#xff0c;binary_crossentropy\u662f\u6807\u51c6\u7684\u635f\u5931\u51fd\u6570\u3002<\/li>\n<li>\u4f18\u5316\u5668&#xff1a;Adam\u4f9d\u7136\u662f\u6211\u4eec\u7a33\u5065\u9ad8\u6548\u7684\u9009\u62e9\u3002<\/li>\n<li>\u8bc4\u4f30\u6307\u6807&#xff1a;\u6211\u4eec\u6700\u5173\u5fc3\u7684\u662faccuracy&#xff08;\u51c6\u786e\u7387&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u8bad\u7ec3&#xff1a;\u8c03\u7528model.fit()\u65b9\u6cd5&#xff0c;\u4f20\u5165\u8bad\u7ec3\u6570\u636e\u548c\u9a8c\u8bc1\u6570\u636e\u3002\u6211\u4eec\u53ef\u4ee5\u8bbe\u7f6e\u8bad\u7ec3\u7684\u8f6e\u6b21&#xff08;epochs&#xff09;\u548c\u6279\u91cf\u5927\u5c0f&#xff08;batch_size&#xff09;&#xff0c;\u5e76\u5229\u7528\u56de\u8c03\u51fd\u6570&#xff08;\u5982EarlyStopping&#xff09;\u6765\u9632\u6b62\u8fc7\u62df\u5408&#xff0c;\u81ea\u52a8\u5728\u6700\u4f73\u70b9\u505c\u6b62\u8bad\u7ec3\u3002<\/li>\n<p> import tensorflow as tf<br \/>\nfrom tensorflow.keras.datasets import imdb<br \/>\nfrom tensorflow.keras.preprocessing.sequence import pad_sequences<\/p>\n<p># 1. \u52a0\u8f7d\u6570\u636e<br \/>\nVOCAB_SIZE &#061; 10000<br \/>\n(train_data, train_labels), (test_data, test_labels) &#061; imdb.load_data(num_words&#061;VOCAB_SIZE)<\/p>\n<p># 2. \u586b\u5145\u5e8f\u5217<br \/>\nMAX_LENGTH &#061; 256<br \/>\ntrain_data &#061; pad_sequences(train_data, maxlen&#061;MAX_LENGTH, padding&#061;&#039;post&#039;, truncating&#061;&#039;post&#039;)<br \/>\ntest_data &#061; pad_sequences(test_data, maxlen&#061;MAX_LENGTH, padding&#061;&#039;post&#039;, truncating&#061;&#039;post&#039;)<\/p>\n<p># (\u63a5\u7eed\u4e0a\u4e00\u8282\u5b9a\u4e49\u7684\u6a21\u578b)<br \/>\n# 3. \u7f16\u8bd1\u6a21\u578b<br \/>\nmodel.compile(loss&#061;&#039;binary_crossentropy&#039;, optimizer&#061;&#039;adam&#039;, metrics&#061;[&#039;accuracy&#039;])<\/p>\n<p># 4. \u8bad\u7ec3\u6a21\u578b<br \/>\nhistory &#061; model.fit(train_data, train_labels, epochs&#061;10, batch_size&#061;64,<br \/>\n                    validation_data&#061;(test_data, test_labels))\n <\/li>\n<li>\n<p>\u7ed3\u679c\u5206\u6790 \u8bad\u7ec3\u5b8c\u6210\u540e&#xff0c;\u6211\u4eec\u4e0d\u80fd\u4ec5\u4ec5\u6ee1\u8db3\u4e8e\u4e00\u4e2a\u6570\u5b57&#xff0c;\u6bd4\u5982\u201c90%\u7684\u51c6\u786e\u7387\u201d\u3002\u771f\u6b63\u7684\u7406\u89e3\u6e90\u4e8e\u6df1\u5165\u7684\u5206\u6790\u3002<\/p>\n<ul>\n<li>\u53ef\u89c6\u5316\u5b66\u4e60\u66f2\u7ebf&#xff1a;\u7ed8\u5236history\u5bf9\u8c61\u4e2d\u7684\u8bad\u7ec3\u635f\u5931\/\u51c6\u786e\u7387\u548c\u9a8c\u8bc1\u635f\u5931\/\u51c6\u786e\u7387\u66f2\u7ebf\u3002\u8fd9\u662f\u8bca\u65ad\u8fc7\u62df\u5408\u6216\u6b20\u62df\u5408\u6700\u76f4\u89c2\u7684\u5de5\u5177\u3002<\/li>\n<li>\u8d28\u6027\u5206\u6790&#xff08;Qualitative Analysis&#xff09;&#xff1a;\u8fd9\u662f\u6700\u6709\u8da3\u7684\u90e8\u5206\u3002\u6211\u4eec\u5e94\u8be5\u7528\u81ea\u5df1\u8bbe\u8ba1\u7684\u3001\u5177\u6709\u6311\u6218\u6027\u7684\u8bc4\u8bba\u6765\u201c\u62f7\u95ee\u201d\u6a21\u578b&#xff0c;\u4ee5\u63a2\u7a76\u5176\u80fd\u529b\u7684\u8fb9\u754c&#xff1a;\n<ul>\n<li>\u5e26\u6709\u8f6c\u6298\u7684\u8bc4\u8bba&#xff1a;\u201c\u8fd9\u90e8\u7535\u5f71\u7684\u5f00\u5934\u5f88\u65e0\u804a&#xff0c;\u4f46\u6211\u6ca1\u60f3\u5230\u7ed3\u5c40\u5982\u6b64\u7cbe\u5f69&#xff01;\u201d\u2014\u2014\u6a21\u578b\u80fd\u5426\u6b63\u786e\u6355\u6349\u5230\u540e\u534a\u90e8\u5206\u7684\u5173\u952e\u60c5\u611f&#xff1f;<\/li>\n<li>\u5e26\u6709\u53cd\u8bbd\u7684\u8bc4\u8bba&#xff1a;\u201c\u54e6&#xff0c;\u771f\u662f\u2018\u592a\u68d2\u4e86\u2019&#xff0c;\u6211\u751f\u547d\u4e2d\u53c8\u6d6a\u8d39\u4e86\u4e24\u4e2a\u5c0f\u65f6\u3002\u201d\u2014\u2014\u8fd9\u5bf9\u4e8e\u53ea\u770b\u8bcd\u8bed\u8868\u9762\u610f\u4e49\u7684\u6a21\u578b\u662f\u6781\u5927\u7684\u6311\u6218&#xff0c;\u5f88\u53ef\u80fd\u88ab\u8bef\u5224\u4e3a\u6b63\u9762\u3002<\/li>\n<li>\u5e26\u6709\u5426\u5b9a\u8bcd\u7684\u8bc4\u8bba&#xff1a;\u201c\u6211\u5e76\u4e0d\u89c9\u5f97\u8fd9\u90e8\u7535\u5f71\u4e0d\u597d\u3002\u201d\u2014\u2014\u6a21\u578b\u662f\u5426\u80fd\u6b63\u786e\u7406\u89e3\u53cc\u91cd\u5426\u5b9a&#xff1f; \u901a\u8fc7\u89c2\u5bdf\u6a21\u578b\u5728\u8fd9\u4e9b\u4f8b\u5b50\u4e0a\u7684\u8868\u73b0&#xff0c;\u6211\u4eec\u53ef\u4ee5\u66f4\u6df1\u523b\u5730\u7406\u89e3\u5b83\u7684\u4f18\u52bf&#xff08;\u6bd4\u5982\u6355\u6349\u5173\u952e\u8bcd&#xff09;\u548c\u5c40\u9650\u6027&#xff08;\u6bd4\u5982\u96be\u4ee5\u7406\u89e3\u590d\u6742\u7684\u8bed\u8a00\u7ed3\u6784\u548c\u8bbd\u523a&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5b66\u4ee5\u81f4\u7528 \u4e00\u4e2a\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5176\u6700\u7ec8\u5f52\u5bbf\u662f\u5e94\u7528\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u5176\u4fdd\u5b58\u4e0b\u6765&#xff0c;\u5e76\u7f16\u5199\u4e00\u4e2a\u7b80\u5355\u7684\u9884\u6d4b\u51fd\u6570\u3002\u8fd9\u4e2a\u51fd\u6570\u63a5\u6536\u4e00\u6bb5\u65b0\u7684\u3001\u672a\u7ecf\u5904\u7406\u7684\u82f1\u6587\u6587\u672c&#xff0c;\u7136\u540e&#xff1a;<\/p>\n<li>\u4f7f\u7528\u6211\u4eec\u8bad\u7ec3\u65f6\u7528\u7684Tokenizer\u5bf9\u5176\u8fdb\u884c\u5206\u8bcd\u548c\u6574\u6570\u7f16\u7801\u3002<\/li>\n<li>\u4f7f\u7528pad_sequences\u8fdb\u884c\u586b\u5145\u3002<\/li>\n<li>\u8c03\u7528model.predict()\u65b9\u6cd5\u5f97\u5230\u4e00\u4e2a0\u52301\u4e4b\u95f4\u7684\u6982\u7387\u503c\u3002<\/li>\n<li>\u6839\u636e\u8fd9\u4e2a\u6982\u7387\u503c&#xff08;\u4f8b\u5982&#xff0c;\u5927\u4e8e0.5\u5219\u4e3a\u6b63\u9762&#xff09;&#xff0c;\u8fd4\u56de\u6700\u7ec8\u7684\u4eba\u7c7b\u53ef\u8bfb\u7684\u9884\u6d4b\u7ed3\u679c\u3002 \u81f3\u6b64&#xff0c;\u6211\u4eec\u7684\u60c5\u611f\u5206\u6790\u5668\u5c31\u4ece\u4e00\u4e2a\u5b9e\u9a8c\u54c1&#xff0c;\u53d8\u6210\u4e86\u4e00\u4e2a\u53ef\u4ee5\u96c6\u6210\u5230\u4efb\u4f55\u5e94\u7528\u4e2d\u7684\u5b9e\u7528\u5de5\u5177\u3002<\/li>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>12.2 \u673a\u5668\u7ffb\u8bd1&#xff1a;\u6784\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u4e2d\u82f1\u7ffb\u8bd1\u6a21\u578b \u2014\u2014 \u8de8\u8d8a\u8bed\u8a00\u7684\u5df4\u522b\u5854<\/h4>\n<p>\u60c5\u611f\u5206\u6790\u662f\u201c\u7406\u89e3\u201d\u8bed\u8a00&#xff0c;\u800c\u673a\u5668\u7ffb\u8bd1\u5219\u662f\u201c\u751f\u6210\u201d\u8bed\u8a00&#xff0c;\u5176\u96be\u5ea6\u548c\u590d\u6742\u5ea6\u90fd\u63d0\u5347\u4e86\u4e00\u4e2a\u91cf\u7ea7\u3002\u6211\u4eec\u5c06\u8981\u6311\u6218\u7684&#xff0c;\u662f\u4eba\u5de5\u667a\u80fd\u9886\u57df\u6700\u7ecf\u5178\u7684\u4efb\u52a1\u4e4b\u4e00&#xff1a;\u6784\u5efa\u4e00\u4e2a\u80fd\u5c06\u4e2d\u6587\u7ffb\u8bd1\u6210\u82f1\u6587\u7684\u6a21\u578b\u3002<\/p>\n<h5>12.2.1 \u9879\u76ee\u6784\u60f3&#xff1a;\u7f16\u7801\u5668-\u89e3\u7801\u5668&#xff08;Encoder-Decoder&#xff09;\u67b6\u6784\u7684\u667a\u6167<\/h5>\n<ul>\n<li>\n<p>\u4efb\u52a1\u7684\u672c\u8d28 \u673a\u5668\u7ffb\u8bd1\u7684\u6838\u5fc3&#xff0c;\u662f\u5c06\u4e00\u4e2a\u6e90\u8bed\u8a00\u5e8f\u5217&#xff08;\u5982\u4e2d\u6587\u53e5\u5b50&#xff09;\u6620\u5c04\u5230\u4e00\u4e2a\u76ee\u6807\u8bed\u8a00\u5e8f\u5217&#xff08;\u5982\u82f1\u6587\u53e5\u5b50&#xff09;\u3002\u8fd9\u4e24\u4e2a\u5e8f\u5217\u7684\u957f\u5ea6\u901a\u5e38\u4e0d\u540c&#xff0c;\u4e14\u8bed\u6cd5\u7ed3\u6784\u8fe5\u5f02&#xff0c;\u65e0\u6cd5\u50cf\u60c5\u611f\u5206\u6790\u90a3\u6837\u76f4\u63a5\u8fdb\u884c\u5206\u7c7b\u3002<\/p>\n<\/li>\n<li>\n<p>Encoder-Decoder\u6846\u67b6 \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898&#xff0c;\u7814\u7a76\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u6781\u5176\u4f18\u7f8e\u4e14\u5f3a\u5927\u7684\u6846\u67b6\u2014\u2014\u7f16\u7801\u5668-\u89e3\u7801\u5668&#xff08;Encoder-Decoder&#xff09;&#xff0c;\u4e5f\u79f0\u4e3a**\u5e8f\u5217\u5230\u5e8f\u5217&#xff08;Sequence-to-Sequence, Seq2Seq&#xff09;**\u6a21\u578b\u3002<\/p>\n<ul>\n<li>\u7f16\u7801\u5668&#xff08;Encoder&#xff09;&#xff1a;\u5b83\u7684\u4efb\u52a1\u662f\u201c\u9605\u8bfb\u201d\u5e76\u201c\u7406\u89e3\u201d\u6574\u4e2a\u6e90\u8bed\u8a00\u53e5\u5b50\u3002\u5b83\u901a\u5e38\u662f\u4e00\u4e2aRNN&#xff08;\u5982LSTM\u6216GRU&#xff09;&#xff0c;\u4f1a\u9010\u8bcd\u8bfb\u53d6\u6e90\u8bed\u8a00\u5e8f\u5217\u3002\u5f53\u8bfb\u5b8c\u6700\u540e\u4e00\u4e2a\u8bcd\u540e&#xff0c;\u7f16\u7801\u5668\u4f1a\u8f93\u51fa\u4e00\u4e2a\u6700\u7ec8\u7684\u4e0a\u4e0b\u6587\u5411\u91cf&#xff08;Context Vector&#xff09;\u3002\u8fd9\u4e2a\u5411\u91cf&#xff0c;\u5c31\u50cf\u662f\u7f16\u7801\u5668\u5bf9\u6574\u4e2a\u53e5\u5b50\u7684\u201c\u601d\u60f3\u201d\u6216\u201c\u6458\u8981\u201d&#xff0c;\u5b83\u88ab\u671f\u671b\u8574\u542b\u4e86\u6e90\u53e5\u5b50\u7684\u5168\u90e8\u8bed\u4e49\u4fe1\u606f\u3002<\/li>\n<li>\u89e3\u7801\u5668&#xff08;Decoder&#xff09;&#xff1a;\u5b83\u7684\u4efb\u52a1\u662f\u6839\u636e\u7f16\u7801\u5668\u63d0\u4f9b\u7684\u201c\u601d\u60f3\u201d&#xff0c;\u751f\u6210\u76ee\u6807\u8bed\u8a00\u53e5\u5b50\u3002\u5b83\u4e5f\u662f\u4e00\u4e2aRNN&#xff0c;\u4f46\u5b83\u7684\u5de5\u4f5c\u65b9\u5f0f\u66f4\u50cf\u4e00\u4e2a\u8bed\u8a00\u751f\u6210\u5668\u3002\u5b83\u63a5\u6536\u7f16\u7801\u5668\u8f93\u51fa\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u4f5c\u4e3a\u5176\u521d\u59cb\u72b6\u6001&#xff0c;\u7136\u540e\u4e00\u4e2a\u8bcd\u4e00\u4e2a\u8bcd\u5730\u751f\u6210\u76ee\u6807\u8bed\u8a00\u5e8f\u5217\u3002\u5728\u751f\u6210\u6bcf\u4e2a\u8bcd\u65f6&#xff0c;\u5b83\u90fd\u4f1a\u8003\u8651\u4e4b\u524d\u5df2\u7ecf\u751f\u6210\u7684\u8bcd\u548c\u5f53\u524d\u7684\u9690\u85cf\u72b6\u6001\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e2a\u6846\u67b6\u7684\u667a\u6167\u5728\u4e8e&#xff0c;\u5b83\u5c06\u590d\u6742\u7684\u7ffb\u8bd1\u4efb\u52a1&#xff0c;\u89e3\u8026\u6210\u4e86\u201c\u7406\u89e3\u201d\u548c\u201c\u751f\u6210\u201d\u4e24\u4e2a\u76f8\u5bf9\u72ec\u7acb\u7684\u9636\u6bb5&#xff0c;\u5e76\u901a\u8fc7\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u4f5c\u4e3a\u6865\u6881&#xff0c;\u8fde\u63a5\u4e86\u4e24\u79cd\u4e0d\u540c\u7684\u8bed\u8a00\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>12.2.2 \u57fa\u4e8eRNN\u7684Seq2Seq\u6a21\u578b\u5b9e\u73b0<\/h5>\n<ul>\n<li>\n<p>\u7f16\u7801\u5668&#xff08;Encoder&#xff09; \u5b9e\u73b0\u8d77\u6765\u975e\u5e38\u76f4\u63a5\u3002\u6211\u4eec\u4f7f\u7528\u4e00\u4e2aLSTM\u7f51\u7edc\u3002\u8f93\u5165\u662f\u4e00\u4e2a\u4e2d\u6587\u53e5\u5b50\u7684\u8bcd\u5411\u91cf\u5e8f\u5217\u3002\u6211\u4eec\u4e0d\u5173\u5fc3\u8fd9\u4e2aLSTM\u5728\u6bcf\u4e2a\u65f6\u95f4\u6b65\u7684\u8f93\u51fa&#xff0c;\u53ea\u5173\u5fc3\u5b83\u5728\u8bfb\u53d6\u5b8c\u6574\u4e2a\u53e5\u5b50\u540e\u7684\u6700\u540e\u4e00\u4e2a\u9690\u85cf\u72b6\u6001&#xff08;hidden state&#xff09;\u548c\u7ec6\u80de\u72b6\u6001&#xff08;cell state&#xff09;\u3002\u8fd9\u4e24\u4e2a\u72b6\u6001\u5411\u91cf\u5c06\u88ab\u6253\u5305&#xff0c;\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u5411\u91cf\u4f20\u9012\u7ed9\u89e3\u7801\u5668\u3002<\/p>\n<\/li>\n<li>\n<p>\u89e3\u7801\u5668&#xff08;Decoder&#xff09; \u89e3\u7801\u5668\u7684\u5b9e\u73b0\u8981\u590d\u6742\u4e00\u4e9b\u3002<\/p>\n<li>\u5b83\u4e5f\u662f\u4e00\u4e2aLSTM&#xff0c;\u5176\u521d\u59cb\u72b6\u6001\u88ab\u8bbe\u7f6e\u4e3a\u7f16\u7801\u5668\u4f20\u6765\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u3002<\/li>\n<li>\u5b83\u7684\u8f93\u5165&#xff0c;\u5728\u6bcf\u4e2a\u65f6\u95f4\u6b65&#xff0c;\u662f\u76ee\u6807\u8bed\u8a00\u5e8f\u5217\u4e2d\u524d\u4e00\u4e2a\u771f\u5b9e\u7684\u8bcd\u3002\u4f8b\u5982&#xff0c;\u8981\u751f\u6210\u82f1\u6587\u53e5\u5b50&#034;I am a student&#034;&#xff0c;\u5728\u751f\u6210&#034;am&#034;\u65f6&#xff0c;\u6211\u4eec\u4f1a\u7ed9\u89e3\u7801\u5668\u8f93\u5165&#034;I&#034;\u3002\u8fd9\u79cd\u4f7f\u7528\u771f\u5b9e\u76ee\u6807\u8bcd\u8bed\u4f5c\u4e3a\u4e0b\u4e00\u6b65\u8f93\u5165\u6765\u5f15\u5bfc\u8bad\u7ec3\u7684\u6280\u5de7&#xff0c;\u88ab\u79f0\u4e3a\u6559\u5e08\u5f3a\u5236&#xff08;Teacher Forcing&#xff09;\u3002\u5b83\u80fd\u6781\u5927\u5730\u7a33\u5b9a\u548c\u52a0\u901f\u6a21\u578b\u7684\u6536\u655b\u3002<\/li>\n<li>\u89e3\u7801\u5668LSTM\u7684\u6bcf\u4e2a\u65f6\u95f4\u6b65\u7684\u8f93\u51fa&#xff0c;\u4f1a\u7ecf\u8fc7\u4e00\u4e2aDense\u5c42&#xff08;\u5176\u5927\u5c0f\u4e3a\u76ee\u6807\u8bed\u8a00\u8bcd\u6c47\u8868\u7684\u5927\u5c0f&#xff09;&#xff0c;\u518d\u901a\u8fc7\u4e00\u4e2asoftmax\u6fc0\u6d3b\u51fd\u6570&#xff0c;\u6765\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u5e94\u8be5\u662f\u8bcd\u6c47\u8868\u4e2d\u7684\u54ea\u4e00\u4e2a\u3002<\/li>\n<\/li>\n<\/ul>\n<h5>12.2.3 \u6ce8\u610f\u529b\u673a\u5236\u7684\u5f15\u5165&#xff1a;\u8ba9\u6a21\u578b\u201c\u4e13\u6ce8\u201d\u8d77\u6765<\/h5>\n<ul>\n<li>\n<p>Seq2Seq\u7684\u74f6\u9888 \u57fa\u7840\u7684Seq2Seq\u6a21\u578b\u5b58\u5728\u4e00\u4e2a\u660e\u663e\u7684\u74f6\u9888&#xff1a;\u5b83\u8bd5\u56fe\u5c06\u6e90\u53e5\u5b50\u7684\u6240\u6709\u4fe1\u606f&#xff0c;\u90fd\u538b\u7f29\u5230\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u4e2d\u3002\u5bf9\u4e8e\u77ed\u53e5\u5b50&#xff0c;\u8fd9\u6216\u8bb8\u53ef\u884c&#xff1b;\u4f46\u5bf9\u4e8e\u957f\u800c\u590d\u6742\u7684\u53e5\u5b50&#xff0c;\u8fd9\u4e2a\u5c0f\u5c0f\u7684\u5411\u91cf\u5f88\u5bb9\u6613\u6210\u4e3a\u4fe1\u606f\u4f20\u9012\u7684\u201c\u74f6\u9888\u201d&#xff0c;\u5bfc\u81f4\u4fe1\u606f\u4e22\u5931&#xff0c;\u7ffb\u8bd1\u8d28\u91cf\u6025\u5267\u4e0b\u964d\u3002<\/p>\n<\/li>\n<li>\n<p>Attention\u7684\u539f\u7406 \u6ce8\u610f\u529b\u673a\u5236&#xff08;Attention Mechanism&#xff09;\u7684\u63d0\u51fa&#xff0c;\u662f\u673a\u5668\u7ffb\u8bd1\u4e43\u81f3\u6574\u4e2a\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u91cc\u7a0b\u7891\u4e8b\u4ef6\u3002\u5b83\u7684\u601d\u60f3\u975e\u5e38\u7b26\u5408\u4eba\u7c7b\u7684\u76f4\u89c9&#xff1a;\u5f53\u6211\u4eec\u5728\u8fdb\u884c\u7ffb\u8bd1\u65f6&#xff0c;\u6211\u4eec\u5e76\u4e0d\u4f1a\u770b\u5b8c\u6574\u4e2a\u4e2d\u6587\u53e5\u5b50\u540e&#xff0c;\u5c31\u5b8c\u5168\u51ed\u8bb0\u5fc6\u53bb\u5199\u82f1\u6587\u53e5\u5b50\u3002\u76f8\u53cd&#xff0c;\u5728\u7ffb\u8bd1\u82f1\u6587\u53e5\u5b50\u7684\u67d0\u4e2a\u90e8\u5206\u65f6&#xff0c;\u6211\u4eec\u7684\u6ce8\u610f\u529b\u4f1a\u52a8\u6001\u5730\u805a\u7126\u5728\u4e2d\u6587\u6e90\u53e5\u7684\u76f8\u5e94\u90e8\u5206\u3002<\/p>\n<p>Attention\u673a\u5236\u6b63\u662f\u6a21\u62df\u4e86\u8fd9\u4e00\u70b9\u3002\u5b83\u4e0d\u518d\u4f9d\u8d56\u5355\u4e00\u7684\u4e0a\u4e0b\u6587\u5411\u91cf&#xff0c;\u800c\u662f&#xff1a;<\/p>\n<li>\u4fdd\u7559\u7f16\u7801\u5668\u6240\u6709\u65f6\u95f4\u6b65\u7684\u8f93\u51fa&#xff08;\u800c\u4e0d\u4ec5\u4ec5\u662f\u6700\u540e\u4e00\u4e2a&#xff09;\u3002<\/li>\n<li>\u5728\u89e3\u7801\u5668\u751f\u6210\u6bcf\u4e00\u4e2a\u76ee\u6807\u8bcd\u65f6&#xff0c;\u5b83\u90fd\u4f1a\u6267\u884c\u4e00\u4e2a\u989d\u5916\u7684\u8ba1\u7b97&#xff1a; a. \u5c06\u89e3\u7801\u5668\u5f53\u524d\u7684\u9690\u85cf\u72b6\u6001&#xff0c;\u4e0e\u7f16\u7801\u5668\u6240\u6709\u65f6\u95f4\u6b65\u7684\u8f93\u51fa\u8fdb\u884c\u4e00\u6b21\u201c\u5339\u914d\u5ea6\u201d\u8ba1\u7b97&#xff0c;\u5f97\u5230\u4e00\u4e2a\u6ce8\u610f\u529b\u5206\u6570&#xff08;attention scores&#xff09;\u3002 b. \u5bf9\u8fd9\u4e9b\u5206\u6570\u8fdb\u884csoftmax\u5f52\u4e00\u5316&#xff0c;\u5f97\u5230\u4e00\u4e2a\u6ce8\u610f\u529b\u6743\u91cd&#xff08;attention weights&#xff09;\u5206\u5e03\u3002\u8fd9\u4e2a\u6743\u91cd\u5206\u5e03&#xff0c;\u5c31\u4ee3\u8868\u4e86\u5728\u751f\u6210\u5f53\u524d\u8fd9\u4e2a\u76ee\u6807\u8bcd\u65f6&#xff0c;\u5e94\u8be5\u5bf9\u6e90\u53e5\u5b50\u7684\u54ea\u4e9b\u90e8\u5206\u6295\u5165\u591a\u5c11\u201c\u6ce8\u610f\u529b\u201d\u3002 c. \u7528\u8fd9\u4e9b\u6743\u91cd\u5bf9\u7f16\u7801\u5668\u7684\u6240\u6709\u8f93\u51fa\u8fdb\u884c\u52a0\u6743\u6c42\u548c&#xff0c;\u5f97\u5230\u4e00\u4e2a\u52a8\u6001\u7684\u3001\u4e3a\u5f53\u524d\u65f6\u95f4\u6b65\u91cf\u8eab\u5b9a\u5236\u7684\u4e0a\u4e0b\u6587\u5411\u91cf\u3002<\/li>\n<li>\u5c06\u8fd9\u4e2a\u52a8\u6001\u7684\u4e0a\u4e0b\u6587\u5411\u91cf&#xff0c;\u4e0e\u89e3\u7801\u5668\u5f53\u524d\u7684\u8f93\u5165\u62fc\u63a5\u5728\u4e00\u8d77&#xff0c;\u518d\u9001\u5165\u89e3\u7801\u5668\u7684\u6838\u5fc3LSTM\u5355\u5143\u3002<\/li>\n<p>\u901a\u8fc7Attention\u673a\u5236&#xff0c;\u89e3\u7801\u5668\u5728\u6bcf\u4e00\u6b65\u90fd\u80fd\u201c\u56de\u5934\u770b\u201d&#xff0c;\u5e76\u7cbe\u51c6\u5730\u4ece\u6e90\u53e5\u5b50\u4e2d\u63d0\u53d6\u6700\u76f8\u5173\u7684\u4fe1\u606f&#xff0c;\u4ece\u800c\u6781\u5927\u5730\u63d0\u5347\u4e86\u957f\u53e5\u5b50\u7684\u7ffb\u8bd1\u8d28\u91cf\u3002\u5b83\u8d4b\u4e88\u4e86\u6a21\u578b\u4e00\u79cd\u201c\u4e13\u6ce8\u201d\u4e8e\u5c40\u90e8\u7684\u80fd\u529b\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>12.2.4 \u6570\u636e\u51c6\u5907\u4e0e\u8bad\u7ec3\u7ec6\u8282<\/h5>\n<ul>\n<li>\n<p>\u6570\u636e\u96c6 \u673a\u5668\u7ffb\u8bd1\u9700\u8981\u5e73\u884c\u8bed\u6599\u5e93&#xff08;Parallel Corpus&#xff09;&#xff0c;\u5373\u6210\u5bf9\u7684\u3001\u4e92\u4e3a\u7ffb\u8bd1\u7684\u53e5\u5b50\u3002\u6709\u8bb8\u591a\u516c\u5f00\u7684\u4e2d\u82f1\u5e73\u884c\u8bed\u6599\u5e93\u53ef\u4f9b\u4f7f\u7528&#xff0c;\u4f8b\u5982Tatoeba\u9879\u76ee\u3001UN Parallel Corpus\u7b49\u3002<\/p>\n<\/li>\n<li>\n<p>\u6587\u672c\u9884\u5904\u7406<\/p>\n<li>\u4e3a\u6e90\u8bed\u8a00&#xff08;\u4e2d\u6587&#xff09;\u548c\u76ee\u6807\u8bed\u8a00&#xff08;\u82f1\u6587&#xff09;\u5206\u522b\u6784\u5efa\u8bcd\u6c47\u8868\u548c\u5206\u8bcd\u5668\u3002\u4e2d\u6587\u5206\u8bcd\u901a\u5e38\u4f7f\u7528jieba\u7b49\u5e93\u3002<\/li>\n<li>\u5728\u6bcf\u4e2a\u53e5\u5b50\u7684\u5f00\u5934\u548c\u7ed3\u5c3e&#xff0c;\u6dfb\u52a0\u7279\u6b8a\u7684\u8d77\u59cb\u7b26&#xff08;\u5982&lt;start&gt;&#xff09;\u548c\u7ed3\u675f\u7b26&#xff08;&lt;end&gt;&#xff09;\u3002\u8fd9\u4e3a\u89e3\u7801\u5668\u63d0\u4f9b\u4e86\u660e\u786e\u7684\u751f\u6210\u8d77\u70b9\u548c\u7ec8\u70b9\u4fe1\u53f7\u3002<\/li>\n<li>\u5c06\u6240\u6709\u53e5\u5b50\u586b\u5145\u5230\u7edf\u4e00\u7684\u957f\u5ea6\u3002<\/li>\n<\/li>\n<li>\n<p>\u8bad\u7ec3\u4e0e\u63a8\u7406\u7684\u533a\u522b<\/p>\n<ul>\n<li>\u8bad\u7ec3\u65f6&#xff1a;\u6211\u4eec\u4f7f\u7528\u201c\u6559\u5e08\u5f3a\u5236\u201d&#xff0c;\u5c06\u771f\u5b9e\u7684\u3001\u5b8c\u6574\u7684\u82f1\u6587\u76ee\u6807\u5e8f\u5217\u5582\u7ed9\u89e3\u7801\u5668\u3002<\/li>\n<li>\u63a8\u7406\u65f6&#xff08;\u7ffb\u8bd1\u4e00\u4e2a\u5168\u65b0\u7684\u4e2d\u6587\u53e5\u5b50&#xff09;&#xff1a;\u6211\u4eec\u6ca1\u6709\u771f\u5b9e\u7684\u82f1\u6587\u53e5\u5b50\u3002\u8fd9\u65f6&#xff0c;\u89e3\u7801\u5668\u5fc5\u987b**\u81ea\u56de\u5f52\u5730&#xff08;autoregressive&#xff09;**\u5de5\u4f5c&#xff1a;\n<li>\u5c06\u4e2d\u6587\u53e5\u5b50\u8f93\u5165\u7f16\u7801\u5668&#xff0c;\u5f97\u5230\u4e0a\u4e0b\u6587\u5411\u91cf\u3002<\/li>\n<li>\u5c06&lt;start&gt;\u6807\u8bb0\u4f5c\u4e3a\u89e3\u7801\u5668\u7684\u7b2c\u4e00\u4e2a\u8f93\u5165\u3002<\/li>\n<li>\u89e3\u7801\u5668\u9884\u6d4b\u51fa\u7b2c\u4e00\u4e2a\u8bcd&#xff08;\u4f8b\u5982&#034;I&#034;&#xff09;\u3002<\/li>\n<li>\u5c06\u521a\u521a\u9884\u6d4b\u51fa\u7684&#034;I&#034;&#xff0c;\u4f5c\u4e3a\u89e3\u7801\u5668\u4e0b\u4e00\u4e2a\u65f6\u95f4\u6b65\u7684\u8f93\u5165\u3002<\/li>\n<li>\u89e3\u7801\u5668\u9884\u6d4b\u51fa\u7b2c\u4e8c\u4e2a\u8bcd&#xff08;\u4f8b\u5982&#034;am&#034;&#xff09;\u3002<\/li>\n<li>&#8230;\u8fd9\u4e2a\u8fc7\u7a0b\u4e0d\u65ad\u5faa\u73af&#xff0c;\u76f4\u5230\u89e3\u7801\u5668\u9884\u6d4b\u51fa&lt;end&gt;\u6807\u8bb0&#xff0c;\u6216\u8005\u8fbe\u5230\u9884\u8bbe\u7684\u6700\u5927\u957f\u5ea6\u4e3a\u6b62\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u6784\u5efa\u4e00\u4e2a\u9ad8\u8d28\u91cf\u7684\u7ffb\u8bd1\u6a21\u578b\u662f\u4e00\u9879\u590d\u6742\u7684\u5de5\u7a0b&#xff0c;\u4f46\u901a\u8fc7\u5b9e\u73b0\u4e00\u4e2a\u5e26Attention\u7684Seq2Seq\u6a21\u578b&#xff0c;\u8bfb\u8005\u5c06\u80fd\u6df1\u523b\u7406\u89e3\u73b0\u4ee3NLP\u4e2d\u5e8f\u5217\u751f\u6210\u4efb\u52a1\u7684\u6838\u5fc3\u601d\u60f3\u3002<\/p>\n<hr \/>\n<h4>12.3 \u667a\u80fd\u95ee\u7b54\u673a\u5668\u4eba&#xff1a;\u57fa\u4e8e\u77e5\u8bc6\u5e93\u7684\u95ee\u7b54\u7cfb\u7edf \u2014\u2014 \u8d4b\u4e88\u673a\u5668\u201c\u8bb0\u5fc6\u201d\u4e0e\u201c\u903b\u8f91\u201d<\/h4>\n<p>\u6700\u540e\u4e00\u4e2a\u9879\u76ee&#xff0c;\u6211\u4eec\u5c06\u6784\u5efa\u4e00\u4e2a\u5728\u5de5\u4e1a\u754c\u5e94\u7528\u6781\u4e3a\u5e7f\u6cdb\u7684\u7cfb\u7edf\u2014\u2014\u95ee\u7b54\u673a\u5668\u4eba&#xff08;Chatbot&#xff09;\u3002\u5b83\u80fd\u6839\u636e\u7528\u6237\u7684\u95ee\u9898&#xff0c;\u4ece\u4e00\u4e2a\u7ed9\u5b9a\u7684\u77e5\u8bc6\u5e93\u4e2d\u627e\u5230\u5e76\u8fd4\u56de\u6700\u76f8\u5173\u7684\u7b54\u6848\u3002<\/p>\n<h5>12.3.1 \u9879\u76ee\u5b9a\u4f4d&#xff1a;\u6211\u4eec\u8981\u505a\u4ec0\u4e48\u6837\u7684\u673a\u5668\u4eba&#xff1f;<\/h5>\n<ul>\n<li>\n<p>Chatbot\u7684\u5206\u7c7b<\/p>\n<ul>\n<li>\u95f2\u804a\u578b&#xff1a;\u76ee\u6807\u662f\u8fdb\u884c\u5f00\u653e\u3001\u6d41\u7545\u3001\u6709\u8da3\u7684\u5bf9\u8bdd&#xff0c;\u5982\u5fae\u8f6f\u5c0f\u51b0\u3002<\/li>\n<li>\u4efb\u52a1\u578b&#xff1a;\u65e8\u5728\u5e2e\u52a9\u7528\u6237\u5b8c\u6210\u7279\u5b9a\u4efb\u52a1&#xff0c;\u5982\u8ba2\u9910\u3001\u67e5\u8be2\u5929\u6c14\u3002<\/li>\n<li>\u95ee\u7b54\u578b&#xff1a;\u4e13\u6ce8\u4e8e\u6839\u636e\u4e8b\u5b9e\u77e5\u8bc6\u56de\u7b54\u7528\u6237\u7684\u95ee\u9898\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u672c\u6b21\u76ee\u6807 \u6211\u4eec\u5c06\u6784\u5efa\u4e00\u4e2a\u68c0\u7d22\u5f0f&#xff08;Retrieval-based&#xff09;\u7684\u95ee\u7b54\u673a\u5668\u4eba\u3002\u5b83\u4e0d\u201c\u751f\u6210\u201d\u65b0\u7684\u7b54\u6848&#xff0c;\u800c\u662f\u4ece\u4e00\u4e2a\u9884\u5148\u51c6\u5907\u597d\u7684\u77e5\u8bc6\u5e93&#xff08;Knowledge Base, KB&#xff09;&#xff08;\u4f8b\u5982&#xff0c;\u4e00\u4e2a\u5305\u542b\u4e0a\u767e\u6761\u201c\u95ee\u9898-\u7b54\u6848\u201d\u5bf9\u7684FAQ\u6587\u6863&#xff09;\u4e2d&#xff0c;\u68c0\u7d22\u51fa\u4e0e\u7528\u6237\u95ee\u9898\u6700\u5339\u914d\u7684\u90a3\u4e2a\u7b54\u6848\u3002\u8fd9\u5728\u667a\u80fd\u5ba2\u670d\u3001\u4f01\u4e1a\u5185\u90e8\u77e5\u8bc6\u67e5\u8be2\u7b49\u573a\u666f\u4e2d\u975e\u5e38\u5b9e\u7528\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>12.3.2 \u6838\u5fc3\u6280\u672f&#xff1a;\u4ece\u6d77\u91cf\u6587\u672c\u4e2d\u68c0\u7d22\u7b54\u6848<\/h5>\n<p>\u6838\u5fc3\u95ee\u9898\u662f&#xff1a;\u5982\u4f55\u8861\u91cf\u7528\u6237\u63d0\u51fa\u7684\u95ee\u9898&#xff0c;\u4e0e\u77e5\u8bc6\u5e93\u4e2d\u54ea\u4e2a\u201c\u6807\u51c6\u95ee\u9898\u201d\u6700\u76f8\u4f3c&#xff1f;<\/p>\n<ul>\n<li>\n<p>\u4fe1\u606f\u68c0\u7d22\u7684\u7ecf\u5178\u65b9\u6cd5&#xff1a;TF-IDF<\/p>\n<ul>\n<li>\u601d\u60f3&#xff1a;TF-IDF&#xff08;\u8bcd\u9891-\u9006\u6587\u6863\u9891\u7387&#xff09;\u662f\u4e00\u79cd\u7ecf\u5178\u7684\u7edf\u8ba1\u65b9\u6cd5\u3002\u5b83\u8ba4\u4e3a&#xff0c;\u4e00\u4e2a\u8bcd\u7684\u91cd\u8981\u6027&#xff0c;\u4e0e\u5b83\u5728\u4e00\u7bc7\u6587\u7ae0\u4e2d\u51fa\u73b0\u7684\u6b21\u6570&#xff08;TF&#xff09;\u6210\u6b63\u6bd4&#xff0c;\u4e0e\u5b83\u5728\u6574\u4e2a\u8bed\u6599\u5e93\u4e2d\u51fa\u73b0\u7684\u9891\u7387&#xff08;IDF&#xff09;\u6210\u53cd\u6bd4\u3002\u4e00\u4e2a\u8bcd\u5982\u679c\u5728\u4e00\u7bc7\u6587\u7ae0\u4e2d\u9891\u7e41\u51fa\u73b0&#xff0c;\u4f46\u5728\u5176\u4ed6\u6587\u7ae0\u4e2d\u5f88\u5c11\u51fa\u73b0&#xff0c;\u90a3\u4e48\u5b83\u5f88\u53ef\u80fd\u5c31\u662f\u8fd9\u7bc7\u6587\u7ae0\u7684\u5173\u952e\u8bcd\u3002<\/li>\n<li>\u6d41\u7a0b&#xff1a;\u6211\u4eec\u53ef\u4ee5\u4e3a\u7528\u6237\u95ee\u9898\u548c\u77e5\u8bc6\u5e93\u4e2d\u7684\u6bcf\u4e2a\u95ee\u9898&#xff0c;\u90fd\u8ba1\u7b97\u4e00\u4e2aTF-IDF\u5411\u91cf\u3002\u7136\u540e&#xff0c;\u901a\u8fc7\u8ba1\u7b97\u7528\u6237\u95ee\u9898\u5411\u91cf\u4e0e\u6240\u6709\u77e5\u8bc6\u5e93\u95ee\u9898\u5411\u91cf\u4e4b\u95f4\u7684\u4f59\u5f26\u76f8\u4f3c\u5ea6&#xff08;Cosine Similarity&#xff09;&#xff0c;\u6765\u627e\u5230\u6700\u76f8\u4f3c\u7684\u90a3\u4e2a\u3002<\/li>\n<li>\u5c40\u9650&#xff1a;TF-IDF\u5b8c\u5168\u57fa\u4e8e\u8bcd\u9891&#xff0c;\u65e0\u6cd5\u7406\u89e3\u8bed\u4e49\u3002\u4f8b\u5982&#xff0c;\u7528\u6237\u95ee\u201c\u7535\u8111\u5f00\u4e0d\u4e86\u673a\u600e\u4e48\u529e&#xff1f;\u201d&#xff0c;\u77e5\u8bc6\u5e93\u91cc\u662f\u201c\u7b14\u8bb0\u672c\u65e0\u6cd5\u542f\u52a8\u5982\u4f55\u5904\u7406&#xff1f;\u201d&#xff0c;\u5c3d\u7ba1\u610f\u601d\u76f8\u540c&#xff0c;\u4f46\u56e0\u4e3a\u5173\u952e\u8bcd\u4e0d\u5339\u914d&#xff0c;TF-IDF\u53ef\u80fd\u65e0\u6cd5\u627e\u5230\u6b63\u786e\u7b54\u6848\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u7684\u8bed\u4e49\u5339\u914d \u4e3a\u4e86\u514b\u670dTF-IDF\u7684\u5c40\u9650&#xff0c;\u6211\u4eec\u5fc5\u987b\u8fdb\u5165**\u8bed\u4e49&#xff08;Semantic&#xff09;**\u5c42\u9762\u3002<\/p>\n<ul>\n<li>\u53e5\u5b50\u5d4c\u5165&#xff08;Sentence Embeddings&#xff09;&#xff1a;\u8fd9\u6b63\u662f\u6211\u4eec\u9700\u8981\u7684\u6280\u672f\u3002\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e2a\u5f3a\u5927\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b&#xff08;\u5982BERT&#xff09;&#xff0c;\u5c06\u7528\u6237\u7684\u95ee\u9898\u548c\u77e5\u8bc6\u5e93\u4e2d\u7684\u6240\u6709\u6807\u51c6\u95ee\u9898&#xff0c;\u90fd\u7f16\u7801\u6210\u4e00\u4e2a\u9ad8\u7ef4\u7684\u3001\u7a20\u5bc6\u7684\u53e5\u5b50\u5d4c\u5165\u5411\u91cf\u3002\u8fd9\u4e9b\u5411\u91cf&#xff0c;\u6355\u6349\u4e86\u53e5\u5b50\u7684\u6df1\u5c42\u8bed\u4e49\u4fe1\u606f\u3002<\/li>\n<li>\u5411\u91cf\u76f8\u4f3c\u5ea6\u8ba1\u7b97&#xff1a;\u540c\u6837&#xff0c;\u6211\u4eec\u8ba1\u7b97\u7528\u6237\u95ee\u9898\u5411\u91cf\u4e0e\u77e5\u8bc6\u5e93\u4e2d\u6240\u6709\u95ee\u9898\u5411\u91cf\u7684\u4f59\u5f26\u76f8\u4f3c\u5ea6\u3002\u4f46\u8fd9\u4e00\u6b21&#xff0c;\u7531\u4e8e\u5411\u91cf\u8574\u542b\u4e86\u8bed\u4e49&#xff0c;\u5373\u4f7f\u4e24\u4e2a\u53e5\u5b50\u7684\u63aa\u8f9e\u5b8c\u5168\u4e0d\u540c&#xff0c;\u53ea\u8981\u610f\u601d\u76f8\u8fd1&#xff0c;\u5b83\u4eec\u7684\u5411\u91cf\u5728\u7a7a\u95f4\u4e2d\u7684\u4f4d\u7f6e\u4e5f\u4f1a\u975e\u5e38\u63a5\u8fd1&#xff0c;\u4ece\u800c\u5f97\u5230\u5f88\u9ad8\u7684\u76f8\u4f3c\u5ea6\u5206\u6570\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>12.3.3 \u5b9e\u6218&#xff1a;\u4f7f\u7528\u9884\u8bad\u7ec3BERT\u6a21\u578b\u6784\u5efa\u95ee\u7b54\u7cfb\u7edf<\/h5>\n<p>\u4ece\u5934\u8bad\u7ec3\u4e00\u4e2aBERT\u6a21\u578b\u662f\u4e0d\u73b0\u5b9e\u7684\u3002\u6211\u4eec\u5c06\u76f4\u63a5\u4f7f\u7528\u5f00\u6e90\u793e\u533a\u5c01\u88c5\u597d\u7684\u5f3a\u5927\u5de5\u5177\u3002<\/p>\n<ul>\n<li>\n<p>Sentence-Transformers\u5e93\u4ecb\u7ecd sentence-transformers\u662f\u4e00\u4e2a\u57fa\u4e8ePyTorch\u548cTransformers\u7684Python\u5e93\u3002\u5b83\u63d0\u4f9b\u4e86\u5927\u91cf\u5728\u201c\u8bed\u4e49\u76f8\u4f3c\u5ea6\u201d\u4efb\u52a1\u4e0a\u5fae\u8c03\u8fc7\u7684\u9884\u8bad\u7ec3\u6a21\u578b&#xff0c;\u53ef\u4ee5\u6781\u5176\u65b9\u4fbf\u5730\u751f\u6210\u9ad8\u8d28\u91cf\u7684\u53e5\u5b50\u5d4c\u5165\u3002<\/p>\n<\/li>\n<li>\n<p>\u6784\u5efa\u77e5\u8bc6\u5e93\u7d22\u5f15 \u4e3a\u4e86\u63d0\u9ad8\u6548\u7387&#xff0c;\u6211\u4eec\u4e0d\u80fd\u5728\u6bcf\u6b21\u7528\u6237\u63d0\u95ee\u65f6&#xff0c;\u90fd\u91cd\u65b0\u8ba1\u7b97\u77e5\u8bc6\u5e93\u4e2d\u6240\u6709\u95ee\u9898\u7684\u5411\u91cf\u3002\u6b63\u786e\u7684\u505a\u6cd5\u662f\u79bb\u7ebf\u9884\u5904\u7406&#xff1a;<\/p>\n<li>\u9009\u62e9\u4e00\u4e2a\u5408\u9002\u7684\u9884\u8bad\u7ec3\u6a21\u578b&#xff08;\u4f8b\u5982&#xff0c;&#039;paraphrase-multilingual-MiniLM-L12-v2&#039;&#xff0c;\u4e00\u4e2a\u652f\u6301\u591a\u8bed\u8a00\u4e14\u6027\u80fd\u4f18\u5f02\u7684\u6a21\u578b&#xff09;\u3002<\/li>\n<li>\u5c06\u77e5\u8bc6\u5e93\u4e2d\u6240\u6709\u7684\u95ee\u9898\u6587\u672c&#xff0c;\u4e00\u6b21\u6027\u5730\u5168\u90e8\u7f16\u7801\u6210\u53e5\u5b50\u5d4c\u5165\u5411\u91cf\u3002<\/li>\n<li>\u5c06\u8fd9\u4e9b\u5411\u91cf&#xff08;\u6211\u4eec\u79f0\u4e4b\u4e3a\u77e5\u8bc6\u5e93\u7d22\u5f15&#xff09;\u8fde\u540c\u5bf9\u5e94\u7684\u7b54\u6848&#xff0c;\u4e00\u8d77\u5b58\u50a8\u8d77\u6765\u3002<\/li>\n<\/li>\n<li>\n<p>\u5b9e\u73b0\u95ee\u7b54\u6d41\u7a0b \u5f53\u4e00\u4e2a\u5728\u7ebf\u8bf7\u6c42\u5230\u6765\u65f6&#xff0c;\u6d41\u7a0b\u5982\u4e0b&#xff1a;<\/p>\n<li>\u63a5\u6536\u7528\u6237\u8f93\u5165\u7684\u95ee\u9898\u3002<\/li>\n<li>\u4f7f\u7528\u76f8\u540c\u7684\u9884\u8bad\u7ec3\u6a21\u578b&#xff0c;\u5c06\u8fd9\u4e2a\u95ee\u9898\u5b9e\u65f6\u7f16\u7801\u6210\u4e00\u4e2a\u5411\u91cf\u3002<\/li>\n<li>\u4f7f\u7528sentence_transformers.util.cos_sim\u7b49\u5de5\u5177&#xff0c;\u5feb\u901f\u8ba1\u7b97\u8be5\u95ee\u9898\u5411\u91cf\u4e0e\u6211\u4eec\u9884\u5148\u5b58\u50a8\u7684\u77e5\u8bc6\u5e93\u7d22\u5f15\u4e2d\u6240\u6709\u5411\u91cf\u7684\u4f59\u5f26\u76f8\u4f3c\u5ea6\u3002<\/li>\n<li>\u627e\u5230\u76f8\u4f3c\u5ea6\u6700\u9ad8\u7684\u90a3\u4e2a\u7d22\u5f15&#xff0c;\u5e76\u8fd4\u56de\u5176\u5bf9\u5e94\u7684\u9884\u8bbe\u7b54\u6848\u3002<\/li>\n<p> from sentence_transformers import SentenceTransformer, util<\/p>\n<p># 1. \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b<br \/>\nmodel &#061; SentenceTransformer(&#039;paraphrase-multilingual-MiniLM-L12-v2&#039;)<\/p>\n<p># 2. \u77e5\u8bc6\u5e93 (QA\u5bf9)<br \/>\nknowledge_base_questions &#061; [&#034;\u7b14\u8bb0\u672c\u65e0\u6cd5\u542f\u52a8\u5982\u4f55\u5904\u7406&#xff1f;&#034;, &#034;\u5982\u4f55\u8fde\u63a5\u5230\u516c\u53f8WiFi&#xff1f;&#034;, &#8230;]<br \/>\nknowledge_base_answers &#061; [&#034;\u8bf7\u68c0\u67e5\u7535\u6e90\u9002\u914d\u5668&#8230;&#034;, &#034;\u8bf7\u9009\u62e9SSID\u4e3a&#8230;&#034;, &#8230;]<\/p>\n<p># 3. \u79bb\u7ebf\u6784\u5efa\u77e5\u8bc6\u5e93\u7d22\u5f15<br \/>\nkb_question_embeddings &#061; model.encode(knowledge_base_questions, convert_to_tensor&#061;True)<\/p>\n<p># 4. \u5728\u7ebf\u95ee\u7b54\u6d41\u7a0b<br \/>\nuser_question &#061; &#034;\u6211\u7684\u7535\u8111\u5f00\u4e0d\u4e86\u673a\u600e\u4e48\u529e&#xff1f;&#034;<br \/>\nuser_question_embedding &#061; model.encode(user_question, convert_to_tensor&#061;True)<\/p>\n<p># 5. \u8ba1\u7b97\u4f59\u5f26\u76f8\u4f3c\u5ea6<br \/>\ncos_scores &#061; util.cos_sim(user_question_embedding, kb_question_embeddings)[0]<\/p>\n<p># 6. \u627e\u5230\u6700\u76f8\u4f3c\u7684\u95ee\u9898\u7d22\u5f15<br \/>\ntop_result_index &#061; cos_scores.argmax()<\/p>\n<p># 7. \u8fd4\u56de\u5bf9\u5e94\u7684\u7b54\u6848<br \/>\nprint(f&#034;\u6700\u5339\u914d\u7684\u95ee\u9898: {knowledge_base_questions[top_result_index]}&#034;)<br \/>\nprint(f&#034;\u7b54\u6848: {knowledge_base_answers[top_result_index]}&#034;)\n <\/li>\n<\/ul>\n<h5>12.3.4 \u7cfb\u7edf\u7684\u4f18\u5316\u4e0e\u5c55\u671b<\/h5>\n<ul>\n<li>\u63d0\u5347\u68c0\u7d22\u6548\u7387&#xff1a;\u5f53\u77e5\u8bc6\u5e93\u53d8\u5f97\u975e\u5e38\u5e9e\u5927&#xff08;\u4f8b\u5982\u767e\u4e07\u7ea7\u522b&#xff09;\u65f6&#xff0c;\u9010\u4e00\u8ba1\u7b97\u4f59\u5f26\u76f8\u4f3c\u5ea6\u4f1a\u53d8\u5f97\u5f88\u6162\u3002\u8fd9\u65f6&#xff0c;\u6211\u4eec\u9700\u8981\u4e13\u95e8\u7684\u5411\u91cf\u76f8\u4f3c\u5ea6\u641c\u7d22\u5f15\u64ce&#xff0c;\u5982Facebook\u5f00\u6e90\u7684FAISS\u5e93\u3002\u5b83\u53ef\u4ee5\u6784\u5efa\u5411\u91cf\u7d22\u5f15&#xff0c;\u5b9e\u73b0\u6beb\u79d2\u7ea7\u7684\u6d77\u91cf\u5411\u91cf\u68c0\u7d22\u3002<\/li>\n<li>\u4ece\u68c0\u7d22\u5f0f\u5230\u751f\u6210\u5f0f&#xff1a;\u6211\u4eec\u5b9e\u73b0\u7684\u68c0\u7d22\u5f0f\u95ee\u7b54\u7cfb\u7edf&#xff0c;\u7b54\u6848\u662f\u56fa\u5b9a\u7684\u3002\u800c\u5f53\u524d\u66f4\u524d\u6cbf\u7684&#xff0c;\u662f\u4ee5ChatGPT\u4e3a\u4ee3\u8868\u7684\u751f\u6210\u5f0f&#xff08;Generative&#xff09;\u95ee\u7b54\u7cfb\u7edf\u3002\u5b83\u4eec\u57fa\u4e8e\u8d85\u5927\u89c4\u6a21\u7684\u8bed\u8a00\u6a21\u578b&#xff08;LLM&#xff09;&#xff0c;\u80fd\u591f\u6839\u636e\u4e0a\u4e0b\u6587\u548c\u80cc\u666f\u77e5\u8bc6&#xff0c;\u52a8\u6001\u5730\u751f\u6210\u6d41\u7545\u3001\u81ea\u7136\u3001\u5168\u65b0\u7684\u7b54\u6848\u3002\u8fd9\u4ee3\u8868\u4e86\u95ee\u7b54\u673a\u5668\u4eba\u53d1\u5c55\u7684\u672a\u6765\u65b9\u5411&#xff0c;\u5176\u80cc\u540e\u662f\u66f4\u5f3a\u5927\u7684Transformer\u67b6\u6784\u548c\u66f4\u6d77\u91cf\u7684\u9884\u8bad\u7ec3\u6570\u636e\u3002<\/li>\n<\/ul>\n<hr \/>\n<p>\u5c0f\u7ed3&#xff1a;\u8bed\u8a00\u662f\u601d\u60f3\u7684\u5c45\u6240<\/p>\n<p>\u5728\u672c\u7ae0\u7684\u63a2\u7d22\u4e2d&#xff0c;\u6211\u4eec\u4ece\u6355\u6349\u6587\u5b57\u7684\u201c\u6e29\u5ea6\u201d&#xff0c;\u5230\u8de8\u8d8a\u8bed\u8a00\u7684\u201c\u5df4\u522b\u5854\u201d&#xff0c;\u518d\u5230\u6784\u5efa\u80fd\u4e0e\u6211\u4eec\u5bf9\u8bdd\u7684\u201c\u4f19\u4f34\u201d&#xff0c;\u4e00\u6b65\u6b65\u6df1\u5165\u4e86\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u6838\u5fc3\u5730\u5e26\u3002<\/p>\n<ul>\n<li>\u6211\u4eec\u901a\u8fc7\u60c5\u611f\u5206\u6790\u9879\u76ee&#xff0c;\u638c\u63e1\u4e86NLP\u4efb\u52a1\u7684\u57fa\u672c\u6d41\u7a0b&#xff1a;\u6587\u672c\u9884\u5904\u7406\u3001\u8bcd\u5d4c\u5165\u8868\u793a&#xff0c;\u4ee5\u53ca\u5982\u4f55\u4f7f\u7528RNN\/CNN\u5bf9\u6587\u672c\u8fdb\u884c\u5206\u7c7b\u3002<\/li>\n<li>\u5728\u673a\u5668\u7ffb\u8bd1\u9879\u76ee\u4e2d&#xff0c;\u6211\u4eec\u6df1\u5165\u5b66\u4e60\u4e86\u91cc\u7a0b\u7891\u5f0f\u7684Encoder-Decoder\u67b6\u6784\u548cAttention\u673a\u5236&#xff0c;\u7406\u89e3\u4e86\u5e8f\u5217\u751f\u6210\u4efb\u52a1\u7684\u7cbe\u9ad3\u3002<\/li>\n<li>\u6700\u540e&#xff0c;\u5728\u95ee\u7b54\u673a\u5668\u4eba\u9879\u76ee\u4e2d&#xff0c;\u6211\u4eec\u63a5\u89e6\u4e86\u5de5\u4e1a\u754c\u5e94\u7528\u5e7f\u6cdb\u7684\u8bed\u4e49\u68c0\u7d22\u601d\u60f3&#xff0c;\u5e76\u5229\u7528\u5f3a\u5927\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b&#xff08;\u5982BERT&#xff09;&#xff0c;\u6784\u5efa\u4e86\u4e00\u4e2a\u8fdc\u6bd4\u4f20\u7edf\u5173\u952e\u8bcd\u5339\u914d\u66f4\u667a\u80fd\u7684\u7cfb\u7edf\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e09\u4e2a\u9879\u76ee&#xff0c;\u5206\u522b\u4ee3\u8868\u4e86NLP\u9886\u57df\u7684\u6587\u672c\u5206\u7c7b\u3001\u5e8f\u5217\u5230\u5e8f\u5217\u751f\u6210\u548c\u8bed\u4e49\u7406\u89e3\u4e0e\u5339\u914d\u8fd9\u4e09\u5927\u6838\u5fc3\u65b9\u5411\u3002\u5b83\u4eec\u5171\u540c\u63ed\u793a\u4e86\u4e00\u4e2a\u771f\u7406&#xff1a;\u8bed\u8a00\u662f\u601d\u60f3\u7684\u5c45\u6240\u3002\u6211\u4eec\u6240\u6709\u7684\u52aa\u529b&#xff0c;\u90fd\u662f\u4e3a\u4e86\u8ba9\u673a\u5668\u80fd\u591f\u66f4\u597d\u5730\u8fdb\u5165\u8fd9\u4e2a\u5c45\u6240&#xff0c;\u53bb\u7406\u89e3\u3001\u53bb\u63d0\u70bc\u3001\u53bb\u8fd0\u7528\u5176\u4e2d\u8574\u542b\u7684\u65e0\u7a77\u667a\u6167\u3002<\/p>\n<p>NLP\u662f\u4e00\u4e2a\u65e5\u65b0\u6708\u5f02\u7684\u9886\u57df&#xff0c;\u65b0\u7684\u6a21\u578b\u3001\u65b0\u7684\u601d\u60f3\u5c42\u51fa\u4e0d\u7a77\u3002\u4f46\u4e07\u53d8\u4e0d\u79bb\u5176\u5b97&#xff0c;\u672c\u7ae0\u6240\u5b9e\u8df5\u7684\u6838\u5fc3\u6982\u5ff5\u2014\u2014\u5411\u91cf\u8868\u793a\u3001\u5e8f\u5217\u5efa\u6a21\u3001\u6ce8\u610f\u529b\u673a\u5236\u3001\u8bed\u4e49\u5339\u914d\u2014\u2014\u5c06\u662f\u60a8\u672a\u6765\u5b66\u4e60\u66f4\u524d\u6cbf\u6280\u672f\u65f6&#xff0c;\u5fc3\u4e2d\u6700\u575a\u5b9e\u7684\u57fa\u77f3\u3002\u613f\u60a8\u4fdd\u6301\u8fd9\u4efd\u597d\u5947\u4e0e\u70ed\u60c5&#xff0c;\u7ee7\u7eed\u5728\u8fd9\u7247\u8bed\u8a00\u7684\u6d77\u6d0b\u4e2d\u4e58\u98ce\u7834\u6d6a&#xff0c;\u7528\u4ee3\u7801&#xff0c;\u53bb\u6784\u5efa\u80fd\u4e0e\u4eba\u7c7b\u8fdb\u884c\u66f4\u6df1\u5c42\u6b21\u4ea4\u6d41\u7684\u3001\u771f\u6b63\u610f\u4e49\u4e0a\u7684\u667a\u80fd\u4f53\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u5341\u4e09\u7ae0&#xff1a;\u9879\u76ee\u5b9e\u6218&#xff1a;\u5176\u4ed6\u9886\u57df<\/h3>\n<p>\u8de8\u8d8a\u8fb9\u754c&#xff0c;\u8d4b\u80fd\u4e07\u8c61<\/p>\n<p>\u5728\u4e4b\u524d\u7684\u7ae0\u8282\u4e2d&#xff0c;\u6211\u4eec\u5df2\u7ecf\u6559\u4f1a\u4e86\u673a\u5668\u5982\u4f55\u53bb\u201c\u770b\u201d\u61c2\u4e16\u754c&#xff08;\u8ba1\u7b97\u673a\u89c6\u89c9&#xff09;&#xff0c;\u5982\u4f55\u53bb\u201c\u542c\u201d\u61c2\u548c\u201c\u8bf4\u201d\u51fa\u601d\u60f3&#xff08;\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff09;\u3002\u6211\u4eec\u8ba9\u673a\u5668\u62e5\u6709\u4e86\u7c7b\u4f3c\u4eba\u7c7b\u7684\u611f\u77e5\u4e0e\u4ea4\u6d41\u80fd\u529b\u3002\u7136\u800c&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u7586\u57df\u8fdc\u4e0d\u6b62\u4e8e\u6b64\u3002\u5b83\u7684\u529b\u91cf&#xff0c;\u5982\u540c\u4e00\u80a1\u5f3a\u5927\u7684\u601d\u6f6e&#xff0c;\u6b63\u5728\u6e17\u900f\u548c\u91cd\u5851\u7740\u4f17\u591a\u770b\u4f3c\u4e0eCV\u3001NLP\u622a\u7136\u4e0d\u540c\u7684\u9886\u57df\u3002<\/p>\n<p>\u672c\u7ae0&#xff0c;\u6211\u4eec\u5c06\u8e0f\u4e0a\u4e00\u6b21\u8de8\u8d8a\u8fb9\u754c\u7684\u63a2\u7d22\u4e4b\u65c5\u3002\u6211\u4eec\u5c06\u770b\u5230&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u7684\u6838\u5fc3\u601d\u60f3\u2014\u2014\u65e0\u8bba\u662f\u4ece\u6570\u636e\u4e2d\u81ea\u52a8\u63d0\u53d6\u5c42\u6b21\u5316\u7279\u5f81&#xff0c;\u8fd8\u662f\u5bf9\u5e8f\u5217\u4fe1\u606f\u8fdb\u884c\u5efa\u6a21&#xff0c;\u6291\u6216\u662f\u5bf9\u672a\u6765\u7684\u201c\u4ef7\u503c\u201d\u8fdb\u884c\u8bc4\u4f30\u2014\u2014\u90fd\u5177\u6709\u60ca\u4eba\u7684\u666e\u9002\u6027\u3002\u6211\u4eec\u5c06\u805a\u7126\u4e8e\u4e09\u4e2a\u5168\u65b0\u7684\u9886\u57df&#xff0c;\u5b83\u4eec\u5206\u522b\u5bf9\u5e94\u7740\u4e09\u79cd\u6df1\u523b\u7684\u667a\u6167&#xff1a;<\/p>\n<li>\u9884\u6d4b\u672a\u6765&#xff1a;\u6211\u4eec\u5c06\u6df1\u5165\u65f6\u95f4\u7684\u6cb3\u6d41&#xff0c;\u5b66\u4e60\u5982\u4f55\u4ece\u5386\u53f2\u7684\u6d9f\u6f2a\u4e2d&#xff0c;\u9884\u6d4b\u672a\u6765\u7684\u6ce2\u6f9c\u3002<\/li>\n<li>\u7406\u89e3\u9009\u62e9&#xff1a;\u6211\u4eec\u5c06\u63a2\u7d22\u4eba\u5fc3\u7684\u504f\u597d&#xff0c;\u6784\u5efa\u4e00\u4e2a\u6bd4\u60a8\u66f4\u61c2\u60a8\u7684\u63a8\u8350\u201c\u77e5\u5df1\u201d\u3002<\/li>\n<li>\u5b66\u4e60\u51b3\u7b56&#xff1a;\u6211\u4eec\u5c06\u5f15\u5bfc\u4e00\u4e2a\u667a\u80fd\u4f53&#xff0c;\u5728\u865a\u62df\u4e16\u754c\u4e2d\u901a\u8fc7\u4e0d\u65ad\u7684\u8bd5\u9519&#xff0c;\u4ece\u96f6\u5f00\u59cb\u4e60\u5f97\u901a\u5173\u6e38\u620f\u7684\u667a\u6167\u3002<\/li>\n<p>\u8fd9\u4e0d\u4ec5\u662f\u4e00\u6b21\u6280\u672f\u6808\u7684\u62d3\u5c55&#xff0c;\u66f4\u662f\u4e00\u6b21\u601d\u7ef4\u8303\u5f0f\u7684\u5ef6\u4f38\u3002\u5b83\u5c06\u5411\u6211\u4eec\u8bc1\u660e&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u5e76\u975e\u4e00\u7cfb\u5217\u5b64\u7acb\u7684\u7b97\u6cd5&#xff0c;\u800c\u662f\u4e00\u79cd\u89e3\u51b3\u95ee\u9898\u7684\u5f3a\u5927\u4e16\u754c\u89c2\u3002\u51c6\u5907\u597d&#xff0c;\u8ba9\u6211\u4eec\u4e00\u8d77\u89c1\u8bc1\u6df1\u5ea6\u5b66\u4e60\u5982\u4f55\u4e3a\u8fd9\u4e9b\u590d\u6742\u7684\u9886\u57df\u8d4b\u80fd&#xff0c;\u5e76\u4ece\u4e2d\u6c72\u53d6\u66f4\u5e7f\u9614\u3001\u66f4\u6df1\u9083\u7684\u667a\u6167\u3002<\/p>\n<h4>13.1 \u65f6\u95f4\u5e8f\u5217\u9884\u6d4b&#xff1a;\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u6216\u5929\u6c14\u53d8\u5316 \u2014\u2014 \u5728\u65f6\u95f4\u7684\u6cb3\u6d41\u4e2d&#xff0c;\u5bfb\u627e\u672a\u6765\u7684\u6d9f\u6f2a<\/h4>\n<p>\u65f6\u95f4&#xff0c;\u662f\u4e16\u754c\u4e0a\u6700\u516c\u5e73\u4e5f\u6700\u795e\u79d8\u7684\u7ef4\u5ea6\u3002\u5b83\u5355\u5411\u6d41\u6dcc&#xff0c;\u6c38\u4e0d\u56de\u5934\u3002\u800c\u65f6\u95f4\u5e8f\u5217\u6570\u636e&#xff0c;\u6b63\u662f\u8fd9\u6761\u6cb3\u6d41\u4e0a\u7559\u4e0b\u7684\u4e00\u4e32\u4e32\u8db3\u8ff9\u3002\u5b66\u4f1a\u89e3\u8bfb\u8fd9\u4e9b\u8db3\u8ff9&#xff0c;\u5e76\u636e\u6b64\u9884\u6d4b\u672a\u6765\u7684\u8d70\u5411&#xff0c;\u662f\u4eba\u7c7b\u957f\u4e45\u4ee5\u6765\u7684\u6e34\u671b&#xff0c;\u4e5f\u662f\u65e0\u6570\u79d1\u5b66\u4e0e\u5546\u4e1a\u51b3\u7b56\u7684\u57fa\u77f3\u3002<\/p>\n<h5>13.1.1 \u9879\u76ee\u6784\u60f3&#xff1a;\u4e3a\u4f55\u65f6\u95f4\u5982\u6b64\u7279\u6b8a&#xff1f;<\/h5>\n<ul>\n<li>\n<p>\u65f6\u95f4\u5e8f\u5217\u7684\u5b9a\u4e49 \u65f6\u95f4\u5e8f\u5217&#xff08;Time Series&#xff09;\u662f\u4e00\u7ec4\u6309\u7167\u65f6\u95f4\u987a\u5e8f\u6392\u5217\u7684\u6570\u636e\u70b9\u7684\u96c6\u5408\u3002\u5b83\u7684\u6838\u5fc3\u7279\u5f81&#xff0c;\u4e5f\u662f\u5b83\u4e0e\u6211\u4eec\u4e4b\u524d\u5904\u7406\u7684\u72ec\u7acb\u540c\u5206\u5e03\u6570\u636e&#xff08;\u5982\u56fe\u50cf&#xff09;\u6700\u6839\u672c\u7684\u533a\u522b&#xff0c;\u5728\u4e8e\u6570\u636e\u70b9\u4e4b\u95f4\u5b58\u5728\u7740\u5185\u5728\u7684\u65f6\u95f4\u4f9d\u8d56\u6027\u3002\u7b80\u800c\u8a00\u4e4b&#xff0c;\u201c\u8fc7\u53bb\u201d\u4f1a\u5f71\u54cd\u201c\u672a\u6765\u201d\u3002<\/p>\n<\/li>\n<li>\n<p>\u5e94\u7528\u573a\u666f \u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u7684\u5e94\u7528\u65e0\u5904\u4e0d\u5728&#xff0c;\u6df1\u523b\u5730\u5f71\u54cd\u7740\u6211\u4eec\u7684\u751f\u6d3b\u548c\u7ecf\u6d4e\u6d3b\u52a8&#xff1a;<\/p>\n<ul>\n<li>\u91d1\u878d\u5e02\u573a&#xff1a;\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u3001\u6c47\u7387\u3001\u6307\u6570\u7684\u672a\u6765\u8d70\u52bf\u3002<\/li>\n<li>\u6c14\u8c61\u5b66&#xff1a;\u9884\u62a5\u672a\u6765\u51e0\u5c0f\u65f6\u6216\u51e0\u5929\u7684\u6c14\u6e29\u3001\u964d\u96e8\u91cf\u3001\u98ce\u901f\u3002<\/li>\n<li>\u5de5\u4e1a\u751f\u4ea7&#xff1a;\u9884\u6d4b\u670d\u52a1\u5668\u7684\u8d1f\u8f7d\u3001\u4ea7\u54c1\u7684\u9500\u91cf\u3001\u7535\u7f51\u7684\u8017\u7535\u91cf&#xff0c;\u4ee5\u63d0\u524d\u8fdb\u884c\u8d44\u6e90\u8c03\u5ea6\u3002<\/li>\n<li>\u533b\u7597\u5065\u5eb7&#xff1a;\u76d1\u63a7\u75c5\u4eba\u7684\u5fc3\u7387\u3001\u8840\u7cd6\u7b49\u751f\u7406\u6307\u6807&#xff0c;\u9884\u8b66\u5f02\u5e38\u72b6\u51b5\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6311\u6218 \u4e00\u6761\u770b\u4f3c\u7b80\u5355\u7684\u65f6\u95f4\u5e8f\u5217\u66f2\u7ebf&#xff0c;\u5f80\u5f80\u662f\u591a\u79cd\u590d\u6742\u6a21\u5f0f\u7684\u53e0\u52a0&#xff1a;<\/p>\n<ul>\n<li>\u8d8b\u52bf&#xff08;Trend&#xff09;&#xff1a;\u6570\u636e\u968f\u65f6\u95f4\u5448\u73b0\u51fa\u7684\u957f\u671f\u4e0a\u5347\u6216\u4e0b\u964d\u7684\u5927\u65b9\u5411\u3002<\/li>\n<li>\u5b63\u8282\u6027&#xff08;Seasonality&#xff09;&#xff1a;\u6570\u636e\u4ee5\u56fa\u5b9a\u7684\u5468\u671f&#xff08;\u5982\u5929\u3001\u5468\u3001\u5e74&#xff09;\u5448\u73b0\u51fa\u7684\u89c4\u5f8b\u6027\u6ce2\u52a8\u3002<\/li>\n<li>\u566a\u58f0&#xff08;Noise&#xff09;&#xff1a;\u4e0d\u89c4\u5219\u7684\u3001\u968f\u673a\u7684\u6ce2\u52a8\u3002 \u4e00\u4e2a\u597d\u7684\u9884\u6d4b\u6a21\u578b&#xff0c;\u9700\u8981\u6709\u80fd\u529b\u4ece\u6df7\u6742\u7684\u4fe1\u53f7\u4e2d&#xff0c;\u89e3\u8026\u5e76\u5b66\u4e60\u5230\u8fd9\u4e9b\u6f5c\u5728\u7684\u6a21\u5f0f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>13.1.2 \u6570\u636e\u51c6\u5907&#xff1a;\u5c06\u65f6\u95f4\u201c\u7a97\u53e3\u5316\u201d<\/h5>\n<p>\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b&#xff0c;\u5c24\u5176\u662f\u76d1\u7763\u5b66\u4e60\u6a21\u578b&#xff0c;\u9700\u8981\u660e\u786e\u7684(\u8f93\u5165, \u8f93\u51fa)\u6570\u636e\u5bf9\u3002\u4f46\u539f\u59cb\u7684\u65f6\u95f4\u5e8f\u5217\u53ea\u662f\u4e00\u6761\u8fde\u7eed\u7684\u7ebf&#xff0c;\u6211\u4eec\u5982\u4f55\u5c06\u5176\u6539\u9020\u4e3a\u76d1\u7763\u5b66\u4e60\u95ee\u9898\u5462&#xff1f;\u7b54\u6848\u662f\u7a97\u53e3\u5316&#xff08;Windowing&#xff09;\u3002<\/p>\n<ul>\n<li>\n<p>\u6570\u636e\u96c6 \u4e3a\u7b80\u5316\u95ee\u9898&#xff0c;\u6211\u4eec\u5148\u4ece\u4e00\u4e2a**\u5355\u53d8\u91cf&#xff08;Univariate&#xff09;**\u65f6\u95f4\u5e8f\u5217\u5f00\u59cb&#xff0c;\u4f8b\u5982\u67d0\u53ea\u80a1\u7968\u8fde\u7eed200\u5929\u7684\u6bcf\u65e5\u6536\u76d8\u4ef7\u3002<\/p>\n<\/li>\n<li>\n<p>\u6570\u636e\u9884\u5904\u7406 \u5728\u5904\u7406\u65f6\u95f4\u5e8f\u5217\u65f6&#xff0c;**\u5f52\u4e00\u5316&#xff08;Normalization&#xff09;**\u5c24\u4e3a\u91cd\u8981\u3002\u5c06\u6570\u636e\u7f29\u653e\u5230\u4e00\u4e2a\u8f83\u5c0f\u7684\u533a\u95f4&#xff08;\u59820\u52301\u6216-1\u52301&#xff09;&#xff0c;\u53ef\u4ee5\u5e2e\u52a9\u795e\u7ecf\u7f51\u7edc\u6a21\u578b&#xff08;\u5c24\u5176\u662fRNN&#xff09;\u66f4\u5feb\u3001\u66f4\u7a33\u5b9a\u5730\u6536\u655b\u3002\u5e38\u7528\u7684\u65b9\u6cd5\u662fMinMaxScaler\u3002\u91cd\u8981\u7684\u662f&#xff0c;\u6211\u4eec\u5e94\u8be5\u53ea\u7528\u8bad\u7ec3\u96c6\u7684\u6570\u636e\u6765\u8ba1\u7b97\u7f29\u653e\u7684\u53c2\u6570&#xff08;min\u548cmax&#xff09;&#xff0c;\u7136\u540e\u518d\u7528\u8fd9\u4e9b\u53c2\u6570\u53bb\u8f6c\u6362\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6&#xff0c;\u4ee5\u9632\u6b62\u672a\u6765\u4fe1\u606f\u7684\u6cc4\u9732\u3002<\/p>\n<\/li>\n<li>\n<p>\u521b\u5efa\u65f6\u95f4\u7a97\u53e3 \u8fd9\u662f\u5c06\u65f6\u95f4\u5e8f\u5217\u95ee\u9898\u201c\u7ffb\u8bd1\u201d\u6210\u76d1\u7763\u5b66\u4e60\u8bed\u8a00\u7684\u5173\u952e\u6b65\u9aa4\u3002\u6211\u4eec\u50cf\u4e00\u4e2a\u79fb\u52a8\u7684\u201c\u7a97\u53e3\u201d\u4e00\u6837&#xff0c;\u5728\u65f6\u95f4\u5e8f\u5217\u4e0a\u6ed1\u52a8&#xff0c;\u6765\u751f\u6210\u6211\u4eec\u7684\u6570\u636e\u96c6\u3002<\/p>\n<ul>\n<li>\u7a97\u53e3\u5927\u5c0f&#xff08;Window Size&#xff09;&#xff1a;\u5b9a\u4e49\u6211\u4eec\u7528\u591a\u957f\u7684\u5386\u53f2\u6570\u636e\u4f5c\u4e3a\u8f93\u5165\u3002\u4f8b\u5982&#xff0c;window_size &#061; 30\u3002<\/li>\n<li>\u9884\u6d4b\u6b65\u957f&#xff08;Horizon&#xff09;&#xff1a;\u5b9a\u4e49\u6211\u4eec\u8981\u9884\u6d4b\u672a\u6765\u591a\u8fdc\u7684\u6570\u636e\u3002\u4f8b\u5982&#xff0c;horizon &#061; 1\u3002<\/li>\n<li>\u6ed1\u52a8\u8fc7\u7a0b&#xff1a;\n<li>\u53d6\u7b2c1\u5929\u5230\u7b2c30\u5929\u7684\u6570\u636e\u4f5c\u4e3a\u7b2c\u4e00\u4e2a\u8f93\u5165X_1&#xff0c;\u7b2c31\u5929\u7684\u6570\u636e\u4f5c\u4e3a\u7b2c\u4e00\u4e2a\u76ee\u6807y_1\u3002<\/li>\n<li>\u7a97\u53e3\u5411\u53f3\u6ed1\u52a8\u4e00\u5929&#xff0c;\u53d6\u7b2c2\u5929\u5230\u7b2c31\u5929\u7684\u6570\u636e\u4f5c\u4e3a\u7b2c\u4e8c\u4e2a\u8f93\u5165X_2&#xff0c;\u7b2c32\u5929\u7684\u6570\u636e\u4f5c\u4e3a\u7b2c\u4e8c\u4e2a\u76ee\u6807y_2\u3002<\/li>\n<li>&#8230;\u4ee5\u6b64\u7c7b\u63a8&#xff0c;\u76f4\u5230\u6570\u636e\u672b\u5c3e\u3002<\/li>\n<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u4e00\u6761\u957f\u5ea6\u4e3a200\u7684\u65f6\u95f4\u5e8f\u5217&#xff0c;\u5c31\u88ab\u6211\u4eec\u8f6c\u6362\u6210\u4e86\u5927\u7ea6170\u4e2a(\u8f93\u5165\u5e8f\u5217, \u76ee\u6807\u503c)\u7684\u6570\u636e\u5bf9&#xff0c;\u53ef\u4ee5\u88ab\u4efb\u4f55\u76d1\u7763\u5b66\u4e60\u6a21\u578b\u4f7f\u7528\u4e86\u3002<\/p>\n<\/li>\n<\/ul>\n<p>import numpy as np<\/p>\n<p># \u5047\u8bbeseries\u662f\u6211\u4eec\u7684\u65f6\u95f4\u5e8f\u5217\u6570\u636e (numpy array)<br \/>\ndef windowed_dataset(series, window_size, batch_size, shuffle_buffer):<br \/>\n    dataset &#061; tf.data.Dataset.from_tensor_slices(series)<br \/>\n    dataset &#061; dataset.window(window_size &#043; 1, shift&#061;1, drop_remainder&#061;True)<br \/>\n    dataset &#061; dataset.flat_map(lambda window: window.batch(window_size &#043; 1))<br \/>\n    dataset &#061; dataset.shuffle(shuffle_buffer)<br \/>\n    dataset &#061; dataset.map(lambda window: (window[:-1], window[-1]))<br \/>\n    dataset &#061; dataset.batch(batch_size).prefetch(1)<br \/>\n    return dataset<\/p>\n<p># \u4f7f\u7528\u793a\u4f8b<br \/>\n# window_size &#061; 30<br \/>\n# train_set &#061; windowed_dataset(train_data, window_size, 32, 1000)<\/p>\n<p>tf.data API\u63d0\u4f9b\u4e86\u9ad8\u6548\u3001\u7075\u6d3b\u7684\u7a97\u53e3\u5316\u5de5\u5177&#xff0c;\u662f\u5904\u7406\u5927\u89c4\u6a21\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u7684\u9996\u9009\u3002<\/p>\n<h5>13.1.3 \u6a21\u578b\u7684\u6f14\u8fdb&#xff1a;\u4eceRNN\u5230\u4e13\u7528\u67b6\u6784<\/h5>\n<p>\u7a97\u53e3\u5316\u540e\u7684\u6570\u636e&#xff0c;\u5176\u8f93\u5165\u90e8\u5206\u662f\u4e00\u4e2a\u5e8f\u5217&#xff0c;\u8fd9\u6b63\u662fRNN\u548c1D CNN\u5927\u663e\u8eab\u624b\u7684\u821e\u53f0\u3002<\/p>\n<ul>\n<li>\n<p>\u57fa\u4e8eLSTM\u7684\u6a21\u578b \u8fd9\u662f\u6700\u76f4\u89c2\u7684\u601d\u8def\u3002LSTM\u88ab\u8bbe\u8ba1\u7528\u6765\u6355\u6349\u5e8f\u5217\u4e2d\u7684\u65f6\u95f4\u4f9d\u8d56\u5173\u7cfb&#xff0c;\u56e0\u6b64\u975e\u5e38\u9002\u5408\u5904\u7406\u65f6\u95f4\u7a97\u53e3\u6570\u636e\u3002<\/p>\n<ul>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;\u4e00\u4e2a\u7b80\u5355\u7684\u6a21\u578b\u53ef\u4ee5\u662fInput\u5c42 -&gt;\u00a0LSTM\u5c42 -&gt;\u00a0Dense\u5c42\u3002\u5982\u679c\u5e8f\u5217\u6a21\u5f0f\u590d\u6742&#xff0c;\u53ef\u4ee5\u5806\u53e0\u591a\u4e2aLSTM\u5c42\u3002\u7531\u4e8e\u6211\u4eec\u7684\u76ee\u6807\u662f\u9884\u6d4b\u4e00\u4e2a\u6570\u503c&#xff0c;\u6240\u4ee5\u6700\u540e\u7684Dense\u5c42\u53ea\u6709\u4e00\u4e2a\u795e\u7ecf\u5143&#xff0c;\u5e76\u4e14\u901a\u5e38\u4e0d\u4f7f\u7528\u6fc0\u6d3b\u51fd\u6570\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u57fa\u4e8e1D CNN\u7684\u6a21\u578b \u4e00\u7ef4\u5377\u79ef\u7f51\u7edc\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u4e00\u79cd\u9ad8\u6548\u7684\u6a21\u5f0f\u68c0\u6d4b\u5668\u3002\u5728\u65f6\u95f4\u5e8f\u5217\u4e2d&#xff0c;\u5b83\u53ef\u4ee5\u5feb\u901f\u5730\u6355\u6349\u5230\u5c40\u90e8\u7684\u4e0a\u5347\/\u4e0b\u964d\u6a21\u5f0f\u3001\u5c0f\u7684\u6ce2\u5cf0\u6ce2\u8c37\u7b49\u3002<\/p>\n<ul>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;Input\u5c42 -&gt;\u00a0Conv1D\u5c42 -&gt;\u00a0Dense\u5c42\u3002Conv1D\u7684\u5377\u79ef\u6838\u4f1a\u5728\u65f6\u95f4\u7ef4\u5ea6\u4e0a\u6ed1\u52a8&#xff0c;\u63d0\u53d6\u51fa\u5173\u952e\u7684\u5c40\u90e8\u7279\u5f81\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7ed3\u5408CNN\u4e0eRNN&#xff1a;\u5f3a\u5f3a\u8054\u5408 \u4e00\u4e2a\u66f4\u9ad8\u7ea7\u3001\u4e5f\u5e38\u5e38\u66f4\u6709\u6548\u7684\u7b56\u7565&#xff0c;\u662f\u5c06CNN\u548cRNN\u7ed3\u5408\u8d77\u6765\u3002<\/p>\n<ul>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;Input\u00a0-&gt;\u00a0Conv1D\u00a0-&gt;\u00a0LSTM\u00a0-&gt;\u00a0Dense\u3002<\/li>\n<li>\u5de5\u4f5c\u539f\u7406&#xff1a;\n<li>\u9996\u5148&#xff0c;\u4f7f\u7528\u4e00\u4e2aConv1D\u5c42\u5bf9\u8f93\u5165\u7684\u957f\u5e8f\u5217\u8fdb\u884c\u4e00\u6b21\u201c\u9884\u5904\u7406\u201d\u3002\u5377\u79ef\u5c42\u53ef\u4ee5\u9ad8\u6548\u5730\u8bc6\u522b\u51fa\u5e8f\u5217\u4e2d\u7684\u5404\u79cd\u5c40\u90e8\u6a21\u5f0f&#xff0c;\u5e76\u5bf9\u5e8f\u5217\u8fdb\u884c\u5e73\u6ed1\u548c\u964d\u91c7\u6837&#xff0c;\u8f93\u51fa\u4e00\u4e2a\u66f4\u77ed\u3001\u4f46\u4fe1\u606f\u66f4\u201c\u6d53\u7f29\u201d\u7684\u7279\u5f81\u5e8f\u5217\u3002<\/li>\n<li>\u7136\u540e&#xff0c;\u5c06\u8fd9\u4e2a\u7279\u5f81\u5e8f\u5217\u9001\u5165\u4e00\u4e2aLSTM\u5c42\u3002LSTM\u5219\u5728\u8fd9\u4e9b\u88ab\u63d0\u70bc\u8fc7\u7684\u7279\u5f81\u4e4b\u4e0a&#xff0c;\u8fdb\u4e00\u6b65\u5b66\u4e60\u5b83\u4eec\u4e4b\u95f4\u7684\u957f\u671f\u4f9d\u8d56\u5173\u7cfb\u3002 \u8fd9\u79cd\u201c\u5148CNN\u63d0\u53d6\u5c40\u90e8\u7279\u5f81&#xff0c;\u540eRNN\u6574\u5408\u957f\u671f\u4f9d\u8d56\u201d\u7684\u6df7\u5408\u6a21\u578b\u67b6\u6784&#xff0c;\u5728\u5f88\u591a\u590d\u6742\u7684\u5e8f\u5217\u9884\u6d4b\u4efb\u52a1\u4e0a\u90fd\u8868\u73b0\u51fa\u8272\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>import tensorflow as tf<\/p>\n<p># \u4e00\u4e2aCNN&#043;LSTM\u6df7\u5408\u6a21\u578b\u7684\u4f8b\u5b50<br \/>\nmodel &#061; tf.keras.models.Sequential([<br \/>\n    # \u8f93\u5165\u5f62\u72b6\u4e3a (\u7a97\u53e3\u5927\u5c0f, \u7279\u5f81\u7ef4\u5ea6)&#xff0c;\u8fd9\u91cc\u662f\u5355\u53d8\u91cf&#xff0c;\u6240\u4ee5\u662f1<br \/>\n    tf.keras.layers.Input(shape&#061;[None, 1]),<\/p>\n<p>    # 1. Conv1D\u5c42\u63d0\u53d6\u5c40\u90e8\u6a21\u5f0f<br \/>\n    tf.keras.layers.Conv1D(filters&#061;64, kernel_size&#061;5, strides&#061;1, padding&#061;&#034;causal&#034;,<br \/>\n                           activation&#061;&#034;relu&#034;),<\/p>\n<p>    # 2. LSTM\u5c42\u5b66\u4e60\u957f\u671f\u4f9d\u8d56<br \/>\n    tf.keras.layers.LSTM(64, return_sequences&#061;True),<br \/>\n    tf.keras.layers.LSTM(64),<\/p>\n<p>    # 3. Dense\u5c42\u8f93\u51fa\u9884\u6d4b\u503c<br \/>\n    tf.keras.layers.Dense(30, activation&#061;&#034;relu&#034;),<br \/>\n    tf.keras.layers.Dense(10, activation&#061;&#034;relu&#034;),<br \/>\n    tf.keras.layers.Dense(1)<br \/>\n])<\/p>\n<h5>13.1.4 \u8bad\u7ec3\u3001\u8bc4\u4f30\u4e0e\u73b0\u5b9e\u8003\u91cf<\/h5>\n<ul>\n<li>\n<p>\u8bc4\u4f30\u6307\u6807 \u8fd9\u662f\u4e00\u4e2a\u56de\u5f52\u4efb\u52a1&#xff0c;\u6211\u4eec\u901a\u5e38\u4f7f\u7528&#xff1a;<\/p>\n<ul>\n<li>\u5747\u65b9\u8bef\u5dee&#xff08;Mean Squared Error, MSE&#xff09;&#xff1a;\u5bf9\u8bef\u5dee\u7684\u5e73\u65b9\u8fdb\u884c\u5e73\u5747&#xff0c;\u5bf9\u5927\u8bef\u5dee\u7684\u60e9\u7f5a\u66f4\u91cd\u3002<\/li>\n<li>\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee&#xff08;Mean Absolute Error, MAE&#xff09;&#xff1a;\u5bf9\u8bef\u5dee\u7684\u7edd\u5bf9\u503c\u8fdb\u884c\u5e73\u5747&#xff0c;\u66f4\u76f4\u89c2\u5730\u53cd\u6620\u4e86\u5e73\u5747\u9884\u6d4b\u8bef\u5dee\u7684\u5927\u5c0f\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u201c\u672a\u6765\u201d\u6570\u636e\u7684\u9677\u9631&#xff1a;\u7edd\u4e0d\u80fd\u968f\u673a\u6253\u4e71&#xff01; \u8fd9\u662f\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4e2d\u6700\u91cd\u8981\u3001\u4e5f\u6700\u5bb9\u6613\u72af\u9519\u7684\u4e00\u70b9\u3002\u5728\u5212\u5206\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6\u65f6&#xff0c;\u6211\u4eec\u5fc5\u987b\u4e25\u683c\u6309\u7167\u65f6\u95f4\u987a\u5e8f\u3002<\/p>\n<ul>\n<li>\u6b63\u786e\u505a\u6cd5&#xff1a;\u53d6\u6570\u636e\u7684\u524d70%\u4f5c\u4e3a\u8bad\u7ec3\u96c6&#xff0c;\u4e2d\u95f4\u768420%\u4f5c\u4e3a\u9a8c\u8bc1\u96c6&#xff0c;\u6700\u540e\u768410%\u4f5c\u4e3a\u6d4b\u8bd5\u96c6\u3002<\/li>\n<li>\u9519\u8bef\u505a\u6cd5&#xff1a;\u50cf\u5904\u7406\u56fe\u50cf\u6570\u636e\u4e00\u6837&#xff0c;\u5c06\u6240\u6709\u7a97\u53e3\u5316\u540e\u7684\u6570\u636e\u5bf9\u968f\u673a\u6253\u4e71&#xff0c;\u518d\u5212\u5206\u8bad\u7ec3\u96c6\u548c\u9a8c\u8bc1\u96c6\u3002\u8fd9\u4f1a\u5bfc\u81f4\u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u201c\u770b\u5230\u201d\u4e86\u5b83\u672c\u4e0d\u5e94\u8be5\u770b\u5230\u7684\u201c\u672a\u6765\u201d\u6570\u636e&#xff0c;\u4ece\u800c\u5f97\u5230\u4e00\u4e2a\u865a\u9ad8\u4e14\u6beb\u65e0\u610f\u4e49\u7684\u8bc4\u4f30\u5206\u6570\u3002\u8fd9\u88ab\u79f0\u4e3a\u6570\u636e\u6cc4\u9732&#xff08;Data Leakage&#xff09;\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u73b0\u5b9e\u7684\u6b8b\u9177&#xff1a;\u4e3a\u4f55\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u5982\u6b64\u4e4b\u96be&#xff1f; \u867d\u7136\u6211\u4eec\u7528\u80a1\u7968\u6570\u636e\u4f5c\u4e3a\u4f8b\u5b50&#xff0c;\u4f46\u5fc5\u987b\u6e05\u9192\u5730\u8ba4\u8bc6\u5230&#xff0c;\u7cbe\u786e\u9884\u6d4b\u80a1\u7968\u4ef7\u683c\u51e0\u4e4e\u662f\u4e0d\u53ef\u80fd\u7684\u3002<\/p>\n<ul>\n<li>\u6709\u6548\u5e02\u573a\u5047\u8bf4&#xff1a;\u7406\u8bba\u8ba4\u4e3a&#xff0c;\u80a1\u7968\u4ef7\u683c\u5df2\u7ecf\u53cd\u6620\u4e86\u6240\u6709\u5df2\u77e5\u7684\u516c\u5f00\u4fe1\u606f\u3002\u672a\u6765\u7684\u4ef7\u683c\u6ce2\u52a8&#xff0c;\u4e3b\u8981\u7531\u65e0\u6cd5\u9884\u6d4b\u7684\u65b0\u4fe1\u606f&#xff08;\u5982\u7a81\u53d1\u65b0\u95fb\u3001\u653f\u7b56\u53d8\u5316&#xff09;\u9a71\u52a8\u3002<\/li>\n<li>\u968f\u673a\u6e38\u8d70\u7406\u8bba&#xff1a;\u8be5\u7406\u8bba\u8ba4\u4e3a&#xff0c;\u80a1\u7968\u4ef7\u683c\u7684\u77ed\u671f\u53d8\u5316\u662f\u968f\u673a\u7684&#xff0c;\u65e0\u6cd5\u6839\u636e\u5386\u53f2\u4ef7\u683c\u8fdb\u884c\u6709\u6548\u9884\u6d4b\u3002<\/li>\n<li>\u6211\u4eec\u6a21\u578b\u7684\u771f\u6b63\u4ef7\u503c&#xff1a;\u5bf9\u4e8e\u91d1\u878d\u5e02\u573a&#xff0c;\u6211\u4eec\u7684\u6a21\u578b\u5b66\u4e60\u5230\u7684&#xff0c;\u66f4\u591a\u662f\u57fa\u4e8e\u5386\u53f2\u6ce2\u52a8\u6a21\u5f0f\u7684\u7edf\u8ba1\u89c4\u5f8b\u548c\u53ef\u80fd\u6027&#xff0c;\u800c\u975e\u786e\u5b9a\u6027\u7684\u56e0\u679c\u5173\u7cfb\u3002\u5b83\u53ef\u4ee5\u4f5c\u4e3a\u4e00\u4e2a\u8f85\u52a9\u51b3\u7b56\u7684\u4fe1\u53f7&#xff08;\u4f8b\u5982&#xff0c;\u5224\u65ad\u5e02\u573a\u6ce2\u52a8\u6027\u3001\u8bc6\u522b\u8d8b\u52bf&#xff09;&#xff0c;\u4f46\u7edd\u4e0d\u80fd\u4f5c\u4e3a\u552f\u4e00\u7684\u4ea4\u6613\u4f9d\u636e\u3002\u76f8\u6bd4\u4e4b\u4e0b&#xff0c;\u9884\u6d4b\u5929\u6c14\u3001\u9500\u91cf\u7b49\u5177\u6709\u66f4\u5f3a\u7269\u7406\u6216\u793e\u4f1a\u89c4\u5f8b\u7684\u5e8f\u5217&#xff0c;\u6a21\u578b\u7684\u53ef\u9760\u6027\u4f1a\u9ad8\u5f97\u591a\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u638c\u63e1\u4e86\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b&#xff0c;\u5c31\u5982\u540c\u62e5\u6709\u4e86\u4e00\u67b6\u53ef\u4ee5\u63a2\u7d22\u65f6\u95f4\u7ef4\u5ea6\u7684\u671b\u8fdc\u955c\u3002\u867d\u7136\u5b83\u65e0\u6cd5\u770b\u6e05\u672a\u6765\u7684\u6bcf\u4e00\u4e2a\u7ec6\u8282&#xff0c;\u4f46\u5374\u80fd\u4e3a\u6211\u4eec\u63ed\u793a\u901a\u5f80\u672a\u6765\u7684\u3001\u53ef\u80fd\u6027\u6700\u9ad8\u7684\u8def\u5f84\u3002<\/p>\n<h4>13.2 \u63a8\u8350\u7cfb\u7edf&#xff1a;\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u6784\u5efa\u7535\u5f71\u6216\u5546\u54c1\u63a8\u8350\u5f15\u64ce<\/h4>\n<p>\u5728\u4e00\u4e2a\u4fe1\u606f\u7206\u70b8\u7684\u65f6\u4ee3&#xff0c;\u6211\u4eec\u9762\u4e34\u7684\u4e0d\u518d\u662f\u4fe1\u606f\u532e\u4e4f&#xff0c;\u800c\u662f\u4fe1\u606f\u8fc7\u8f7d\u3002\u65e0\u8bba\u662f\u9762\u5bf9\u89c6\u9891\u7f51\u7ad9\u4e0a\u767e\u4e07\u7684\u5f71\u7247&#xff0c;\u8fd8\u662f\u7535\u5546\u5e73\u53f0\u4e0a\u4ebf\u7684\u5546\u54c1&#xff0c;\u5982\u4f55\u5feb\u901f\u627e\u5230\u81ea\u5df1\u611f\u5174\u8da3\u7684\u5185\u5bb9&#xff0c;\u6210\u4e3a\u4e00\u4e2a\u5de8\u5927\u7684\u6311\u6218\u3002\u63a8\u8350\u7cfb\u7edf&#xff0c;\u6b63\u662f\u8fd9\u4e2a\u65f6\u4ee3\u7684\u201c\u4fe1\u606f\u5bfc\u822a\u5458\u201d\u3002<\/p>\n<h5>13.2.1 \u9879\u76ee\u6982\u8ff0&#xff1a;\u4fe1\u606f\u8fc7\u8f7d\u65f6\u4ee3\u7684\u201c\u5bfc\u822a\u5458\u201d<\/h5>\n<ul>\n<li>\n<p>\u63a8\u8350\u7cfb\u7edf\u7684\u6838\u5fc3\u4ef7\u503c \u63a8\u8350\u7cfb\u7edf\u7684\u4f7f\u547d&#xff0c;\u662f\u5728\u201c\u7528\u6237\u201d\u548c\u201c\u7269\u54c1\u201d\u4e4b\u95f4\u5efa\u7acb\u4e00\u5ea7\u9ad8\u6548\u7684\u6865\u6881\u3002<\/p>\n<ul>\n<li>\u5bf9\u7528\u6237\u800c\u8a00&#xff1a;\u5b83\u80fd\u53d1\u6398\u7528\u6237\u7684\u6f5c\u5728\u5174\u8da3&#xff0c;\u63d0\u5347\u4f53\u9a8c&#xff0c;\u5e26\u6765\u201c\u53d1\u73b0\u201d\u7684\u60ca\u559c\u3002<\/li>\n<li>\u5bf9\u5e73\u53f0\u800c\u8a00&#xff1a;\u5b83\u80fd\u589e\u52a0\u7528\u6237\u7c98\u6027&#xff0c;\u63d0\u5347\u70b9\u51fb\u7387\u3001\u8f6c\u5316\u7387&#xff0c;\u662f\u73b0\u4ee3\u4e92\u8054\u7f51\u5546\u4e1a\u6a21\u5f0f\u7684\u6838\u5fc3\u5f15\u64ce\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u7ecf\u5178\u65b9\u6cd5\u56de\u987e&#xff1a;\u534f\u540c\u8fc7\u6ee4 \u5728\u6df1\u5ea6\u5b66\u4e60\u6d41\u884c\u4e4b\u524d&#xff0c;**\u534f\u540c\u8fc7\u6ee4&#xff08;Collaborative Filtering&#xff09;**\u662f\u63a8\u8350\u7cfb\u7edf\u9886\u57df\u6700\u6210\u529f\u3001\u5e94\u7528\u6700\u5e7f\u6cdb\u7684\u601d\u60f3\u3002\u5b83\u7684\u6838\u5fc3\u5047\u8bbe\u662f\u201c\u7269\u4ee5\u7c7b\u805a&#xff0c;\u4eba\u4ee5\u7fa4\u5206\u201d\u3002<\/p>\n<ul>\n<li>\u57fa\u4e8e\u7528\u6237\u7684\u534f\u540c\u8fc7\u6ee4&#xff08;User-based CF&#xff09;&#xff1a;\u627e\u5230\u4e0e\u60a8\u54c1\u5473\u76f8\u4f3c\u7684\u7528\u6237&#xff08;\u201c\u8fd1\u90bb\u201d&#xff09;&#xff0c;\u7136\u540e\u5c06\u4ed6\u4eec\u559c\u6b22\u3001\u800c\u60a8\u8fd8\u6ca1\u770b\u8fc7\u7684\u7269\u54c1\u63a8\u8350\u7ed9\u60a8\u3002\u5176\u903b\u8f91\u662f&#xff1a;\u201c\u6211\u7684\u670b\u53cb\u559c\u6b22\u7684\u4e1c\u897f&#xff0c;\u6211\u53ef\u80fd\u4e5f\u559c\u6b22\u3002\u201d<\/li>\n<li>\u57fa\u4e8e\u7269\u54c1\u7684\u534f\u540c\u8fc7\u6ee4&#xff08;Item-based CF&#xff09;&#xff1a;\u627e\u5230\u4e0e\u60a8\u8fc7\u53bb\u559c\u6b22\u7684\u7269\u54c1\u76f8\u4f3c\u7684\u5176\u4ed6\u7269\u54c1\u3002\u5176\u903b\u8f91\u662f&#xff1a;\u201c\u559c\u6b22\u300a\u76d7\u68a6\u7a7a\u95f4\u300b\u7684\u4eba&#xff0c;\u901a\u5e38\u4e5f\u559c\u6b22\u300a\u661f\u9645\u7a7f\u8d8a\u300b\u3002\u201d \u534f\u540c\u8fc7\u6ee4\u601d\u60f3\u6734\u7d20\u800c\u6709\u6548&#xff0c;\u4f46\u5b83\u5b58\u5728\u4e00\u4e9b\u95ee\u9898&#xff0c;\u5982\u6570\u636e\u7a00\u758f\u6027\u3001\u96be\u4ee5\u5904\u7406\u65b0\u7528\u6237\u6216\u65b0\u7269\u54c1&#xff08;\u51b7\u542f\u52a8\u95ee\u9898&#xff09;\u7b49\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>13.2.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u5207\u5165\u70b9&#xff1a;\u77e9\u9635\u5206\u89e3\u4e0e\u5d4c\u5165\u601d\u60f3<\/h5>\n<p>\u6df1\u5ea6\u5b66\u4e60\u4e3a\u63a8\u8350\u7cfb\u7edf\u5e26\u6765\u4e86\u66f4\u5f3a\u5927\u3001\u66f4\u7075\u6d3b\u7684\u5efa\u6a21\u5de5\u5177\u3002\u5176\u5207\u5165\u70b9&#xff0c;\u6b63\u662f\u5c06\u534f\u540c\u8fc7\u6ee4\u7684\u6838\u5fc3\u601d\u60f3\u2014\u2014\u77e9\u9635\u5206\u89e3&#xff08;Matrix Factorization&#xff09;\u2014\u2014\u7528\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u8a00\u91cd\u65b0\u8be0\u91ca\u3002<\/p>\n<ul>\n<li>\n<p>\u77e9\u9635\u5206\u89e3 \u6211\u4eec\u53ef\u4ee5\u5c06\u6240\u6709\u7528\u6237\u7684\u8bc4\u5206\u6570\u636e&#xff0c;\u60f3\u8c61\u6210\u4e00\u4e2a\u5de8\u5927\u7684\u3001\u7a00\u758f\u7684\u201c\u7528\u6237-\u7269\u54c1\u201d\u8bc4\u5206\u77e9\u9635\u3002\u77e9\u9635\u5206\u89e3\u7684\u76ee\u6807&#xff0c;\u662f\u627e\u5230\u4e24\u4e2a\u4f4e\u7ef4\u7684\u3001\u7a20\u5bc6\u7684\u6f5c\u5728\u7279\u5f81&#xff08;Latent Factor&#xff09;\u77e9\u9635&#xff1a;\u4e00\u4e2a\u7528\u6237\u7279\u5f81\u77e9\u9635\u548c\u4e00\u4e2a\u7269\u54c1\u7279\u5f81\u77e9\u9635\u3002\u5f53\u8fd9\u4e24\u4e2a\u77e9\u9635\u76f8\u4e58\u65f6&#xff0c;\u80fd\u8fd1\u4f3c\u5730\u8fd8\u539f\u51fa\u539f\u59cb\u7684\u8bc4\u5206\u77e9\u9635\u3002<\/p>\n<ul>\n<li>\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u672c\u8d28\u4e0a\u5c31\u662f\u5728\u4e3a\u6bcf\u4e00\u4e2a\u7528\u6237\u548c\u6bcf\u4e00\u4e2a\u7269\u54c1&#xff0c;\u5b66\u4e60\u4e00\u4e2a\u5d4c\u5165\u5411\u91cf&#xff08;Embedding&#xff09;\u3002\u8fd9\u4e2a\u5411\u91cf&#xff0c;\u4ee3\u8868\u4e86\u8be5\u7528\u6237\u6216\u7269\u54c1\u5728\u67d0\u4e2a\u62bd\u8c61\u7684\u201c\u7279\u5f81\u7a7a\u95f4\u201d\u4e2d\u7684\u4f4d\u7f6e\u3002\u4f8b\u5982&#xff0c;\u4e00\u4e2a\u7528\u6237\u7684\u5411\u91cf\u53ef\u80fd\u7f16\u7801\u4e86\u4ed6\u7684\u201c\u79d1\u5e7b\u504f\u597d\u5ea6\u201d\u3001\u201c\u559c\u5267\u538c\u6076\u5ea6\u201d\u7b49&#xff1b;\u4e00\u4e2a\u7535\u5f71\u7684\u5411\u91cf\u5219\u7f16\u7801\u4e86\u5b83\u7684\u201c\u79d1\u5e7b\u6210\u5206\u201d\u3001\u201c\u6587\u827a\u5c5e\u6027\u201d\u7b49\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\u5b9e\u73b0\u77e9\u9635\u5206\u89e3 \u6211\u4eec\u53ef\u4ee5\u7528\u4e00\u4e2a\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc\u6765\u5b8c\u6210\u8fd9\u4e2a\u4efb\u52a1&#xff0c;\u8fd9\u79cd\u6a21\u578b\u901a\u5e38\u88ab\u79f0\u4e3a**\u795e\u7ecf\u534f\u540c\u8fc7\u6ee4&#xff08;Neural Collaborative Filtering, NCF&#xff09;**\u7684\u4e00\u79cd\u7b80\u5316\u5f62\u5f0f\u3002<\/p>\n<li>\u8f93\u5165&#xff1a;\u6a21\u578b\u7684\u8f93\u5165\u662f\u4e00\u4e2a(\u7528\u6237ID, \u7269\u54c1ID)\u7684\u6570\u636e\u5bf9\u3002<\/li>\n<li>\u5d4c\u5165\u5c42&#xff1a;\n<ul>\n<li>\u521b\u5efa\u4e00\u4e2a\u7528\u6237\u5d4c\u5165\u5c42&#xff0c;\u5c06\u7528\u6237ID\u6620\u5c04\u4e3a\u4e00\u4e2a\u7528\u6237\u5d4c\u5165\u5411\u91cf\u3002<\/li>\n<li>\u521b\u5efa\u4e00\u4e2a\u7269\u54c1\u5d4c\u5165\u5c42&#xff0c;\u5c06\u7269\u54c1ID\u6620\u5c04\u4e3a\u4e00\u4e2a\u7269\u54c1\u5d4c\u5165\u5411\u91cf\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u4ea4\u4e92&#xff1a;\u5c06\u5f97\u5230\u7684\u7528\u6237\u5411\u91cf\u548c\u7269\u54c1\u5411\u91cf&#xff0c;\u8fdb\u884c\u70b9\u79ef&#xff08;Dot Product&#xff09;\u64cd\u4f5c\u3002\u8fd9\u4e2a\u70b9\u79ef\u7684\u7ed3\u679c&#xff0c;\u5c31\u88ab\u6211\u4eec\u89c6\u4f5c\u6a21\u578b\u5bf9\u8be5\u7528\u6237\u7ed9\u8be5\u7269\u54c1\u7684\u9884\u6d4b\u8bc4\u5206\u3002<\/li>\n<li>\u8bad\u7ec3&#xff1a;\u5c06\u6a21\u578b\u7684\u9884\u6d4b\u8bc4\u5206\u4e0e\u771f\u5b9e\u7684\u8bc4\u5206\u8fdb\u884c\u6bd4\u8f83&#xff08;\u4f8b\u5982&#xff0c;\u4f7f\u7528\u5747\u65b9\u8bef\u5dee\u4f5c\u4e3a\u635f\u5931&#xff09;&#xff0c;\u901a\u8fc7\u53cd\u5411\u4f20\u64ad\u6765\u540c\u65f6\u5b66\u4e60\u7528\u6237\u548c\u7269\u54c1\u7684\u5d4c\u5165\u5411\u91cf\u3002<\/li>\n<p>\u8fd9\u79cd\u65b9\u6cd5\u7684\u4f18\u96c5\u4e4b\u5904\u5728\u4e8e&#xff0c;\u5b83\u5c06\u63a8\u8350\u95ee\u9898&#xff0c;\u5de7\u5999\u5730\u8f6c\u5316\u4e3a\u4e86\u4e00\u4e2a\u6807\u51c6\u7684\u76d1\u7763\u5b66\u4e60\u4efb\u52a1&#xff0c;\u5e76\u81ea\u7136\u5730\u4e3a\u6bcf\u4e2a\u7528\u6237\u548c\u7269\u54c1\u5b66\u4e60\u5230\u4e86\u5bcc\u6709\u610f\u4e49\u7684\u4f4e\u7ef4\u8868\u793a\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>13.2.3 \u6784\u5efa\u4e00\u4e2a\u66f4\u5f3a\u5927\u7684\u6df7\u5408\u63a8\u8350\u6a21\u578b<\/h5>\n<p>\u57fa\u4e8eID\u7684\u534f\u540c\u8fc7\u6ee4\u6a21\u578b\u867d\u7136\u6709\u6548&#xff0c;\u4f46\u5b83\u5ffd\u7565\u4e86\u5927\u91cf\u6709\u4ef7\u503c\u7684\u8fb9\u4fe1\u606f&#xff08;Side Information&#xff09;\u3002<\/p>\n<ul>\n<li>\n<p>\u8d85\u8d8aID \u4e00\u4e2a\u7528\u6237&#xff0c;\u9664\u4e86ID&#xff0c;\u8fd8\u6709\u5e74\u9f84\u3001\u6027\u522b\u3001\u804c\u4e1a\u7b49\u7528\u6237\u7279\u5f81\u3002\u4e00\u4ef6\u7269\u54c1&#xff0c;\u9664\u4e86ID&#xff0c;\u8fd8\u6709\u7c7b\u522b\u3001\u54c1\u724c\u3001\u4ef7\u683c\u3001\u6807\u7b7e\u7b49\u7269\u54c1\u7279\u5f81\u3002\u5c06\u8fd9\u4e9b\u7279\u5f81\u878d\u5165\u6a21\u578b&#xff0c;\u4e0d\u4ec5\u80fd\u63d0\u5347\u63a8\u8350\u7684\u51c6\u786e\u6027&#xff0c;\u8fd8\u80fd\u6781\u5927\u5730\u7f13\u89e3**\u51b7\u542f\u52a8&#xff08;Cold Start&#xff09;**\u95ee\u9898\u2014\u2014\u5373\u5f53\u4e00\u4e2a\u65b0\u7528\u6237\u6216\u65b0\u7269\u54c1\u51fa\u73b0\u65f6&#xff0c;\u7531\u4e8e\u6ca1\u6709\u4efb\u4f55\u4ea4\u4e92\u5386\u53f2&#xff0c;\u4f20\u7edf\u534f\u540c\u8fc7\u6ee4\u5c06\u675f\u624b\u65e0\u7b56&#xff0c;\u4f46\u6df7\u5408\u6a21\u578b\u4f9d\u7136\u53ef\u4ee5\u6839\u636e\u5176\u7279\u5f81\u8fdb\u884c\u521d\u6b65\u63a8\u8350\u3002<\/p>\n<\/li>\n<li>\n<p>\u6df7\u5408\u6a21\u578b&#xff08;Hybrid Model&#xff09; \u6211\u4eec\u53ef\u4ee5\u6784\u5efa\u4e00\u4e2a\u66f4\u590d\u6742\u7684\u795e\u7ecf\u7f51\u7edc\u6765\u878d\u5408\u8fd9\u4e9b\u591a\u6e90\u4fe1\u606f\u3002<\/p>\n<li>\u8f93\u5165&#xff1a;\u6a21\u578b\u7684\u8f93\u5165\u53d8\u5f97\u66f4\u52a0\u4e30\u5bcc&#xff0c;\u5305\u62ec\u7528\u6237ID\u3001\u7269\u54c1ID&#xff0c;\u4ee5\u53ca\u5404\u79cd\u7528\u6237\u548c\u7269\u54c1\u7684\u5143\u6570\u636e\u7279\u5f81\u3002<\/li>\n<li>\u7279\u5f81\u5904\u7406&#xff1a;\n<ul>\n<li>ID\u7c7b\u7279\u5f81&#xff1a;\u50cf\u4e4b\u524d\u4e00\u6837&#xff0c;\u901a\u8fc7\u5404\u81ea\u7684Embedding\u5c42\u8f6c\u6362\u4e3a\u5d4c\u5165\u5411\u91cf\u3002<\/li>\n<li>\u7c7b\u522b\u7279\u5f81&#xff08;\u5982\u7535\u5f71\u7c7b\u578b&#xff09;&#xff1a;\u540c\u6837\u53ef\u4ee5\u4f7f\u7528Embedding\u5c42\u3002<\/li>\n<li>\u6570\u503c\u7279\u5f81&#xff08;\u5982\u7528\u6237\u5e74\u9f84&#xff09;&#xff1a;\u53ef\u4ee5\u8fdb\u884c\u5f52\u4e00\u5316\u540e\u76f4\u63a5\u4f7f\u7528&#xff0c;\u6216\u5206\u7bb1\u540e\u5f53\u4f5c\u7c7b\u522b\u7279\u5f81\u5904\u7406\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u7279\u5f81\u878d\u5408&#xff1a;\u5c06\u5904\u7406\u540e\u7684\u6240\u6709\u7279\u5f81\u5411\u91cf**\u62fc\u63a5&#xff08;Concatenate&#xff09;**\u5728\u4e00\u8d77&#xff0c;\u5f62\u6210\u4e00\u4e2a\u5de8\u5927\u7684\u3001\u4fe1\u606f\u4e30\u5bcc\u7684\u5355\u4e00\u5411\u91cf\u3002<\/li>\n<li>\u6df1\u5ea6\u7f51\u7edc&#xff1a;\u5c06\u8fd9\u4e2a\u878d\u5408\u540e\u7684\u5411\u91cf&#xff0c;\u9001\u5165\u4e00\u4e2a\u6807\u51c6\u7684\u591a\u5c42\u611f\u77e5\u673a&#xff08;MLP&#xff09;\u4e2d\u3002\u8fd9\u4e2aMLP\u7684\u4f5c\u7528&#xff0c;\u662f\u5b66\u4e60\u8fd9\u4e9b\u4e0d\u540c\u6765\u6e90\u7279\u5f81\u4e4b\u95f4\u590d\u6742\u7684\u3001\u975e\u7ebf\u6027\u7684\u4ea4\u4e92\u5173\u7cfb\u3002<\/li>\n<li>\u8f93\u51fa&#xff1a;\u7f51\u7edc\u7684\u6700\u540e\u4e00\u5c42\u662f\u4e00\u4e2a\u5355\u4e00\u7684\u795e\u7ecf\u5143&#xff0c;\u8f93\u51fa\u6700\u7ec8\u7684\u9884\u6d4b\u8bc4\u5206\u3002<\/li>\n<p>\u8fd9\u79cd\u6df7\u5408\u6a21\u578b\u67b6\u6784\u6781\u5176\u7075\u6d3b&#xff0c;\u53ef\u4ee5\u65b9\u4fbf\u5730\u878d\u5165\u4efb\u610f\u7c7b\u578b\u7684\u7279\u5f81&#xff0c;\u662f\u5de5\u4e1a\u754c\u63a8\u8350\u7cfb\u7edf\u5e38\u7528\u7684\u5efa\u6a21\u8303\u5f0f\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>13.2.4 \u5b9e\u8df5\u4e0e\u8bc4\u4f30<\/h5>\n<ul>\n<li>\n<p>\u6570\u636e\u96c6 MovieLens\u662f\u7531GroupLens\u7814\u7a76\u5c0f\u7ec4\u53d1\u5e03\u7684\u4e00\u7cfb\u5217\u5173\u4e8e\u7535\u5f71\u8bc4\u5206\u7684\u6570\u636e\u96c6&#xff0c;\u662f\u63a8\u8350\u7cfb\u7edf\u7814\u7a76\u6700\u7ecf\u5178\u7684\u201cHello, World\u201d\u6570\u636e\u96c6\u3002\u5b83\u5305\u542b\u4e86\u7528\u6237\u5bf9\u7535\u5f71\u7684\u8bc4\u5206\u3001\u7535\u5f71\u7684\u7c7b\u578b\u6807\u7b7e\u4ee5\u53ca\u7528\u6237\u7684\u57fa\u672c\u4fe1\u606f\u3002<\/p>\n<\/li>\n<li>\n<p>\u8bc4\u4f30\u6307\u6807 \u8fd9\u662f\u4e00\u4e2a\u56de\u5f52\u4efb\u52a1&#xff0c;\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528&#xff1a;<\/p>\n<ul>\n<li>\u5747\u65b9\u6839\u8bef\u5dee&#xff08;Root Mean Squared Error, RMSE&#xff09;<\/li>\n<li>\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee&#xff08;Mean Absolute Error, MAE&#xff09;\u00a0\u8fd9\u4e24\u4e2a\u6307\u6807\u90fd\u7528\u4e8e\u8861\u91cf\u6a21\u578b\u9884\u6d4b\u8bc4\u5206\u4e0e\u771f\u5b9e\u8bc4\u5206\u4e4b\u95f4\u7684\u5dee\u8ddd\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4ece\u8bc4\u5206\u9884\u6d4b\u5230Top-N\u63a8\u8350 \u5728\u771f\u5b9e\u5e94\u7528\u4e2d&#xff0c;\u6211\u4eec\u901a\u5e38\u4e0d\u662f\u8981\u9884\u6d4b\u7528\u6237\u5bf9\u67d0\u4e2a\u7279\u5b9a\u7269\u54c1\u7684\u8bc4\u5206&#xff0c;\u800c\u662f\u8981\u4e3a\u4ed6\u751f\u6210\u4e00\u4e2a\u63a8\u8350\u5217\u8868&#xff08;Top-N Recommendation&#xff09;\u3002<\/p>\n<ul>\n<li>\u5b9e\u73b0\u65b9\u6cd5&#xff1a;\u5f53\u4e00\u4e2a\u7528\u6237\u6765\u8bbf\u65f6&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u53bb\u8ba1\u7b97\u4ed6\u5bf9\u6240\u6709&#xff08;\u6216\u4e00\u4e2a\u5019\u9009\u5b50\u96c6&#xff09;\u4ed6\u5c1a\u672a\u4ea4\u4e92\u8fc7\u7684\u7269\u54c1\u7684\u9884\u6d4b\u8bc4\u5206\u3002\u7136\u540e&#xff0c;\u5c06\u8fd9\u4e9b\u9884\u6d4b\u8bc4\u5206\u4ece\u9ad8\u5230\u4f4e\u6392\u5e8f&#xff0c;\u9009\u53d6\u524dN\u4e2a\u7269\u54c1&#xff0c;\u4f5c\u4e3a\u6700\u7ec8\u7684\u63a8\u8350\u5217\u8868\u5448\u73b0\u7ed9\u7528\u6237\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u63a8\u8350\u7cfb\u7edf\u662f\u6280\u672f\u4e0e\u5546\u4e1a\u7ed3\u5408\u6700\u7d27\u5bc6\u7684\u9886\u57df\u4e4b\u4e00\u3002\u901a\u8fc7\u8fd9\u4e2a\u9879\u76ee&#xff0c;\u6211\u4eec\u4e0d\u4ec5\u5b66\u4e60\u4e86\u5982\u4f55\u7528\u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u5d4c\u5165\u201d\u601d\u60f3\u6765\u7406\u89e3\u7528\u6237\u548c\u7269\u54c1&#xff0c;\u66f4\u7aa5\u89c1\u4e86\u9a71\u52a8\u73b0\u4ee3\u4e92\u8054\u7f51\u4e2a\u6027\u5316\u4f53\u9a8c\u7684\u6838\u5fc3\u6280\u672f\u3002<\/p>\n<hr \/>\n<h4>13.3 \u5f3a\u5316\u5b66\u4e60\u5165\u95e8&#xff1a;\u4f7f\u7528\u6df1\u5ea6Q\u7f51\u7edc&#xff08;DQN&#xff09;\u73a9\u8f6c\u7b80\u5355\u6e38\u620f \u2014\u2014 \u5728\u8bd5\u9519\u4e2d\u5b66\u4e60\u667a\u6167<\/h4>\n<p>\u6211\u4eec\u672c\u7ae0\u7684\u6700\u540e\u4e00\u4e2a\u9879\u76ee&#xff0c;\u5c06\u8fdb\u5165\u4e00\u4e2a\u601d\u60f3\u8303\u5f0f\u622a\u7136\u4e0d\u540c\u7684\u9886\u57df\u2014\u2014\u5f3a\u5316\u5b66\u4e60&#xff08;Reinforcement Learning, RL&#xff09;\u3002\u5728\u8fd9\u91cc&#xff0c;\u6211\u4eec\u4e0d\u518d\u4e3a\u6a21\u578b\u63d0\u4f9b\u201c\u6b63\u786e\u7b54\u6848\u201d&#xff0c;\u800c\u662f\u8bbe\u5b9a\u4e00\u4e2a\u201c\u76ee\u6807\u201d&#xff0c;\u7136\u540e\u653e\u624b\u8ba9\u5b83\u5728\u4e0e\u73af\u5883\u7684\u4e92\u52a8\u4e2d&#xff0c;\u81ea\u5df1\u53bb\u5b66\u4e60\u5982\u4f55\u8fbe\u6210\u8fd9\u4e2a\u76ee\u6807\u3002\u8fd9\u662f\u4e00\u79cd\u66f4\u63a5\u8fd1\u751f\u7269\u5b66\u4e60\u672c\u8d28\u7684\u3001\u66f4\u4e3b\u52a8\u7684\u667a\u6167\u3002<\/p>\n<h5>13.3.1 \u601d\u60f3\u7684\u8f6c\u53d8&#xff1a;\u4ece\u76d1\u7763\u5b66\u4e60\u5230\u5f3a\u5316\u5b66\u4e60<\/h5>\n<ul>\n<li>\n<p>\u5f3a\u5316\u5b66\u4e60\u7684\u6838\u5fc3\u8303\u5f0f \u60f3\u8c61\u4e00\u4e2a\u6b63\u5728\u5b66\u4e60\u9a91\u81ea\u884c\u8f66\u7684\u5b69\u5b50\u3002\u60a8\u4e0d\u4f1a\u544a\u8bc9\u4ed6\u6bcf\u4e00\u6beb\u79d2\u5e94\u8be5\u4ee5\u4ec0\u4e48\u89d2\u5ea6\u8f6c\u52a8\u8f66\u628a\u3001\u7528\u591a\u5927\u7684\u529b\u6c14\u8e6c\u811a\u8e0f\u677f\u3002\u60a8\u53ea\u4f1a\u7ed9\u4ed6\u4e00\u4e2a\u76ee\u6807&#xff08;\u201c\u522b\u6454\u5012&#xff0c;\u5f80\u524d\u8d70\u201d&#xff09;&#xff0c;\u7136\u540e\u8ba9\u4ed6\u81ea\u5df1\u53bb\u5c1d\u8bd5\u3002\u4ed6\u4f1a\u6454\u5012&#xff08;\u8d1f\u5956\u52b1&#xff09;&#xff0c;\u4e5f\u4f1a\u6210\u529f\u524d\u8fdb&#xff08;\u6b63\u5956\u52b1&#xff09;\u3002\u901a\u8fc7\u4e0d\u65ad\u7684\u8bd5\u9519&#xff0c;\u4ed6\u6700\u7ec8\u4f1a\u5b66\u4f1a\u3002\u8fd9\u5c31\u662f\u5f3a\u5316\u5b66\u4e60\u3002<\/p>\n<ul>\n<li>\u667a\u80fd\u4f53&#xff08;Agent&#xff09;&#xff1a;\u5b66\u4e60\u8005\u548c\u51b3\u7b56\u8005&#xff08;\u5b69\u5b50\/\u6211\u4eec\u7684\u7b97\u6cd5&#xff09;\u3002<\/li>\n<li>\u73af\u5883&#xff08;Environment&#xff09;&#xff1a;\u667a\u80fd\u4f53\u6240\u5904\u7684\u5916\u90e8\u4e16\u754c&#xff08;\u7269\u7406\u4e16\u754c\/\u6e38\u620f\u4e16\u754c&#xff09;\u3002<\/li>\n<li>\u72b6\u6001&#xff08;State&#xff09;&#xff1a;\u5bf9\u73af\u5883\u5728\u67d0\u4e00\u65f6\u523b\u7684\u63cf\u8ff0&#xff08;\u81ea\u884c\u8f66\u7684\u4f4d\u7f6e\u3001\u901f\u5ea6\u3001\u503e\u89d2&#xff09;\u3002<\/li>\n<li>\u884c\u52a8&#xff08;Action&#xff09;&#xff1a;\u667a\u80fd\u4f53\u53ef\u4ee5\u91c7\u53d6\u7684\u64cd\u4f5c&#xff08;\u8f6c\u52a8\u8f66\u628a\u3001\u8e6c\u811a\u8e0f\u677f&#xff09;\u3002<\/li>\n<li>\u5956\u52b1&#xff08;Reward&#xff09;&#xff1a;\u73af\u5883\u5bf9\u667a\u80fd\u4f53\u884c\u52a8\u7684\u53cd\u9988\u4fe1\u53f7&#xff08;\u524d\u8fdb\u5f97\u5230\u6b63\u5956\u52b1&#xff0c;\u6454\u5012\u5f97\u5230\u8d1f\u5956\u52b1&#xff09;\u3002<\/li>\n<li>\u7b56\u7565&#xff08;Policy&#xff09;&#xff1a;\u667a\u80fd\u4f53\u7684\u201c\u5927\u8111\u201d&#xff0c;\u5373\u4e00\u4e2a\u4ece\u72b6\u6001\u5230\u884c\u52a8\u7684\u6620\u5c04&#xff0c;\u51b3\u5b9a\u4e86\u5728\u7279\u5b9a\u72b6\u6001\u4e0b\u5e94\u8be5\u91c7\u53d6\u4ec0\u4e48\u884c\u52a8\u3002 RL\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u5b66\u4e60\u4e00\u4e2a\u6700\u4f18\u7b56\u7565&#xff0c;\u4ee5\u6700\u5927\u5316\u957f\u671f\u7d2f\u79ef\u5956\u52b1\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4e0e\u76d1\u7763\u5b66\u4e60\u7684\u533a\u522b<\/p>\n<ul>\n<li>\u53cd\u9988\u4fe1\u53f7&#xff1a;\u76d1\u7763\u5b66\u4e60\u6709\u660e\u786e\u7684\u3001\u5373\u65f6\u7684\u201c\u6807\u7b7e\u201d&#xff0c;\u800cRL\u53ea\u6709\u5ef6\u8fdf\u7684\u3001\u7a00\u758f\u7684\u201c\u5956\u52b1\u201d\u3002\u60a8\u53ef\u80fd\u505a\u4e86\u5f88\u591a\u6b65\u6b63\u786e\u7684\u64cd\u4f5c&#xff0c;\u6700\u540e\u624d\u5f97\u5230\u4e00\u4e2a\u5956\u52b1\u3002<\/li>\n<li>\u6570\u636e\u6765\u6e90&#xff1a;\u76d1\u7763\u5b66\u4e60\u7684\u6570\u636e\u662f\u7ed9\u5b9a\u7684\u3001\u9759\u6001\u7684&#xff0c;\u800cRL\u7684\u6570\u636e\u662f\u667a\u80fd\u4f53\u901a\u8fc7\u4e0e\u73af\u5883\u4e3b\u52a8\u4ea4\u4e92\u751f\u6210\u7684&#xff0c;\u662f\u4e00\u4e2a\u52a8\u6001\u7684\u8fc7\u7a0b\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>13.3.2 Q-Learning\u4e0e\u6df1\u5ea6Q\u7f51\u7edc&#xff08;DQN&#xff09;\u7684\u7cbe\u9ad3<\/h5>\n<ul>\n<li>\n<p>Q-Learning \u8fd9\u662f\u4e00\u79cd\u7ecf\u5178\u7684\u3001\u57fa\u4e8e\u4ef7\u503c&#xff08;Value-based&#xff09;\u7684RL\u7b97\u6cd5\u3002\u5b83\u4e0d\u76f4\u63a5\u5b66\u4e60\u7b56\u7565&#xff0c;\u800c\u662f\u5b66\u4e60\u4e00\u4e2a\u52a8\u4f5c\u4ef7\u503c\u51fd\u6570Q(s, a)\u3002<\/p>\n<ul>\n<li>Q(s, a)\u7684\u542b\u4e49&#xff1a;\u5b83\u4ee3\u8868\u4e86\u5728\u72b6\u6001s\u4e0b&#xff0c;\u91c7\u53d6\u884c\u52a8a&#xff0c;\u5e76\u4e14\u4e4b\u540e\u90fd\u9075\u5faa\u6700\u4f18\u7b56\u7565&#xff0c;\u6240\u80fd\u83b7\u5f97\u7684\u672a\u6765\u603b\u56de\u62a5\u7684\u671f\u671b\u503c\u3002<\/li>\n<li>\u51b3\u7b56&#xff1a;\u4e00\u65e6\u6211\u4eec\u6709\u4e86\u6700\u4f18\u7684Q\u51fd\u6570&#xff0c;\u7b56\u7565\u5c31\u53d8\u5f97\u975e\u5e38\u7b80\u5355&#xff1a;\u5728\u4efb\u4f55\u72b6\u6001s\u4e0b&#xff0c;\u6211\u4eec\u53ea\u9700\u8981\u9009\u62e9\u90a3\u4e2a\u80fd\u4f7fQ(s, a)\u503c\u6700\u5927\u7684\u884c\u52a8a\u5373\u53ef\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6df1\u5ea6Q\u7f51\u7edc&#xff08;DQN&#xff09; \u5bf9\u4e8e\u7b80\u5355\u7684\u73af\u5883&#xff08;\u5982\u8ff7\u5bab&#xff09;&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528\u4e00\u4e2a\u8868\u683c&#xff08;Q-Table&#xff09;\u6765\u5b58\u50a8\u6240\u6709(s, a)\u5bf9\u7684Q\u503c\u3002\u4f46\u5bf9\u4e8e\u590d\u6742\u73af\u5883&#xff0c;\u6bd4\u5982\u96c5\u8fbe\u5229&#xff08;Atari&#xff09;\u6e38\u620f&#xff0c;\u5176\u72b6\u6001\u662f\u539f\u59cb\u7684\u5c4f\u5e55\u50cf\u7d20&#xff0c;\u72b6\u6001\u7a7a\u95f4\u51e0\u4e4e\u662f\u65e0\u9650\u7684\u3002\u8fd9\u65f6&#xff0c;\u8868\u683c\u65b9\u6cd5\u5931\u6548\u4e86\u3002 **DQN&#xff08;Deep Q-Network&#xff09;**\u7684\u9769\u547d\u6027\u8d21\u732e\u5728\u4e8e&#xff1a;\u4f7f\u7528\u4e00\u4e2a\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u6765\u8fd1\u4f3c\u8fd9\u4e2aQ\u51fd\u6570\u3002<\/p>\n<ul>\n<li>\u7f51\u7edc\u8f93\u5165&#xff1a;\u73af\u5883\u7684\u72b6\u6001&#xff08;\u4f8b\u5982&#xff0c;\u6e38\u620f\u753b\u9762\u7684\u51e0\u5e27\u56fe\u50cf&#xff09;\u3002<\/li>\n<li>\u7f51\u7edc\u8f93\u51fa&#xff1a;\u8be5\u72b6\u6001\u4e0b&#xff0c;\u6bcf\u4e00\u4e2a\u53ef\u80fd\u52a8\u4f5c\u7684Q\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>13.3.3 DQN\u7684\u4e24\u5927\u201c\u6cd5\u5b9d\u201d&#xff1a;\u7ecf\u9a8c\u56de\u653e\u4e0e\u76ee\u6807\u7f51\u7edc<\/h5>\n<p>\u76f4\u63a5\u7528\u795e\u7ecf\u7f51\u7edc\u62df\u5408Q\u51fd\u6570\u4f1a\u5bfc\u81f4\u8bad\u7ec3\u975e\u5e38\u4e0d\u7a33\u5b9a\u30022015\u5e74&#xff0c;DeepMind\u63d0\u51fa\u7684DQN\u5f15\u5165\u4e86\u4e24\u4e2a\u5173\u952e\u6280\u672f&#xff0c;\u6210\u529f\u5730\u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898&#xff0c;\u5f00\u542f\u4e86\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u7684\u65f6\u4ee3\u3002<\/p>\n<ul>\n<li>\n<p>\u7ecf\u9a8c\u56de\u653e&#xff08;Experience Replay&#xff09; \u667a\u80fd\u4f53\u4e0e\u73af\u5883\u4ea4\u4e92\u4ea7\u751f\u7684\u6570\u636e\u662f\u9ad8\u5ea6\u65f6\u95f4\u76f8\u5173\u7684&#xff0c;\u76f4\u63a5\u6309\u987a\u5e8f\u5b66\u4e60\u4f1a\u5bfc\u81f4\u6a21\u578b\u9677\u5165\u5c40\u90e8\u6700\u4f18\u3002<\/p>\n<ul>\n<li>\u505a\u6cd5&#xff1a;\u667a\u80fd\u4f53\u5c06\u81ea\u5df1\u7ecf\u5386\u8fc7\u7684\u6bcf\u4e00\u6b65\u2014\u2014\u5373(\u72b6\u6001, \u52a8\u4f5c, \u5956\u52b1, \u4e0b\u4e00\u72b6\u6001)\u56db\u5143\u7ec4\u2014\u2014\u90fd\u5b58\u50a8\u5230\u4e00\u4e2a\u56fa\u5b9a\u5927\u5c0f\u7684\u201c\u8bb0\u5fc6\u6c60\u201d&#xff08;Replay Buffer&#xff09;\u4e2d\u3002\u5728\u8bad\u7ec3\u65f6&#xff0c;\u6211\u4eec\u4e0d\u518d\u4f7f\u7528\u521a\u521a\u4ea7\u751f\u7684\u6570\u636e&#xff0c;\u800c\u662f\u4ece\u8fd9\u4e2a\u8bb0\u5fc6\u6c60\u4e2d\u968f\u673a\u62bd\u53d6\u4e00\u4e2a\u5c0f\u6279\u91cf&#xff08;mini-batch&#xff09;\u7684\u6570\u636e\u6765\u8fdb\u884c\u5b66\u4e60\u3002<\/li>\n<li>\u597d\u5904&#xff1a;1. \u6253\u7834\u4e86\u6570\u636e\u76f8\u5173\u6027&#xff0c;\u4f7f\u8bad\u7ec3\u6837\u672c\u66f4\u63a5\u8fd1\u72ec\u7acb\u540c\u5206\u5e03\u30022. \u63d0\u9ad8\u4e86\u6570\u636e\u5229\u7528\u7387&#xff0c;\u4e00\u6761\u7ecf\u9a8c\u53ef\u4ee5\u88ab\u91cd\u590d\u5b66\u4e60\u591a\u6b21\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u76ee\u6807\u7f51\u7edc&#xff08;Target Network&#xff09; \u5728\u8ba1\u7b97Q\u5b66\u4e60\u7684\u76ee\u6807\u503c\u65f6&#xff0c;\u6211\u4eec\u9700\u8981\u7528\u5230Q\u7f51\u7edc\u81ea\u8eab\u7684\u9884\u6d4b&#xff0c;\u8fd9\u4f1a\u5bfc\u81f4\u4e00\u4e2a\u201c\u8ffd\u9010\u81ea\u5df1\u5c3e\u5df4\u201d\u7684\u95ee\u9898&#xff0c;\u4f7f\u5f97\u5b66\u4e60\u76ee\u6807\u4e0d\u65ad\u6447\u6446&#xff0c;\u96be\u4ee5\u6536\u655b\u3002<\/p>\n<ul>\n<li>\u505a\u6cd5&#xff1a;\u6211\u4eec\u521b\u5efa\u4e24\u4e2a\u7ed3\u6784\u5b8c\u5168\u76f8\u540c\u7684Q\u7f51\u7edc\u3002\u4e00\u4e2a\u662f\u6211\u4eec\u6b63\u5728\u79ef\u6781\u8bad\u7ec3\u7684\u4e3b\u7f51\u7edc&#xff08;Main Network&#xff09;&#xff0c;\u53e6\u4e00\u4e2a\u662f\u76ee\u6807\u7f51\u7edc&#xff08;Target Network&#xff09;\u3002\u5728\u8ba1\u7b97\u76ee\u6807Q\u503c\u65f6&#xff0c;\u6211\u4eec\u4f7f\u7528\u56fa\u5b9a\u7684\u76ee\u6807\u7f51\u7edc&#xff1b;\u5728\u66f4\u65b0\u6743\u91cd\u65f6&#xff0c;\u6211\u4eec\u53ea\u66f4\u65b0\u4e3b\u7f51\u7edc\u3002\u6bcf\u9694\u4e00\u5b9a\u7684\u6b65\u6570&#xff0c;\u6211\u4eec\u518d\u5c06\u4e3b\u7f51\u7edc\u7684\u6743\u91cd\u590d\u5236\u7ed9\u76ee\u6807\u7f51\u7edc\u3002<\/li>\n<li>\u597d\u5904&#xff1a;\u8fd9\u76f8\u5f53\u4e8e\u5728\u4e00\u6bb5\u65f6\u95f4\u5185&#xff0c;\u5c06\u5b66\u4e60\u76ee\u6807\u201c\u51bb\u7ed3\u201d\u4f4f&#xff0c;\u5927\u5927\u589e\u52a0\u4e86\u8bad\u7ec3\u7684\u7a33\u5b9a\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>13.3.4 \u5b9e\u6218&#xff1a;\u4f7f\u7528DQN\u73a9\u8f6cCartPole\u6216Atari\u6e38\u620f<\/h5>\n<ul>\n<li>\n<p>\u73af\u5883 \u6211\u4eec\u5c06\u4f7f\u7528Gymnasium&#xff08;\u7531OpenAI Gym\u53d1\u5c55\u800c\u6765&#xff09;\u8fd9\u4e2aPython\u5e93\u3002\u5b83\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u5927\u91cf\u6807\u51c6\u5316\u7684RL\u6d4b\u8bd5\u73af\u5883&#xff0c;\u4ece\u7b80\u5355\u7684CartPole&#xff08;\u63a7\u5236\u5c0f\u8f66\u4e0a\u7684\u6746\u5b50\u4fdd\u6301\u5e73\u8861&#xff09;\u5230\u590d\u6742\u7684Atari\u6e38\u620f&#xff0c;\u90fd\u5c01\u88c5\u4e86\u7edf\u4e00\u7684API\u63a5\u53e3\u3002<\/p>\n<\/li>\n<li>\n<p>\u7f16\u7801\u5b9e\u73b0 \u6784\u5efa\u4e00\u4e2aDQN\u667a\u80fd\u4f53&#xff0c;\u9700\u8981\u5b9e\u73b0\u4ee5\u4e0b\u6838\u5fc3\u7ec4\u4ef6&#xff1a;<\/p>\n<li>Q\u7f51\u7edc&#xff1a;\u4e00\u4e2a\u7b80\u5355\u7684\u5168\u8fde\u63a5\u7f51\u7edc&#xff08;\u5bf9\u4e8eCartPole&#xff09;\u6216\u5377\u79ef\u7f51\u7edc&#xff08;\u5bf9\u4e8eAtari&#xff09;\u3002<\/li>\n<li>\u8bb0\u5fc6\u6c60&#xff1a;\u4e00\u4e2a\u961f\u5217\u6216\u5217\u8868&#xff0c;\u7528\u4e8e\u5b58\u50a8\u7ecf\u9a8c\u3002<\/li>\n<li>\u52a8\u4f5c\u9009\u62e9\u7b56\u7565&#xff1a;\u901a\u5e38\u4f7f\u7528\u03b5-greedy\u7b56\u7565\u3002\u5373\u4ee5\u03b5\u7684\u6982\u7387\u968f\u673a\u9009\u62e9\u4e00\u4e2a\u52a8\u4f5c&#xff08;\u9f13\u52b1\u63a2\u7d22&#xff09;&#xff0c;\u4ee51-\u03b5\u7684\u6982\u7387\u9009\u62e9\u5f53\u524dQ\u503c\u6700\u9ad8\u7684\u52a8\u4f5c&#xff08;\u5229\u7528\u5df2\u77e5\u6700\u4f18\u7b56\u7565&#xff09;\u3002\u03b5\u7684\u503c\u4f1a\u968f\u7740\u8bad\u7ec3\u7684\u8fdb\u884c\u800c\u9010\u6e10\u8870\u51cf\u3002<\/li>\n<li>\u5b66\u4e60\u903b\u8f91&#xff1a;\u4ece\u8bb0\u5fc6\u6c60\u4e2d\u91c7\u6837&#xff0c;\u8ba1\u7b97\u635f\u5931&#xff08;\u901a\u5e38\u662fHuber Loss&#xff09;&#xff0c;\u5e76\u66f4\u65b0\u4e3b\u7f51\u7edc\u3002<\/li>\n<\/li>\n<li>\n<p>\u8bad\u7ec3\u5faa\u73af \u6574\u4e2a\u8fc7\u7a0b\u662f\u4e00\u4e2a\u5927\u7684\u5faa\u73af&#xff1a;<\/p>\n<li>\u521d\u59cb\u5316\u73af\u5883&#xff0c;\u83b7\u5f97\u521d\u59cb\u72b6\u6001\u3002<\/li>\n<li>\u5bf9\u4e8e\u6bcf\u4e00\u8f6e&#xff08;episode&#xff09;&#xff1a; a. \u667a\u80fd\u4f53\u6839\u636e\u5f53\u524d\u72b6\u6001\u548c\u03b5-greedy\u7b56\u7565\u9009\u62e9\u4e00\u4e2a\u52a8\u4f5c\u3002 b. \u5728\u73af\u5883\u4e2d\u6267\u884c\u8be5\u52a8\u4f5c&#xff0c;\u83b7\u5f97\u5956\u52b1\u3001\u4e0b\u4e00\u72b6\u6001\u548c\u662f\u5426\u7ed3\u675f\u7684\u4fe1\u53f7\u3002 c. \u5c06\u8fd9\u4e2a\u7ecf\u9a8c\u5b58\u5165\u8bb0\u5fc6\u6c60\u3002 d. \u4ece\u8bb0\u5fc6\u6c60\u4e2d\u91c7\u6837&#xff0c;\u8fdb\u884c\u4e00\u6b21\u7f51\u7edc\u8bad\u7ec3\u3002 e. \u5468\u671f\u6027\u5730\u66f4\u65b0\u76ee\u6807\u7f51\u7edc\u3002 f. \u76f4\u5230\u8be5\u8f6e\u7ed3\u675f\u3002<\/li>\n<\/li>\n<li>\n<p>\u7ed3\u679c \u6211\u4eec\u901a\u8fc7\u7ed8\u5236\u6bcf\u4e00\u8f6e\u83b7\u5f97\u7684\u7d2f\u79ef\u5956\u52b1\u66f2\u7ebf&#xff0c;\u6765\u89c2\u5bdf\u667a\u80fd\u4f53\u7684\u5b66\u4e60\u8fdb\u7a0b\u3002\u60a8\u4f1a\u770b\u5230\u4e00\u6761\u4ee4\u4eba\u632f\u594b\u7684\u66f2\u7ebf&#xff1a;\u4e00\u5f00\u59cb&#xff0c;\u667a\u80fd\u4f53\u50cf\u4e2a\u65e0\u5934\u82cd\u8747&#xff0c;\u5f97\u5206\u5f88\u4f4e&#xff1b;\u4f46\u968f\u7740\u8bad\u7ec3\u7684\u8fdb\u884c&#xff0c;\u5956\u52b1\u66f2\u7ebf\u4f1a\u7a33\u6b65\u4e0a\u5347&#xff0c;\u6700\u7ec8\u8fbe\u5230\u751a\u81f3\u8d85\u8fc7\u4eba\u7c7b\u6c34\u5e73\u3002\u8fd9\u751f\u52a8\u5730\u5c55\u793a\u4e86\u667a\u80fd\u4f53\u662f\u5982\u4f55\u5728\u7eaf\u7cb9\u7684\u8bd5\u9519\u4e2d&#xff0c;\u6d8c\u73b0\u51fa\u771f\u6b63\u7684\u201c\u667a\u6167\u201d\u7684\u3002<\/p>\n<\/li>\n<\/ul>\n<hr \/>\n<p>\u5c0f\u7ed3&#xff1a;\u667a\u6167\u7684\u5f62\u6001\u662f\u591a\u6837\u7684<\/p>\n<p>\u5728\u672c\u7ae0\u7684\u65c5\u7a0b\u4e2d&#xff0c;\u6211\u4eec\u4e09\u6b21\u8de8\u8d8a\u4e86\u5b66\u79d1\u7684\u8fb9\u754c&#xff0c;\u5c06\u6df1\u5ea6\u5b66\u4e60\u7684\u706b\u79cd\u64ad\u6492\u5230\u4e86\u5168\u65b0\u7684\u571f\u58e4\u3002<\/p>\n<ul>\n<li>\u5728\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4e2d&#xff0c;\u6211\u4eec\u5b66\u4f1a\u4e86\u5982\u4f55\u5c0a\u91cd\u548c\u5904\u7406\u65f6\u95f4\u7684\u7279\u6b8a\u5c5e\u6027&#xff0c;\u5229\u7528RNN\u548cCNN\u7684\u7ec4\u5408&#xff0c;\u5728\u5386\u53f2\u7684\u957f\u6cb3\u4e2d\u5bfb\u627e\u672a\u6765\u7684\u7ebf\u7d22\u3002<\/li>\n<li>\u5728\u63a8\u8350\u7cfb\u7edf\u4e2d&#xff0c;\u6211\u4eec\u8fd0\u7528\u201c\u5d4c\u5165\u201d\u8fd9\u4e00\u6838\u5fc3\u601d\u60f3&#xff0c;\u5c06\u7528\u6237\u548c\u7269\u54c1\u6620\u5c04\u5230\u540c\u4e00\u4e2a\u8bed\u4e49\u7a7a\u95f4&#xff0c;\u4ece\u800c\u6d1e\u5bdf\u4e86\u4eba\u5fc3\u7684\u9009\u62e9\u4e0e\u504f\u597d\u3002<\/li>\n<li>\u5728\u5f3a\u5316\u5b66\u4e60\u4e2d&#xff0c;\u6211\u4eec\u5f7b\u5e95\u98a0\u8986\u4e86\u76d1\u7763\u5b66\u4e60\u7684\u8303\u5f0f&#xff0c;\u901a\u8fc7\u6784\u5efa\u4e00\u4e2a\u5728\u4e0e\u73af\u5883\u4ea4\u4e92\u4e2d\u81ea\u6211\u5b66\u4e60\u7684\u667a\u80fd\u4f53&#xff0c;\u89c1\u8bc1\u4e86\u667a\u6167\u5982\u4f55\u5728\u8bd5\u9519\u4e0e\u5956\u52b1\u7684\u9a71\u52a8\u4e0b\u81ea\u53d1\u6d8c\u73b0\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e09\u4e2a\u9879\u76ee\u5171\u540c\u5411\u6211\u4eec\u63ed\u793a\u4e86\u6df1\u5ea6\u5b66\u4e60\u4f5c\u4e3a\u4e00\u79cd\u901a\u7528\u95ee\u9898\u6c42\u89e3\u8303\u5f0f\u7684\u5f3a\u5927\u5a01\u529b\u3002\u5176\u6838\u5fc3\u601d\u60f3\u2014\u2014\u65e0\u8bba\u662f\u5c42\u6b21\u5316\u7279\u5f81\u8868\u793a\u3001\u5e8f\u5217\u4fe1\u606f\u5efa\u6a21&#xff0c;\u8fd8\u662f\u4ef7\u503c\u51fd\u6570\u8fd1\u4f3c\u2014\u2014\u90fd\u5982\u540c\u53ef\u4ee5\u81ea\u7531\u7ec4\u5408\u7684\u201c\u4e50\u9ad8\u79ef\u6728\u201d&#xff0c;\u80fd\u591f\u88ab\u7075\u6d3b\u5730\u5e94\u7528\u4e8e\u89e3\u51b3\u5404\u79cd\u770b\u4f3c\u6beb\u4e0d\u76f8\u5173\u7684\u590d\u6742\u95ee\u9898\u3002<\/p>\n<p>\u4e0d\u8981\u5c06\u81ea\u5df1\u5c40\u9650\u4e8e\u8ba1\u7b97\u673a\u89c6\u89c9\u6216\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7684\u201c\u8212\u9002\u533a\u201d\u3002\u771f\u6b63\u7684\u521b\u65b0&#xff0c;\u5f80\u5f80\u53d1\u751f\u5728\u601d\u60f3\u7684\u78b0\u649e\u4e0e\u5b66\u79d1\u7684\u4ea4\u53c9\u5730\u5e26\u3002\u5e26\u7740\u5728\u672c\u7ae0\u5b66\u5230\u7684\u5168\u65b0\u89c6\u89d2&#xff0c;\u53bb\u89c2\u5bdf\u60a8\u8eab\u8fb9\u7684\u4e16\u754c&#xff0c;\u53bb\u601d\u8003\u8fd8\u6709\u54ea\u4e9b\u9886\u57df\u3001\u54ea\u4e9b\u95ee\u9898&#xff0c;\u53ef\u4ee5\u88ab\u6df1\u5ea6\u5b66\u4e60\u7684\u667a\u6167\u4e4b\u5149\u6240\u7167\u4eae\u3002\u8fd9\u5c06\u662f\u60a8\u4ece\u4e00\u540d\u4f18\u79c0\u7684\u5b66\u4e60\u8005&#xff0c;\u8715\u53d8\u4e3a\u4e00\u540d\u6770\u51fa\u7684\u521b\u9020\u8005\u7684\u5f00\u59cb\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u5341\u56db\u7ae0&#xff1a;\u6a21\u578b\u90e8\u7f72\u4e0e\u5de5\u7a0b\u5316<\/h3>\n<p>\u6a21\u578b\u7684\u201c\u51fa\u5c71\u201d\u4e4b\u65c5<\/p>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u81f3\u6b64&#xff0c;\u6211\u4eec\u5df2\u7ecf\u4e00\u540c\u8d70\u8fc7\u4e86\u6df1\u5ea6\u5b66\u4e60\u4e16\u754c\u7684\u4e07\u6c34\u5343\u5c71\u3002\u6211\u4eec\u5b66\u4f1a\u4e86\u5982\u4f55\u642d\u5efa\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u9aa8\u67b6\u201d&#xff0c;\u5982\u4f55\u7528\u6570\u636e\u8fd9\u5473\u201c\u4ed9\u8349\u201d\u8fdb\u884c\u6ecb\u517b&#xff0c;\u5982\u4f55\u8fd0\u7528\u201c\u70bc\u4e39\u672f\u201d\u822c\u7684\u4f18\u5316\u6280\u5de7&#xff0c;\u6700\u7ec8\u8bad\u7ec3\u51fa\u6027\u80fd\u5f3a\u5927\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u3002\u5728Jupyter Notebook\u4e2d&#xff0c;\u5f53val_accuracy\u8fbe\u5230\u4e00\u4e2a\u4ee4\u4eba\u6ee1\u610f\u7684\u6570\u5b57\u65f6&#xff0c;\u6211\u4eec\u6216\u8bb8\u4f1a\u611f\u5230\u5fc3\u6ee1\u610f\u8db3\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u8fd9\u53ea\u662f\u6545\u4e8b\u7684\u5f00\u59cb&#xff0c;\u7edd\u975e\u7ed3\u675f\u3002<\/p>\n<p>\u4e00\u4e2a\u5728\u6df1\u5c71\u4e2d\u70bc\u6210\u7684\u7edd\u4e16\u9ad8\u624b&#xff0c;\u82e5\u60f3\u540d\u626c\u5929\u4e0b\u3001\u884c\u4fa0\u4ed7E\u4e49&#xff0c;\u7ec8\u987b\u201c\u51fa\u5c71\u201d&#xff0c;\u6b65\u5165\u6c5f\u6e56\u3002\u540c\u6837&#xff0c;\u4e00\u4e2a\u5728\u5b9e\u9a8c\u5ba4\u91cc\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u82e5\u60f3\u771f\u6b63\u521b\u9020\u4ef7\u503c\u3001\u8d4b\u80fd\u767e\u4e1a&#xff0c;\u4e5f\u5fc5\u987b\u5f00\u542f\u5b83\u7684\u201c\u51fa\u5c71\u201d\u4e4b\u65c5\u3002\u8fd9\u4e2a\u65c5\u7a0b&#xff0c;\u4fbf\u662f\u6a21\u578b\u90e8\u7f72\u4e0e\u5de5\u7a0b\u5316\u3002\u5b83\u5173\u4e4e\u5982\u4f55\u5c06\u6211\u4eec\u667a\u6167\u7684\u7ed3\u6676&#xff0c;\u8f6c\u5316\u4e3a\u4e00\u4e2a\u7a33\u5b9a\u3001\u9ad8\u6548\u3001\u53ef\u6269\u5c55\u3001\u53ef\u7ef4\u62a4\u7684\u751f\u4ea7\u529b\u670d\u52a1&#xff0c;\u53bb\u8fce\u63a5\u771f\u5b9e\u4e16\u754c\u4e2d\u6210\u5343\u4e0a\u4e07\u6b21\u8c03\u7528\u7684\u8003\u9a8c\u3002<\/p>\n<p>\u4e00\u4e2a\u6210\u529f\u7684AI\u5e94\u7528&#xff0c;\u7b97\u6cd5\u7684\u7cbe\u5999\u56fa\u7136\u91cd\u8981&#xff0c;\u4f46\u5b83\u4ec5\u4ec5\u662f\u51b0\u5c71\u7684\u4e00\u89d2\u3002\u6c34\u9762\u4e4b\u4e0b&#xff0c;\u662f\u5e9e\u5927\u800c\u575a\u5b9e\u7684\u5de5\u7a0b\u5b9e\u8df5\u5728\u652f\u6491\u3002\u672c\u7ae0&#xff0c;\u6211\u4eec\u5c06\u805a\u7126\u4e8e\u8fd9\u201c\u6c34\u4e0b\u201d\u7684\u90e8\u5206&#xff0c;\u5b8c\u6210\u4ece\u201c\u70bc\u4e39\u201d\u5230\u201c\u6d4e\u4e16\u201d\u7684\u8fd9\u201c\u6700\u540e\u4e00\u516c\u91cc\u201d\u3002\u6211\u4eec\u5c06\u5b66\u4e60&#xff1a;<\/p>\n<ul>\n<li>\u5982\u4f55\u4e3a\u6211\u4eec\u65e5\u76ca\u201c\u81c3\u80bf\u201d\u7684\u6a21\u578b\u8fdb\u884c\u8f7b\u91cf\u5316\u201c\u7626\u8eab\u201d&#xff0c;\u8ba9\u5b83\u80fd\u98de\u5165\u5bfb\u5e38\u767e\u59d3\u5bb6&#xff0c;\u5728\u624b\u673a\u3001\u5728\u8fb9\u7f18\u8bbe\u5907\u4e0a\u8fd0\u884c\u3002<\/li>\n<li>\u5982\u4f55\u5c06\u6a21\u578b\u90e8\u7f72\u5230\u4e13\u4e1a\u7684\u670d\u52a1\u6846\u67b6\u4e2d&#xff0c;\u8ba9\u5b83\u62e5\u6709\u4e00\u4e2a\u575a\u56fa\u3001\u9ad8\u6027\u80fd\u7684\u201c\u5bb6\u201d\u3002<\/li>\n<li>\u5982\u4f55\u5c06\u5176\u5c01\u88c5\u4e3a\u6807\u51c6\u7684API\u670d\u52a1&#xff0c;\u4e3a\u5b83\u5efa\u7acb\u4e00\u5ea7\u4e0e\u4e16\u754c\u4e07\u7269\u6c9f\u901a\u7684\u201c\u6865\u6881\u201d\u3002<\/li>\n<li>\u6700\u540e&#xff0c;\u6211\u4eec\u5c06\u4e86\u89e3MLOps\u7684\u5b8f\u5927\u7406\u5ff5&#xff0c;\u5b66\u4e60\u5982\u4f55\u7cfb\u7edf\u5316\u5730\u7ba1\u7406AI\u7684\u6574\u4e2a\u751f\u547d\u5468\u671f&#xff0c;\u6210\u4e3a\u4e00\u540d\u771f\u6b63\u7684AI\u201c\u5efa\u7b51\u5e08\u201d\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u8d9f\u65c5\u7a0b&#xff0c;\u662f\u7406\u8bba\u4e0e\u73b0\u5b9e\u7684\u4ea4\u6c47&#xff0c;\u662f\u7b97\u6cd5\u4e0e\u5de5\u7a0b\u7684\u878d\u5408\u3002\u5b83\u5c06\u8003\u9a8c\u6211\u4eec\u7684\u4e25\u8c28\u3001\u7ec6\u81f4\u4e0e\u8fdc\u89c1\u3002\u51c6\u5907\u597d&#xff0c;\u8ba9\u6211\u4eec\u62a4\u9001\u4eb2\u624b\u521b\u9020\u7684\u6a21\u578b&#xff0c;\u5b8c\u6210\u5b83\u4ece\u201c\u5c55\u54c1\u201d\u5230\u201c\u5de5\u5177\u201d\u7684\u4f1f\u5927\u8715\u53d8\u3002<\/p>\n<h4>14.1 \u6a21\u578b\u8f7b\u91cf\u5316&#xff1a;\u4e3a\u6a21\u578b\u201c\u7626\u8eab\u201d&#xff0c;\u4ee5\u9002\u5e94\u66f4\u5e7f\u9614\u7684\u5929\u5730<\/h4>\n<p>\u6211\u4eec\u8ffd\u6c42\u66f4\u6df1\u3001\u66f4\u590d\u6742\u7684\u7f51\u7edc&#xff0c;\u4ee5\u671f\u83b7\u5f97\u66f4\u9ad8\u7684\u7cbe\u5ea6&#xff0c;\u8fd9\u672c\u8eab\u65e0\u53ef\u539a\u975e\u3002\u4f46\u8fd9\u4efd\u201c\u667a\u6167\u201d\u7684\u4ee3\u4ef7&#xff0c;\u662f\u6a21\u578b\u53c2\u6570\u91cf\u7684\u7206\u70b8\u5f0f\u589e\u957f\u548c\u5de8\u5927\u7684\u8ba1\u7b97\u5f00\u9500\u3002\u4e00\u4e2a\u52a8\u8f84\u6570GB\u7684\u6a21\u578b&#xff0c;\u5c31\u50cf\u4e00\u4f4d\u8eab\u7a7f\u91cd\u94e0\u7684\u5c06\u519b&#xff0c;\u867d\u5a01\u529b\u65e0\u7a77&#xff0c;\u5374\u96be\u4ee5\u5728\u72ed\u7a84\u7684\u8857\u5df7\u4e2d\u7075\u6d3b\u4f5c\u6218\u3002\u82e5\u60f3\u8ba9AI\u7684\u80fd\u529b\u904d\u53ca\u624b\u673a\u3001\u667a\u80fd\u6444\u50cf\u5934\u3001\u7269\u8054\u7f51\u8bbe\u5907\u7b49\u201c\u8857\u5934\u5df7\u5c3e\u201d&#xff0c;\u5c31\u5fc5\u987b\u4e3a\u6a21\u578b\u201c\u7626\u8eab\u201d\u3002<\/p>\n<h5>14.1.1 \u4e3a\u4f55\u9700\u8981\u8f7b\u91cf\u5316&#xff1f;\u6027\u80fd\u3001\u6210\u672c\u4e0e\u573a\u666f\u7684\u8003\u91cf<\/h5>\n<p>\u6a21\u578b\u8f7b\u91cf\u5316\u7684\u9a71\u52a8\u529b&#xff0c;\u6e90\u4e8e\u73b0\u5b9e\u4e16\u754c\u4e2d\u65e0\u5904\u4e0d\u5728\u7684\u201c\u9650\u5236\u201d\u3002<\/p>\n<ul>\n<li>\u6027\u80fd\u4e0e\u5ef6\u8fdf&#xff08;Performance &amp; Latency&#xff09;&#xff1a;\u5728\u81ea\u52a8\u9a7e\u9a76\u3001\u5b9e\u65f6\u8bed\u97f3\u8bc6\u522b\u7b49\u573a\u666f&#xff0c;\u96f6\u70b9\u51e0\u79d2\u7684\u5ef6\u8fdf\u90fd\u53ef\u80fd\u5bfc\u81f4\u707e\u96be\u6027\u540e\u679c\u3002\u8f7b\u91cf\u5316\u80fd\u663e\u8457\u52a0\u5feb\u6a21\u578b\u7684\u63a8\u7406\u901f\u5ea6&#xff0c;\u6ee1\u8db3\u4e25\u82db\u7684\u5b9e\u65f6\u6027\u8981\u6c42\u3002<\/li>\n<li>\u6210\u672c\u4e0e\u529f\u8017&#xff08;Cost &amp; Power&#xff09;&#xff1a;\u5728\u4e91\u7aef&#xff0c;\u66f4\u5feb\u7684\u63a8\u7406\u901f\u5ea6\u610f\u5473\u7740\u66f4\u5c11\u7684\u8ba1\u7b97\u8d44\u6e90\u5360\u7528&#xff0c;\u76f4\u63a5\u964d\u4f4e\u670d\u52a1\u6210\u672c\u3002\u5728\u79fb\u52a8\u7aef\u6216\u8fb9\u7f18\u8bbe\u5907\u4e0a&#xff0c;\u66f4\u4f4e\u7684\u8ba1\u7b97\u91cf\u610f\u5473\u7740\u66f4\u5c11\u7684\u80fd\u8017\u548c\u66f4\u957f\u7684\u7eed\u822a\u65f6\u95f4\u3002<\/li>\n<li>\u90e8\u7f72\u73af\u5883&#xff08;Deployment Environment&#xff09;&#xff1a;\u624b\u673a\u3001\u53ef\u7a7f\u6234\u8bbe\u5907\u3001\u5de5\u4e1a\u4f20\u611f\u5668\u7b49\u8fb9\u7f18\u8bbe\u5907\u7684\u8ba1\u7b97\u80fd\u529b\u548c\u5185\u5b58\u90fd\u975e\u5e38\u6709\u9650\u3002\u4e00\u4e2a\u5e9e\u5927\u7684\u6a21\u578b\u6839\u672c\u65e0\u6cd5\u5728\u8fd9\u4e9b\u8bbe\u5907\u4e0a\u8fd0\u884c\u3002<\/li>\n<\/ul>\n<p>\u56e0\u6b64&#xff0c;\u6a21\u578b\u8f7b\u91cf\u5316\u7684\u76ee\u6807\u975e\u5e38\u660e\u786e&#xff1a;\u5728\u5c3d\u53ef\u80fd\u4fdd\u6301\u6a21\u578b\u7cbe\u5ea6\u7684\u524d\u63d0\u4e0b&#xff0c;\u51cf\u5c0f\u6a21\u578b\u4f53\u79ef\u3001\u964d\u4f4e\u8ba1\u7b97\u9700\u6c42\u3001\u52a0\u5feb\u63a8\u7406\u901f\u5ea6\u3002 \u8fd9\u662f\u4e00\u95e8\u5728\u201c\u6027\u80fd\u201d\u4e0e\u201c\u6548\u7387\u201d\u4e4b\u95f4\u5bfb\u6c42\u6700\u4f73\u5e73\u8861\u7684\u827a\u672f\u3002<\/p>\n<h5>14.1.2 \u526a\u679d&#xff08;Pruning&#xff09;&#xff1a;\u526a\u53bb\u201c\u5197\u4f59\u201d\u7684\u667a\u6167<\/h5>\n<p>\u5982\u540c\u56ed\u4e01\u4fee\u526a\u82b1\u6728&#xff0c;\u526a\u53bb\u67af\u679d\u8d25\u53f6&#xff0c;\u65b9\u80fd\u8ba9\u517b\u5206\u96c6\u4e2d\u4e8e\u4e3b\u5e72&#xff0c;\u4f7f\u82b1\u6735\u5f00\u5f97\u66f4\u76db\u3002\u795e\u7ecf\u7f51\u7edc\u7684\u526a\u679d&#xff0c;\u4ea6\u662f\u540c\u7406\u3002<\/p>\n<ul>\n<li>\n<p>\u6838\u5fc3\u601d\u60f3 \u7814\u7a76\u53d1\u73b0&#xff0c;\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e2d\u5b58\u5728\u5927\u91cf\u7684\u53c2\u6570\u5197\u4f59\u3002\u8bb8\u591a\u6743\u91cd\u8fde\u63a5\u7684\u6570\u503c\u975e\u5e38\u63a5\u8fd1\u4e8e\u96f6&#xff0c;\u5b83\u4eec\u5bf9\u7f51\u7edc\u7684\u6700\u7ec8\u8f93\u51fa\u8d21\u732e\u751a\u5fae\u3002\u526a\u679d\u6280\u672f\u7684\u6838\u5fc3&#xff0c;\u5c31\u662f\u8bc6\u522b\u5e76\u5b89\u5168\u5730\u79fb\u9664\u8fd9\u4e9b\u4e0d\u91cd\u8981\u7684\u8fde\u63a5\u6216\u795e\u7ecf\u5143&#xff0c;\u4ece\u800c\u5728\u4e0d\u663e\u8457\u5f71\u54cd\u7cbe\u5ea6\u7684\u524d\u63d0\u4e0b&#xff0c;\u5f97\u5230\u4e00\u4e2a\u66f4\u5c0f\u3001\u66f4\u7a00\u758f\u7684\u7f51\u7edc\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b9e\u73b0\u65b9\u5f0f \u6700\u5e38\u89c1\u7684**\u6743\u91cd\u526a\u679d&#xff08;Weight Pruning&#xff09;**\u6d41\u7a0b&#xff0c;\u5982\u540c\u4e00\u573a\u201c\u672b\u4f4d\u6dd8\u6c70\u201d\u7684\u8003\u6838&#xff1a;<\/p>\n<li>\u8bad\u7ec3\u4e00\u4e2a\u5927\u6a21\u578b&#xff1a;\u9996\u5148&#xff0c;\u6b63\u5e38\u5730\u8bad\u7ec3\u4e00\u4e2a\u5b8c\u6574\u7684\u3001\u5bc6\u96c6\u7684\u201c\u6559\u5e08\u201d\u6a21\u578b&#xff0c;\u4f7f\u5176\u8fbe\u5230\u8f83\u9ad8\u7684\u7cbe\u5ea6\u3002<\/li>\n<li>\u8bc4\u4f30\u6743\u91cd\u91cd\u8981\u6027&#xff1a;\u4e3a\u7f51\u7edc\u4e2d\u7684\u6bcf\u4e00\u4e2a\u6743\u91cd\u8fde\u63a5\u6253\u5206\u3002\u6700\u7b80\u5355\u76f4\u63a5\u7684\u65b9\u6cd5&#xff0c;\u5c31\u662f\u4f7f\u7528\u5176\u7edd\u5bf9\u503c\u7684\u5927\u5c0f\u4f5c\u4e3a\u91cd\u8981\u6027\u7684\u8861\u91cf\u6807\u51c6\u3002\u7edd\u5bf9\u503c\u8d8a\u5927\u7684\u6743\u91cd&#xff0c;\u88ab\u8ba4\u4e3a\u8d8a\u91cd\u8981\u3002<\/li>\n<li>\u79fb\u9664\u4e0d\u91cd\u8981\u7684\u6743\u91cd&#xff1a;\u8bbe\u5b9a\u4e00\u4e2a\u526a\u679d\u9608\u503c&#xff08;\u4f8b\u5982&#xff0c;\u79fb\u9664\u7edd\u5bf9\u503c\u6700\u5c0f\u768420%\u7684\u6743\u91cd&#xff09;&#xff0c;\u5c06\u8fd9\u4e9b\u6743\u91cd\u7684\u503c\u5f3a\u5236\u8bbe\u4e3a0\u3002\u6b64\u65f6&#xff0c;\u6211\u4eec\u7684\u6743\u91cd\u77e9\u9635\u5c31\u53d8\u6210\u4e86\u4e00\u4e2a\u7a00\u758f\u77e9\u9635\u3002<\/li>\n<li>\u5fae\u8c03&#xff08;Fine-tuning&#xff09;&#xff1a;\u526a\u679d\u64cd\u4f5c\u4f1a\u6682\u65f6\u6027\u5730\u964d\u4f4e\u6a21\u578b\u7cbe\u5ea6\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u9700\u8981\u7528\u8f83\u4f4e\u7684\u5b66\u4e60\u7387&#xff0c;\u5728\u539f\u59cb\u6570\u636e\u96c6\u4e0a\u5bf9\u8fd9\u4e2a\u88ab\u526a\u679d\u7684\u7a00\u758f\u6a21\u578b\u8fdb\u884c\u51e0\u8f6e\u989d\u5916\u7684\u8bad\u7ec3\u3002\u8fd9\u4e2a\u8fc7\u7a0b&#xff0c;\u80fd\u8ba9\u5269\u4f59\u7684\u6743\u91cd\u8fdb\u884c\u5fae\u8c03&#xff0c;\u4ee5\u8865\u507f\u88ab\u79fb\u9664\u6743\u91cd\u6240\u9020\u6210\u7684\u5f71\u54cd&#xff0c;\u4ece\u800c\u6062\u590d\u5927\u90e8\u5206\u5931\u53bb\u7684\u7cbe\u5ea6\u3002 \u8fd9\u4e2a\u201c\u8bad\u7ec3-\u526a\u679d-\u5fae\u8c03\u201d\u7684\u5faa\u73af\u53ef\u4ee5\u8fed\u4ee3\u8fdb\u884c&#xff0c;\u76f4\u5230\u6a21\u578b\u5927\u5c0f\u548c\u7cbe\u5ea6\u8fbe\u5230\u7406\u60f3\u7684\u5e73\u8861\u70b9\u3002<\/li>\n<\/li>\n<\/ul>\n<h5>14.1.3 \u91cf\u5316&#xff08;Quantization&#xff09;&#xff1a;\u4ece\u201c\u6d6e\u70b9\u201d\u5230\u201c\u6574\u6570\u201d\u7684\u7cbe\u70bc<\/h5>\n<p>\u5982\u679c\u8bf4\u526a\u679d\u662f\u5728\u201c\u51cf\u5c11\u201d\u53c2\u6570\u7684\u6570\u91cf&#xff0c;\u90a3\u4e48\u91cf\u5316\u5219\u662f\u5728\u201c\u538b\u7f29\u201d\u6bcf\u4e00\u4e2a\u53c2\u6570\u7684\u4f53\u79ef\u3002\u5b83\u662f\u4e00\u95e8\u5c06\u201c\u7c97\u7cae\u7ec6\u4f5c\u201d\u7684\u827a\u672f\u3002<\/p>\n<ul>\n<li>\n<p>\u6838\u5fc3\u601d\u60f3 \u9ed8\u8ba4\u60c5\u51b5\u4e0b&#xff0c;\u795e\u7ecf\u7f51\u7edc\u7684\u6743\u91cd\u548c\u6fc0\u6d3b\u503c\u90fd\u4ee532\u4f4d\u6d6e\u70b9\u6570&#xff08;FP32&#xff09;\u7684\u683c\u5f0f\u5b58\u50a8\u3002\u91cf\u5316\u6280\u672f&#xff0c;\u5c31\u662f\u5c06\u8fd9\u4e9b\u9ad8\u7cbe\u5ea6\u7684\u6d6e\u70b9\u6570&#xff0c;\u7528\u4e00\u79cd\u6620\u5c04\u5173\u7cfb&#xff0c;\u8f6c\u6362\u4e3a\u66f4\u4f4e\u7cbe\u5ea6\u7684\u6570\u503c\u7c7b\u578b&#xff0c;\u6700\u5e38\u89c1\u7684\u662f8\u4f4d\u5b9a\u70b9\u6574\u6570&#xff08;INT8&#xff09;\u3002<\/p>\n<\/li>\n<li>\n<p>\u5e26\u6765\u7684\u597d\u5904 \u8fd9\u79cd\u770b\u4f3c\u7b80\u5355\u7684\u7cbe\u5ea6\u8f6c\u6362&#xff0c;\u5374\u80fd\u5e26\u6765\u5de8\u5927\u7684\u6536\u76ca&#xff1a;<\/p>\n<ul>\n<li>\u6a21\u578b\u4f53\u79ef\u51cf\u5c0f&#xff1a;\u4ece32\u4f4d\u52308\u4f4d&#xff0c;\u6a21\u578b\u7684\u5927\u5c0f\u53ef\u4ee5\u76f4\u63a5\u538b\u7f29\u5230\u539f\u6765\u76841\/4\u3002<\/li>\n<li>\u8ba1\u7b97\u901f\u5ea6\u63d0\u5347&#xff1a;\u8bb8\u591a\u73b0\u4ee3CPU\u548c\u4e13\u7528AI\u82af\u7247&#xff08;\u5982Google\u7684TPU\u3001NVIDIA\u7684Tensor Core&#xff09;\u5bf9\u6574\u6570\u8fd0\u7b97\u7684\u901f\u5ea6&#xff0c;\u8fdc\u5feb\u4e8e\u6d6e\u70b9\u8fd0\u7b97\u3002\u56e0\u6b64&#xff0c;\u91cf\u5316\u540e\u7684\u6a21\u578b\u63a8\u7406\u901f\u5ea6\u53ef\u4ee5\u5f97\u52302\u52304\u500d\u751a\u81f3\u66f4\u9ad8\u7684\u63d0\u5347\u3002<\/li>\n<li>\u5185\u5b58\u4e0e\u529f\u8017\u964d\u4f4e&#xff1a;\u66f4\u5c0f\u7684\u6570\u636e\u7c7b\u578b\u610f\u5473\u7740\u66f4\u5c11\u7684\u5185\u5b58\u5360\u7528\u548c\u66f4\u4f4e\u7684\u6570\u636e\u642c\u8fd0\u529f\u8017\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5b9e\u73b0\u65b9\u5f0f \u91cf\u5316\u4e3b\u8981\u6709\u4e24\u79cd\u4e3b\u6d41\u6280\u672f\u8def\u5f84&#xff1a;<\/p>\n<ul>\n<li>\u8bad\u7ec3\u540e\u91cf\u5316&#xff08;Post-Training Quantization, PTQ&#xff09;&#xff1a;\u8fd9\u662f\u6700\u7b80\u5355\u76f4\u63a5\u7684\u65b9\u5f0f\u3002\u6211\u4eec\u62ff\u4e00\u4e2a\u5df2\u7ecf\u8bad\u7ec3\u597d\u7684FP32\u6a21\u578b&#xff0c;\u901a\u8fc7\u4e00\u5c0f\u90e8\u5206\u6821\u51c6\u6570\u636e\u6765\u786e\u5b9a\u6d6e\u70b9\u6570\u5230\u6574\u6570\u7684\u6620\u5c04\u8303\u56f4&#xff0c;\u7136\u540e\u76f4\u63a5\u5bf9\u6a21\u578b\u8fdb\u884c\u8f6c\u6362\u3002\u5b83\u65e0\u9700\u91cd\u65b0\u8bad\u7ec3&#xff0c;\u64cd\u4f5c\u4fbf\u6377&#xff0c;\u4f46\u53ef\u80fd\u4f1a\u5e26\u6765\u4e00\u5b9a\u7684\u7cbe\u5ea6\u635f\u5931\u3002<\/li>\n<li>\u91cf\u5316\u611f\u77e5\u8bad\u7ec3&#xff08;Quantization-Aware Training, QAT&#xff09;&#xff1a;\u4e3a\u4e86\u5f25\u8865PTQ\u53ef\u80fd\u5e26\u6765\u7684\u7cbe\u5ea6\u635f\u5931&#xff0c;QAT\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5c31\u201c\u6a21\u62df\u201d\u4e86\u91cf\u5316\u64cd\u4f5c\u3002\u5b83\u5728\u6a21\u578b\u7684\u524d\u5411\u4f20\u64ad\u4e2d&#xff0c;\u6a21\u62df\u91cf\u5316\u548c\u53cd\u91cf\u5316\u7684\u8fc7\u7a0b&#xff0c;\u8ba9\u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u5c31\u63d0\u524d\u9002\u5e94\u4e86\u4f4e\u7cbe\u5ea6\u8ba1\u7b97\u53ef\u80fd\u5e26\u6765\u7684\u566a\u58f0\u548c\u8bef\u5dee\u3002\u8fd9\u6837\u8bad\u7ec3\u51fa\u6765\u7684\u6a21\u578b&#xff0c;\u5728\u8f6c\u6362\u4e3a\u771f\u6b63\u7684INT8\u6a21\u578b\u65f6&#xff0c;\u51e0\u4e4e\u6ca1\u6709\u7cbe\u5ea6\u635f\u5931\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>14.1.4 \u77e5\u8bc6\u84b8\u998f&#xff08;Knowledge Distillation&#xff09;&#xff1a;\u8ba9\u201c\u5927\u201d\u6a21\u578b\u6559\u201c\u5c0f\u201d\u6a21\u578b<\/h5>\n<p>\u8fd9\u662f\u4e00\u79cd\u6e90\u4e8e\u201c\u6a21\u4eff\u201d\u548c\u201c\u4f20\u627f\u201d\u7684\u667a\u6167\u3002\u6211\u4eec\u4e0d\u76f4\u63a5\u53bb\u4f18\u5316\u4e00\u4e2a\u5c0f\u6a21\u578b&#xff0c;\u800c\u662f\u8ba9\u4e00\u4e2a\u5b66\u8bc6\u6e0a\u535a\u7684\u201c\u8001\u5e08\u201d&#xff0c;\u6765\u624b\u628a\u624b\u5730\u6559\u4e00\u4e2a\u5929\u8d44\u806a\u9896\u7684\u201c\u5b66\u751f\u201d\u3002<\/p>\n<ul>\n<li>\n<p>\u6838\u5fc3\u601d\u60f3 \u6211\u4eec\u5148\u8bad\u7ec3\u4e00\u4e2a\u6027\u80fd\u5f3a\u5927\u4f46\u7ed3\u6784\u590d\u6742\u3001\u53c2\u6570\u5e9e\u5927\u7684\u6559\u5e08\u6a21\u578b&#xff08;Teacher Model&#xff09;\u3002\u7136\u540e&#xff0c;\u6211\u4eec\u8bbe\u8ba1\u4e00\u4e2a\u7ed3\u6784\u7b80\u5355\u3001\u53c2\u6570\u91cf\u5c0f\u7684\u5b66\u751f\u6a21\u578b&#xff08;Student Model&#xff09;\u3002\u77e5\u8bc6\u84b8\u998f\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u8ba9\u8fd9\u4e2a\u5b66\u751f\u6a21\u578b&#xff0c;\u53bb\u5b66\u4e60\u548c\u6a21\u4eff\u6559\u5e08\u6a21\u578b\u7684\u201c\u884c\u4e3a\u201d&#xff0c;\u4ece\u800c\u4ee5\u5c0f\u5de7\u7684\u8eab\u8eaf&#xff0c;\u8fbe\u5230\u6216\u63a5\u8fd1\u6559\u5e08\u6a21\u578b\u7684\u6027\u80fd\u3002<\/p>\n<\/li>\n<li>\n<p>\u5b9e\u73b0\u65b9\u5f0f \u5982\u4f55\u201c\u6307\u5bfc\u201d\u5462&#xff1f;\u5173\u952e\u5728\u4e8e&#xff0c;\u5b66\u751f\u4e0d\u4ec5\u4ec5\u5b66\u4e60\u5ba2\u89c2\u771f\u7406&#xff0c;\u66f4\u8981\u5b66\u4e60\u8001\u5e08\u7684\u201c\u601d\u8003\u8fc7\u7a0b\u201d\u3002<\/p>\n<li>\u786c\u6807\u7b7e&#xff08;Hard Labels&#xff09;&#xff1a;\u8fd9\u662f\u6211\u4eec\u901a\u5e38\u7684\u8bad\u7ec3\u76ee\u6807&#xff0c;\u5373\u6570\u636e\u771f\u5b9e\u7684\u6807\u7b7e&#xff08;\u5982\u201c\u732b\u201d\u3001\u201c\u72d7\u201d&#xff09;\u3002\u5b66\u751f\u6a21\u578b\u5f53\u7136\u8981\u5b66\u4e60\u8fd9\u4e2a\u3002<\/li>\n<li>\u8f6f\u6807\u7b7e&#xff08;Soft Labels&#xff09;&#xff1a;\u8fd9\u662f\u77e5\u8bc6\u84b8\u998f\u7684\u6838\u5fc3\u3002\u5b83\u662f\u6307\u6559\u5e08\u6a21\u578b\u5728softmax\u5c42\u8f93\u51fa\u7684\u5b8c\u6574\u7684\u6982\u7387\u5206\u5e03\u3002\u8fd9\u4e2a\u5206\u5e03&#xff0c;\u8574\u542b\u4e86\u6559\u5e08\u6a21\u578b\u66f4\u4e30\u5bcc\u7684\u4fe1\u606f\u3002\u4f8b\u5982&#xff0c;\u5bf9\u4e8e\u4e00\u5f20\u732b\u7684\u56fe\u7247&#xff0c;\u6559\u5e08\u6a21\u578b\u53ef\u80fd\u8ba4\u4e3a\u5b83\u670990%\u7684\u53ef\u80fd\u662f\u732b&#xff0c;\u4f46\u540c\u65f6\u67095%\u7684\u53ef\u80fd\u662f\u72d7&#xff0c;1%\u7684\u53ef\u80fd\u662f\u8001\u864e\u3002\u8fd9\u4e2a\u201c5%\u7684\u72d7\u201d\u548c\u201c1%\u7684\u8001\u864e\u201d\u7684\u4fe1\u606f&#xff0c;\u5c31\u63ed\u793a\u4e86\u6559\u5e08\u6a21\u578b\u8ba4\u4e3a\u201c\u732b\u201d\u4e0e\u201c\u72d7\u201d\u3001\u201c\u8001\u864e\u201d\u4e4b\u95f4\u5b58\u5728\u67d0\u79cd\u76f8\u4f3c\u6027\u3002<\/li>\n<li>\u8054\u5408\u8bad\u7ec3&#xff1a;\u5b66\u751f\u6a21\u578b\u7684\u6700\u7ec8\u635f\u5931\u51fd\u6570&#xff0c;\u662f\u5176\u5728\u786c\u6807\u7b7e\u4e0a\u7684\u635f\u5931\u548c\u5728\u8f6f\u6807\u7b7e\u4e0a\u7684\u635f\u5931\u7684\u4e00\u4e2a\u52a0\u6743\u548c\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f&#xff0c;\u5b66\u751f\u6a21\u578b\u4e0d\u4ec5\u88ab\u5f3a\u5236\u5b66\u4e60\u6b63\u786e\u7684\u7b54\u6848&#xff0c;\u66f4\u88ab\u5f15\u5bfc\u53bb\u6a21\u4eff\u6559\u5e08\u6a21\u578b\u5bf9\u4e0d\u540c\u7c7b\u522b\u4e4b\u95f4\u76f8\u4f3c\u6027\u7684\u7406\u89e3\u3002<\/li>\n<\/li>\n<\/ul>\n<p>\u77e5\u8bc6\u84b8\u998f\u662f\u4e00\u79cd\u975e\u5e38\u6709\u6548\u7684\u6a21\u578b\u538b\u7f29\u548c\u8fc1\u79fb\u5b66\u4e60\u6280\u672f&#xff0c;\u5b83\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u4e00\u79cd\u5728\u4e0d\u727a\u7272\u592a\u591a\u6027\u80fd\u7684\u524d\u63d0\u4e0b&#xff0c;\u83b7\u5f97\u8f7b\u91cf\u5316\u6a21\u578b\u7684\u65b0\u601d\u8def\u3002<\/p>\n<p>\u901a\u8fc7\u526a\u679d\u3001\u91cf\u5316\u3001\u77e5\u8bc6\u84b8\u998f\u8fd9\u4e09\u5927\u201c\u6cd5\u672f\u201d&#xff0c;\u6211\u4eec\u4fbf\u80fd\u5c06\u4e00\u4e2a\u5e9e\u5927\u800c\u7b28\u91cd\u7684\u6a21\u578b&#xff0c;\u4fee\u70bc\u6210\u4e00\u4e2a\u8eab\u8f7b\u5982\u71d5\u3001\u5feb\u5982\u95ea\u7535\u7684\u9ad8\u624b&#xff0c;\u4e3a\u5b83\u63a5\u4e0b\u6765\u7684\u201c\u51fa\u5c71\u201d\u4e4b\u65c5&#xff0c;\u94fa\u5e73\u4e86\u9053\u8def\u3002<\/p>\n<hr \/>\n<h4>14.2 \u6a21\u578b\u90e8\u7f72&#xff1a;\u4e3a\u6a21\u578b\u5bfb\u627e\u4e00\u4e2a\u201c\u5bb6\u201d<\/h4>\n<p>\u6a21\u578b\u90e8\u7f72&#xff0c;\u5c31\u662f\u5c06\u6211\u4eec\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u6587\u4ef6&#xff0c;\u653e\u7f6e\u5230\u4e00\u4e2a\u4e13\u95e8\u7684\u3001\u4e3a\u751f\u4ea7\u73af\u5883\u8bbe\u8ba1\u7684\u8f6f\u4ef6\u7cfb\u7edf\u4e2d\u8fd0\u884c\u7684\u8fc7\u7a0b\u3002\u8fd9\u4e2a\u7cfb\u7edf&#xff0c;\u5c31\u662f\u6a21\u578b\u7684\u201c\u5bb6\u201d\u3002\u5b83\u9700\u8981\u8db3\u591f\u575a\u56fa&#xff0c;\u80fd\u591f\u62b5\u5fa1\u9ad8\u5e76\u53d1\u7684\u8bbf\u95ee\u538b\u529b&#xff1b;\u9700\u8981\u8db3\u591f\u9ad8\u6548&#xff0c;\u80fd\u591f\u63d0\u4f9b\u4f4e\u5ef6\u8fdf\u7684\u63a8\u7406\u54cd\u5e94&#xff1b;\u8fd8\u9700\u8981\u8db3\u591f\u7075\u6d3b&#xff0c;\u80fd\u591f\u65b9\u4fbf\u5730\u7ba1\u7406\u548c\u66f4\u65b0\u6a21\u578b\u3002<\/p>\n<h5>14.2.1 \u8de8\u8d8a\u6846\u67b6\u7684\u9e3f\u6c9f&#xff1a;ONNX&#xff08;Open Neural Network Exchange&#xff09;<\/h5>\n<p>\u5728\u5c06\u6a21\u578b\u201c\u653e\u5165\u201d\u65b0\u5bb6\u4e4b\u524d&#xff0c;\u6211\u4eec\u9996\u5148\u8981\u89e3\u51b3\u4e00\u4e2a\u201c\u8bed\u8a00\u201d\u95ee\u9898\u3002TensorFlow\u8bad\u7ec3\u7684\u6a21\u578b&#xff0c;PyTorch\u4e0d\u8ba4\u8bc6&#xff1b;PyTorch\u4fdd\u5b58\u7684\u6a21\u578b&#xff0c;MXNet\u4e5f\u65e0\u6cd5\u52a0\u8f7d\u3002\u8fd9\u79cd\u6846\u67b6\u95f4\u7684\u58c1\u5792&#xff0c;\u7ed9\u6a21\u578b\u7684\u81ea\u7531\u6d41\u901a\u548c\u90e8\u7f72\u5e26\u6765\u4e86\u5de8\u5927\u7684\u969c\u788d\u3002<\/p>\n<ul>\n<li>\n<p>\u95ee\u9898&#xff1a;\u6846\u67b6\u6797\u7acb&#xff0c;\u5404\u81ea\u4e3a\u653f \u4e0d\u540c\u7684\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6&#xff0c;\u5c31\u50cf\u8bf4\u7740\u4e0d\u540c\u201c\u65b9\u8a00\u201d\u7684\u90e8\u843d&#xff0c;\u5b83\u4eec\u90fd\u6709\u81ea\u5df1\u72ec\u7279\u7684\u6a21\u578b\u4fdd\u5b58\u683c\u5f0f\u3002\u8fd9\u79cd\u4e0d\u7edf\u4e00&#xff0c;\u5bfc\u81f4\u4e00\u4e2a\u6a21\u578b\u88ab\u201c\u9501\u5b9a\u201d\u5728\u4e86\u521b\u9020\u5b83\u7684\u90a3\u4e2a\u6846\u67b6\u751f\u6001\u91cc&#xff0c;\u96be\u4ee5\u8fc1\u79fb\u3002<\/p>\n<\/li>\n<li>\n<p>ONNX\u7684\u89d2\u8272&#xff1a;AI\u4e16\u754c\u7684\u201c\u666e\u901a\u8bdd\u201d \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898&#xff0c;ONNX (Open Neural Network Exchange) \u5e94\u8fd0\u800c\u751f\u3002\u5b83\u5e76\u975e\u4e00\u4e2a\u6846\u67b6&#xff0c;\u800c\u662f\u4e00\u4e2a\u5f00\u653e\u7684\u3001\u4e2d\u7acb\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u8868\u793a\u683c\u5f0f\u3002\u5b83\u5b9a\u4e49\u4e86\u4e00\u5957\u6807\u51c6\u7684\u7b97\u5b50\u548c\u6587\u4ef6\u683c\u5f0f&#xff0c;\u81f4\u529b\u4e8e\u6210\u4e3aAI\u6a21\u578b\u9886\u57df\u7684\u201c\u666e\u901a\u8bdd\u201d\u3002<\/p>\n<ul>\n<li>\u5de5\u4f5c\u6d41\u7a0b&#xff1a;\u6211\u4eec\u53ef\u4ee5\u5c06\u4efb\u4f55\u4e3b\u6d41\u6846\u67b6&#xff08;\u5982TensorFlow, PyTorch, Keras, MXNet&#xff09;\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5bfc\u51fa&#xff08;Export&#xff09;\u4e3a\u7edf\u4e00\u7684.onnx\u683c\u5f0f\u3002\u7136\u540e&#xff0c;\u4efb\u4f55\u652f\u6301ONNX\u6807\u51c6\u7684\u63a8\u7406\u5f15\u64ce&#xff08;Inference Engine&#xff09;&#xff0c;\u5982\u5fae\u8f6f\u7684ONNX Runtime\u3001NVIDIA\u7684TensorRT\u7b49&#xff0c;\u90fd\u53ef\u4ee5\u52a0\u8f7d\u8fd9\u4e2a.onnx\u6587\u4ef6&#xff0c;\u5e76\u5728\u5404\u79cd\u4e0d\u540c\u7684\u786c\u4ef6\u5e73\u53f0&#xff08;CPU, GPU, \u79fb\u52a8\u7aef\u82af\u7247&#xff09;\u4e0a\u8fdb\u884c\u9ad8\u6027\u80fd\u7684\u63a8\u7406\u3002<\/li>\n<li>\u6838\u5fc3\u4ef7\u503c&#xff1a;ONNX\u5c06\u201c\u6a21\u578b\u8bad\u7ec3\u201d\u548c\u201c\u6a21\u578b\u63a8\u7406\u201d\u8fd9\u4e24\u4e2a\u73af\u8282\u8fdb\u884c\u4e86\u89e3\u8026\u3002\u7814\u7a76\u5458\u53ef\u4ee5\u7528\u81ea\u5df1\u6700\u559c\u6b22\u7684\u6846\u67b6\u8fdb\u884c\u521b\u65b0\u548c\u5b9e\u9a8c&#xff0c;\u800c\u5de5\u7a0b\u5e08\u5219\u53ef\u4ee5\u4e13\u6ce8\u4e8e\u7528\u7edf\u4e00\u7684\u5de5\u5177\u6808\u8fdb\u884c\u9ad8\u6548\u3001\u8de8\u5e73\u53f0\u7684\u90e8\u7f72\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5b9e\u8df5&#xff1a;\u5bfc\u51fa\u4e3aONNX \u5c06\u6a21\u578b\u5bfc\u51fa\u4e3aONNX\u901a\u5e38\u975e\u5e38\u7b80\u5355\u3002\u4f8b\u5982&#xff0c;\u5bf9\u4e8e\u4e00\u4e2aPyTorch\u6a21\u578b&#xff1a;<\/p>\n<p> import torch<br \/>\nimport torchvision<\/p>\n<p># \u5047\u8bbedummy_model\u662f\u4e00\u4e2a\u8bad\u7ec3\u597d\u7684PyTorch\u6a21\u578b<br \/>\ndummy_input &#061; torch.randn(1, 3, 224, 224) # \u521b\u5efa\u4e00\u4e2a\u7b26\u5408\u6a21\u578b\u8f93\u5165\u7684\u865a\u62df\u5f20\u91cf<br \/>\ntorch.onnx.export(dummy_model,<br \/>\n                  dummy_input,<br \/>\n                  &#034;model.onnx&#034;, # \u8f93\u51fa\u6587\u4ef6\u540d<br \/>\n                  verbose&#061;False)<\/p>\n<p>\u62e5\u6709\u4e86.onnx\u8fd9\u4e2a\u901a\u7528\u683c\u5f0f\u6587\u4ef6&#xff0c;\u6211\u4eec\u7684\u6a21\u578b\u5c31\u83b7\u5f97\u4e86\u524d\u6240\u672a\u6709\u7684\u201c\u81ea\u7531\u201d&#xff0c;\u53ef\u4ee5\u88ab\u90e8\u7f72\u5230\u66f4\u5e7f\u9614\u7684\u4e16\u754c\u4e2d\u53bb\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>14.2.2 \u5de5\u4e1a\u7ea7\u670d\u52a1\u6846\u67b6&#xff1a;TensorFlow Serving<\/h5>\n<p>\u73b0\u5728&#xff0c;\u6a21\u578b\u6709\u4e86\u201c\u901a\u7528\u8bed\u8a00\u201d&#xff0c;\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u5de5\u4e1a\u7ea7\u7684\u201c\u5bb6\u201d\u6765\u627f\u8f7d\u5b83\u3002TensorFlow Serving\u6b63\u662fGoogle\u4e3a\u5176\u4eb2\u513f\u5b50TensorFlow\u91cf\u8eab\u6253\u9020&#xff0c;\u5e76\u4e45\u7ecf\u751f\u4ea7\u73af\u5883\u8003\u9a8c\u7684\u9ad8\u6027\u80fd\u670d\u52a1\u7cfb\u7edf\u3002<\/p>\n<ul>\n<li>\n<p>\u5b9a\u4f4d&#xff1a;\u4e00\u4e2a\u4e13\u4e3a\u751f\u4ea7\u73af\u5883\u8bbe\u8ba1\u7684\u3001\u7075\u6d3b\u7684\u3001\u9ad8\u6027\u80fd\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u670d\u52a1\u7cfb\u7edf\u3002<\/p>\n<\/li>\n<li>\n<p>\u6838\u5fc3\u7279\u6027&#xff1a;<\/p>\n<ul>\n<li>\u9ad8\u6027\u80fd&#xff1a;\u5e95\u5c42\u7531C&#043;&#043;\u7f16\u5199&#xff0c;\u4e13\u4e3a\u4f4e\u5ef6\u8fdf\u3001\u9ad8\u541e\u5410\u91cf\u7684\u63a8\u7406\u573a\u666f\u4f18\u5316\u3002<\/li>\n<li>\u6a21\u578b\u7248\u672c\u63a7\u5236&#xff1a;\u8fd9\u662f\u5176\u6700\u5f3a\u5927\u7684\u529f\u80fd\u4e4b\u4e00\u3002\u53ef\u4ee5\u4e3a\u4e00\u4e2a\u6a21\u578b\u90e8\u7f72\u591a\u4e2a\u7248\u672c&#xff0c;\u5e76\u53ef\u4ee5\u65e0\u7f1d\u5730\u5207\u6362\u3001\u56de\u6eda&#xff0c;\u751a\u81f3\u8fdb\u884cA\/B\u6d4b\u8bd5&#xff0c;\u800c\u65e0\u9700\u4e2d\u65ad\u670d\u52a1\u3002<\/li>\n<li>\u6613\u4e8e\u90e8\u7f72&#xff1a;\u4e0eDocker\u7b49\u5bb9\u5668\u5316\u6280\u672f\u5b8c\u7f8e\u7ed3\u5408&#xff0c;\u53ef\u4ee5\u901a\u8fc7\u51e0\u884c\u547d\u4ee4\u5c31\u542f\u52a8\u4e00\u4e2a\u7a33\u5b9a\u3001\u53ef\u6269\u5c55\u7684\u9884\u6d4b\u670d\u52a1\u3002<\/li>\n<li>\u591a\u6a21\u578b\u670d\u52a1&#xff1a;\u53ef\u4ee5\u5728\u540c\u4e00\u4e2a\u670d\u52a1\u5b9e\u4f8b\u4e2d&#xff0c;\u540c\u65f6\u52a0\u8f7d\u548c\u63d0\u4f9b\u591a\u4e2a\u4e0d\u540c\u6a21\u578b\u7684\u9884\u6d4b\u670d\u52a1\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u5de5\u4f5c\u6d41\u7a0b&#xff1a;<\/p>\n<li>\u5bfc\u51fa\u6a21\u578b&#xff1a;\u5c06TensorFlow\/Keras\u6a21\u578b\u5bfc\u51fa\u4e3a\u6807\u51c6\u7684SavedModel\u683c\u5f0f\u3002\u8fd9\u662f\u4e00\u4e2a\u5305\u542b\u4e86\u6a21\u578b\u56fe\u7ed3\u6784\u548c\u6743\u91cd\u7684\u76ee\u5f55\u3002<\/li>\n<li>\u542f\u52a8\u670d\u52a1&#xff1a;\u4f7f\u7528\u5b98\u65b9\u63d0\u4f9b\u7684Docker\u955c\u50cf&#xff0c;\u662f\u6700\u7b80\u5355\u7684\u65b9\u5f0f\u3002\u53ea\u9700\u4e00\u6761\u547d\u4ee4&#xff0c;\u6307\u5b9a\u6a21\u578b\u6240\u5728\u7684\u8def\u5f84\u548c\u7aef\u53e3&#xff0c;\u5373\u53ef\u542f\u52a8\u670d\u52a1\u3002\n<p>bash<\/p>\n<p> docker run -p 8501:8501 &#8211;mount type&#061;bind,source&#061;\/path\/to\/my_model\/,target&#061;\/models\/my_model -e MODEL_NAME&#061;my_model -t tensorflow\/serving<br \/>\n&#096;&#096;&#096;    3.  **\u53d1\u8d77\u8bf7\u6c42**&#xff1a;\u670d\u52a1\u542f\u52a8\u540e&#xff0c;\u4f1a\u66b4\u9732\u4e00\u4e2aRESTful API\u6216gRPC\u7aef\u70b9\u3002\u5ba2\u6237\u7aef\u53ef\u4ee5\u901a\u8fc7\u53d1\u9001HTTP\/gRPC\u8bf7\u6c42&#xff0c;\u5c06\u6570\u636e\u53d1\u9001\u5230\u8fd9\u4e2a\u7aef\u70b9&#xff0c;\u5e76\u83b7\u5f97\u9884\u6d4b\u7ed3\u679c\u3002\n <\/li>\n<\/li>\n<\/ul>\n<h5>14.2.3 \u53e6\u4e00\u4e2a\u9009\u62e9&#xff1a;TorchServe<\/h5>\n<p>\u5bf9\u4e8ePyTorch\u7684\u5fe0\u5b9e\u7528\u6237&#xff0c;\u7531PyTorch\u56e2\u961f\u5b98\u65b9\u7ef4\u62a4\u7684TorchServe\u5219\u662f\u4e00\u4e2a\u66f4\u81ea\u7136\u7684\u9009\u62e9\u3002<\/p>\n<ul>\n<li>\u5b9a\u4f4d&#xff1a;\u4e00\u4e2a\u4e13\u95e8\u4e3aPyTorch\u6a21\u578b\u8bbe\u8ba1\u7684\u3001\u7075\u6d3b\u4e14\u6613\u4e8e\u4f7f\u7528\u7684\u670d\u52a1\u5de5\u5177\u3002<\/li>\n<li>\u6838\u5fc3\u7279\u6027&#xff1a;\n<ul>\n<li>\u4e0ePyTorch\u751f\u6001\u7d27\u5bc6\u96c6\u6210&#xff1a;\u5bf9PyTorch\u6a21\u578b\u7684\u652f\u6301\u6700\u4e3a\u539f\u751f\u3002<\/li>\n<li>\u6a21\u578b\u6253\u5305&#xff1a;\u5b83\u8981\u6c42\u5c06\u6a21\u578b\u6587\u4ef6\u3001\u4f9d\u8d56\u9879\u4ee5\u53ca\u81ea\u5b9a\u4e49\u7684\u9884\u5904\u7406\/\u540e\u5904\u7406\u903b\u8f91&#xff0c;\u4e00\u8d77\u6253\u5305\u6210\u4e00\u4e2a.mar&#xff08;Model Archive&#xff09;\u6587\u4ef6&#xff0c;\u8fd9\u79cd\u5c01\u88c5\u4f7f\u5f97\u90e8\u7f72\u66f4\u52a0\u89c4\u8303\u548c\u53ef\u79fb\u690d\u3002<\/li>\n<li>\u529f\u80fd\u5b8c\u5907&#xff1a;\u540c\u6837\u652f\u6301\u6a21\u578b\u7248\u672c\u7ba1\u7406\u3001\u6027\u80fd\u6307\u6807\u76d1\u63a7\u3001REST\/gRPC\u63a5\u53e3\u7b49\u751f\u4ea7\u7ea7\u7279\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u9009\u62e9TF Serving\u8fd8\u662fTorchServe&#xff0c;\u4e3b\u8981\u53d6\u51b3\u4e8e\u60a8\u7684\u6280\u672f\u6808\u548c\u56e2\u961f\u504f\u597d\u3002\u4e24\u8005\u90fd\u662f\u4f18\u79c0\u7684\u3001\u7ecf\u5f97\u8d77\u8003\u9a8c\u7684\u751f\u4ea7\u7ea7\u90e8\u7f72\u65b9\u6848\u3002<\/p>\n<hr \/>\n<h4>14.3 \u5c06\u6a21\u578b\u5c01\u88c5\u4e3aAPI\u670d\u52a1&#xff1a;\u6784\u5efa\u4e0e\u4e16\u754c\u6c9f\u901a\u7684\u6865\u6881<\/h4>\n<p>\u65e0\u8bba\u662fTF Serving\u8fd8\u662fTorchServe&#xff0c;\u5b83\u4eec\u90fd\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5f3a\u5927\u7684\u540e\u7aef\u670d\u52a1\u3002\u73b0\u5728&#xff0c;\u6211\u4eec\u9700\u8981\u6784\u5efa\u4e00\u4e2a\u6807\u51c6\u7684\u201c\u63a5\u53e3\u201d&#xff0c;\u8ba9\u5916\u90e8\u4e16\u754c\u80fd\u591f\u65b9\u4fbf\u5730\u4e0e\u8fd9\u4e2a\u670d\u52a1\u8fdb\u884c\u4ea4\u4e92\u3002\u8fd9\u4e2a\u63a5\u53e3&#xff0c;\u5c31\u662fAPI&#xff08;Application Programming Interface&#xff09;\u3002<\/p>\n<h5>14.3.1 \u4e3a\u4f55\u9700\u8981API&#xff1f;\u89e3\u8026\u4e0e\u6807\u51c6\u5316<\/h5>\n<p>\u5c06\u6a21\u578b\u80fd\u529b\u5c01\u88c5\u6210API&#xff0c;\u662f\u73b0\u4ee3\u8f6f\u4ef6\u5de5\u7a0b\u7684\u6700\u4f73\u5b9e\u8df5\u3002<\/p>\n<ul>\n<li>\u89e3\u8026&#xff1a;\u524d\u7aef\u5e94\u7528&#xff08;\u5982\u7f51\u9875\u3001\u624b\u673aApp&#xff09;\u7684\u5f00\u53d1\u8005&#xff0c;\u65e0\u9700\u5173\u5fc3\u60a8\u7684\u6a21\u578b\u662f\u7528\u4ec0\u4e48\u6846\u67b6\u5199\u7684\u3001\u90e8\u7f72\u5728\u54ea\u91cc\u3002\u4ed6\u4eec\u53ea\u9700\u8981\u77e5\u9053API\u7684\u5730\u5740\u548c\u8c03\u7528\u89c4\u5219&#xff0c;\u5c31\u53ef\u4ee5\u50cf\u8c03\u7528\u4e00\u4e2a\u666e\u901a\u7684\u51fd\u6570\u4e00\u6837&#xff0c;\u83b7\u5f97AI\u7684\u80fd\u529b\u3002\u8fd9\u4f7f\u5f97\u524d\u7aef\u548cAI\u540e\u7aef\u7684\u5f00\u53d1\u53ef\u4ee5\u5b8c\u5168\u5206\u79bb&#xff0c;\u72ec\u7acb\u8fdb\u884c\u3002<\/li>\n<li>\u6807\u51c6\u5316&#xff1a;HTTP\u534f\u8bae\u662f\u4e92\u8054\u7f51\u7684\u901a\u7528\u8bed\u8a00\u3002\u901a\u8fc7RESTful API&#xff0c;\u6211\u4eec\u53ef\u4ee5\u7528\u6700\u6807\u51c6\u3001\u6700\u901a\u7528\u7684\u65b9\u5f0f\u63d0\u4f9b\u670d\u52a1&#xff0c;\u4efb\u4f55\u7f16\u7a0b\u8bed\u8a00\u3001\u4efb\u4f55\u5e73\u53f0\u7684\u5ba2\u6237\u7aef\u90fd\u53ef\u4ee5\u8f7b\u677e\u8c03\u7528\u3002<\/li>\n<\/ul>\n<p>\u6211\u4eec\u5c06\u4f7f\u7528Python\u7684Web\u6846\u67b6&#xff0c;\u6765\u5feb\u901f\u6784\u5efa\u8fd9\u4e2aAPI\u201c\u4e2d\u95f4\u5c42\u201d\u3002<\/p>\n<h5>14.3.2 Flask&#xff1a;\u8f7b\u91cf\u800c\u5f3a\u5927\u7684Python Web\u6846\u67b6<\/h5>\n<p>Flask\u4ee5\u5176\u7b80\u6d01\u3001\u7075\u6d3b\u548c\u201c\u5fae\u201d\u6838\u5fc3\u7684\u8bbe\u8ba1\u54f2\u5b66&#xff0c;\u6210\u4e3a\u8bb8\u591aPython\u5f00\u53d1\u8005\u6784\u5efaWeb\u670d\u52a1\u7684\u9996\u9009&#xff0c;\u5c24\u5176\u9002\u5408\u5feb\u901f\u642d\u5efa\u4e2d\u5c0f\u578bAPI\u670d\u52a1\u3002<\/p>\n<ul>\n<li>\u5b9e\u8df5&#xff1a;\u7528Flask\u521b\u5efa\u4e00\u4e2a\u7b80\u5355\u7684\u56fe\u50cf\u5206\u7c7bAPI\u3002\n<p>python<\/p>\n<p> from flask import Flask, request, jsonify<br \/>\n# \u5047\u8bbe\u6211\u4eec\u6709\u4e00\u4e2a\u5df2\u7ecf\u52a0\u8f7d\u597d\u7684\u6a21\u578b\u548c\u9884\u5904\u7406\u51fd\u6570<br \/>\n# from my_model import load_my_model, preprocess_image, predict<\/p>\n<p>app &#061; Flask(__name__)<br \/>\nmodel &#061; load_my_model()<\/p>\n<p>&#064;app.route(&#034;\/predict&#034;, methods&#061;[&#034;POST&#034;])<br \/>\ndef predict_endpoint():<br \/>\n    if &#039;file&#039; not in request.files:<br \/>\n        return &#034;File not provided&#034;, 400<\/p>\n<p>    file &#061; request.files[&#039;file&#039;]<br \/>\n    image &#061; preprocess_image(file) # \u8bfb\u53d6\u5e76\u9884\u5904\u7406\u56fe\u50cf<br \/>\n    prediction &#061; predict(model, image) # \u6a21\u578b\u9884\u6d4b<\/p>\n<p>    return jsonify(prediction) # \u5c06\u7ed3\u679c\u4ee5JSON\u683c\u5f0f\u8fd4\u56de<\/p>\n<p>if __name__ &#061;&#061; &#034;__main__&#034;:<br \/>\n    app.run(host&#061;&#034;0.0.0.0&#034;, port&#061;5000)<br \/>\n \u53ea\u9700\u77ed\u77ed\u5341\u51e0\u884c\u4ee3\u7801&#xff0c;\u6211\u4eec\u5c31\u521b\u5efa\u4e86\u4e00\u4e2a\u529f\u80fd\u5b8c\u5907\u7684\u9884\u6d4bAPI\u3002\u4efb\u4f55\u5ba2\u6237\u7aef\u53ea\u9700\u5411\/predict\u8fd9\u4e2a\u5730\u5740\u53d1\u9001\u4e00\u4e2a\u5305\u542b\u56fe\u7247\u6587\u4ef6\u7684POST\u8bf7\u6c42&#xff0c;\u5c31\u80fd\u6536\u5230JSON\u683c\u5f0f\u7684\u9884\u6d4b\u7ed3\u679c\u3002<\/li>\n<\/ul>\n<h5>14.3.3 FastAPI&#xff1a;\u73b0\u4ee3\u3001\u9ad8\u6027\u80fd\u7684\u9009\u62e9<\/h5>\n<p>\u8fd1\u5e74\u6765&#xff0c;\u4e00\u4e2a\u540d\u4e3aFastAPI\u7684\u65b0\u661f\u5189\u5189\u5347\u8d77&#xff0c;\u5e76\u8fc5\u901f\u6210\u4e3a\u6784\u5efa\u9ad8\u6027\u80fdAPI\u7684\u4e1a\u754c\u65b0\u5ba0\u3002<\/p>\n<ul>\n<li>\u7279\u70b9&#xff1a;\n<ul>\n<li>\u9ad8\u6027\u80fd&#xff1a;\u5176\u5e95\u5c42\u57fa\u4e8eStarlette\u548cPydantic&#xff0c;\u6027\u80fd\u53ef\u4e0eNode.js\u3001Go\u7b49\u7f16\u8bd1\u578b\u8bed\u8a00\u6784\u5efa\u7684\u670d\u52a1\u76f8\u5ab2\u7f8e\u3002<\/li>\n<li>\u57fa\u4e8e\u7c7b\u578b\u63d0\u793a&#xff1a;\u5b83\u5927\u91cf\u4f7f\u7528Python 3.6&#043;\u7684\u7c7b\u578b\u63d0\u793a\u3002\u53ea\u9700\u8981\u5728\u51fd\u6570\u7b7e\u540d\u4e2d\u58f0\u660e\u6570\u636e\u7c7b\u578b&#xff0c;FastAPI\u5c31\u80fd\u81ea\u52a8\u5b8c\u6210\u6570\u636e\u6821\u9a8c\u3001\u5e8f\u5217\u5316\u548c\u6587\u6863\u751f\u6210\u3002<\/li>\n<li>\u81ea\u52a8\u751f\u6210API\u6587\u6863&#xff1a;\u8fd9\u662fFastAPI\u6700\u4ee4\u4eba\u60ca\u8273\u7684\u7279\u6027\u4e4b\u4e00\u3002\u5b83\u80fd\u6839\u636e\u4ee3\u7801&#xff0c;\u81ea\u52a8\u751f\u6210\u4e00\u4e2a\u4ea4\u4e92\u5f0f\u7684API\u6587\u6863\u9875\u9762&#xff08;\u57fa\u4e8eSwagger UI\u548cReDoc&#xff09;\u3002\u5f00\u53d1\u8005\u53ef\u4ee5\u76f4\u63a5\u5728\u6d4f\u89c8\u5668\u4e2d\u6d4b\u8bd5API&#xff0c;\u6781\u5927\u63d0\u5347\u4e86\u534f\u4f5c\u6548\u7387\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u5bf9\u4e8e\u8ffd\u6c42\u6781\u81f4\u6027\u80fd\u3001\u89c4\u8303\u5316\u5f00\u53d1\u548c\u81ea\u52a8\u5316\u6587\u6863\u7684\u751f\u4ea7\u7ea7\u9879\u76ee&#xff0c;FastAPI\u65e0\u7591\u662f\u6bd4Flask\u66f4\u4f18\u8d8a\u7684\u9009\u62e9\u3002<\/p>\n<hr \/>\n<h4>14.4 MLOps\u7b80\u4ecb&#xff1a;\u7cfb\u7edf\u5316\u5730\u7ba1\u7406AI\u7684\u751f\u547d\u5468\u671f<\/h4>\n<p>\u81f3\u6b64&#xff0c;\u6211\u4eec\u5df2\u7ecf\u6210\u529f\u5730\u5c06\u5355\u4e2a\u6a21\u578b\u90e8\u7f72\u4e0a\u7ebf\u3002\u4f46\u4e00\u4e2a\u771f\u6b63\u7684AI\u4ea7\u54c1&#xff0c;\u5176\u751f\u547d\u8fdc\u672a\u7ed3\u675f\u3002\u6570\u636e\u5728\u53d8&#xff0c;\u4e1a\u52a1\u5728\u53d8&#xff0c;\u6a21\u578b\u4e5f\u9700\u8981\u4e0d\u65ad\u5730\u8fed\u4ee3\u548c\u6f14\u8fdb\u3002\u5982\u4f55\u79d1\u5b66\u3001\u9ad8\u6548\u5730\u7ba1\u7406\u8fd9\u6574\u4e2a\u590d\u6742\u7684\u751f\u547d\u5468\u671f&#xff1f;MLOps\u7ed9\u51fa\u4e86\u7b54\u6848\u3002<\/p>\n<h5>14.4.1 \u4eceDevOps\u5230MLOps&#xff1a;\u5f53\u673a\u5668\u5b66\u4e60\u9047\u89c1\u5de5\u7a0b\u6587\u5316<\/h5>\n<ul>\n<li>MLOps&#xff08;Machine Learning Operations&#xff09;\u662f\u4e00\u5957\u65e8\u5728\u5b9e\u73b0\u673a\u5668\u5b66\u4e60\u5e94\u7528\u5f00\u53d1\u3001\u90e8\u7f72\u548c\u7ef4\u62a4\u6d41\u7a0b\u81ea\u52a8\u5316\u4e0e\u6807\u51c6\u5316\u7684\u539f\u5219\u548c\u5b9e\u8df5\u3002\u5b83\u662f\u6210\u719f\u7684\u8f6f\u4ef6\u5de5\u7a0b\u6587\u5316DevOps\u5728\u673a\u5668\u5b66\u4e60\u9886\u57df\u7684\u81ea\u7136\u5ef6\u4f38\u3002<\/li>\n<li>\u4e3a\u4f55\u9700\u8981MLOps&#xff1a;\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u662f\u4e00\u4e2a\u72ec\u7279\u7684\u201c\u4e09\u4f53\u201d\u3002\u5b83\u4e0d\u4ec5\u4ec5\u7531\u4ee3\u7801\u6784\u6210&#xff0c;\u8fd8\u5305\u542b\u6570\u636e\u548c\u6a21\u578b\u8fd9\u4e24\u4e2a\u6838\u5fc3\u7ec4\u4ef6\u3002\u8fd9\u4e09\u8005\u90fd\u6709\u5404\u81ea\u7684\u751f\u547d\u5468\u671f&#xff0c;\u90fd\u9700\u8981\u88ab\u4e25\u683c\u7ba1\u7406\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u6a21\u578b\u7684\u6027\u80fd\u4f1a\u968f\u7740\u65f6\u95f4\u7684\u63a8\u79fb\u800c\u201c\u8870\u9000\u201d&#xff08;Concept Drift&#xff09;&#xff0c;\u9700\u8981\u5efa\u7acb\u4e00\u5957\u6301\u7eed\u76d1\u63a7\u548c\u518d\u8bad\u7ec3\u7684\u673a\u5236\u3002MLOps\u6b63\u662f\u4e3a\u4e86\u5e94\u5bf9\u8fd9\u79cd\u590d\u6742\u6027\u800c\u751f\u3002<\/li>\n<\/ul>\n<h5>14.4.2 \u6838\u5fc3\u652f\u67f1\u4e4b\u4e00&#xff1a;\u7248\u672c\u63a7\u5236<\/h5>\n<p>\u53ef\u590d\u73b0\u6027\u662f\u79d1\u5b66\u7684\u57fa\u77f3&#xff0c;\u4e5f\u662fMLOps\u7684\u6838\u5fc3\u8ffd\u6c42\u3002\u6211\u4eec\u9700\u8981\u5bf9AI\u7cfb\u7edf\u7684\u4e09\u5927\u7ec4\u4ef6\u8fdb\u884c\u5168\u9762\u7684\u7248\u672c\u63a7\u5236\u3002<\/p>\n<ul>\n<li>\u4ee3\u7801\u7248\u672c\u63a7\u5236&#xff1a;\u4f7f\u7528Git\u3002\u8fd9\u662f\u6240\u6709\u73b0\u4ee3\u8f6f\u4ef6\u5de5\u7a0b\u7684\u7edd\u5bf9\u57fa\u7840&#xff0c;\u65e0\u9700\u591a\u8a00\u3002<\/li>\n<li>\u6570\u636e\u7248\u672c\u63a7\u5236&#xff1a;\u6570\u636e\u662f\u6a21\u578b\u7684\u201c\u98df\u7cae\u201d&#xff0c;\u6570\u636e\u7684\u4efb\u4f55\u53d8\u5316\u90fd\u53ef\u80fd\u5bfc\u81f4\u6a21\u578b\u884c\u4e3a\u7684\u6539\u53d8\u3002\u6211\u4eec\u9700\u8981\u5de5\u5177\u6765\u8ffd\u8e2a\u6570\u636e\u7684\u7248\u672c\u3002DVC (Data Version Control)\u00a0\u5c31\u662f\u4e3a\u6b64\u800c\u751f\u7684\u5de5\u5177\u3002\u5b83\u80fd\u50cfGit\u4e00\u6837&#xff0c;\u5bf9\u5927\u578b\u6570\u636e\u96c6\u8fdb\u884c\u7248\u672c\u7ba1\u7406&#xff0c;\u800c\u65e0\u9700\u5c06\u5e9e\u5927\u7684\u6570\u636e\u6587\u4ef6\u5b58\u5165Git\u4ed3\u5e93\u3002<\/li>\n<li>\u6a21\u578b\u7248\u672c\u63a7\u5236&#xff1a;\u6bcf\u6b21\u8bad\u7ec3&#xff0c;\u6211\u4eec\u90fd\u53ef\u80fd\u5f97\u5230\u4e00\u4e2a\u65b0\u7248\u672c\u7684\u6a21\u578b\u3002\u6211\u4eec\u9700\u8981\u4e00\u4e2a\u7cfb\u7edf\u6765\u8bb0\u5f55\u6bcf\u4e2a\u6a21\u578b\u7248\u672c\u3001\u5b83\u662f\u7531\u54ea\u4e2a\u7248\u672c\u7684\u6570\u636e\u548c\u4ee3\u7801\u8bad\u7ec3\u51fa\u6765\u7684\u3001\u5b83\u7684\u6027\u80fd\u6307\u6807\u5982\u4f55\u3001\u5b83\u5f53\u524d\u5904\u4e8e\u4ec0\u4e48\u72b6\u6001&#xff08;\u5f00\u53d1\u3001\u6682\u5b58\u6216\u751f\u4ea7&#xff09;\u3002MLflow\u7b49\u5de5\u5177\u63d0\u4f9b\u7684**\u6a21\u578b\u6ce8\u518c&#xff08;Model Registry&#xff09;**\u529f\u80fd&#xff0c;\u6b63\u662f\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u7684\u5229\u5668\u3002<\/li>\n<\/ul>\n<h5>14.4.3 \u6838\u5fc3\u652f\u67f1\u4e4b\u4e8c&#xff1a;CI\/CD\/CT\u6d41\u6c34\u7ebf<\/h5>\n<p>MLOps\u81f4\u529b\u4e8e\u5c06AI\u751f\u547d\u5468\u671f\u4e2d\u7684\u4e00\u5207\u90fd\u81ea\u52a8\u5316&#xff0c;\u6784\u5efa\u81ea\u52a8\u5316\u7684\u201c\u6d41\u6c34\u7ebf&#xff08;Pipeline&#xff09;\u201d\u662f\u5176\u6838\u5fc3\u5b9e\u8df5\u3002<\/p>\n<ul>\n<li>CI (Continuous Integration, \u6301\u7eed\u96c6\u6210)&#xff1a;\u5f53\u5f00\u53d1\u8005\u63d0\u4ea4\u65b0\u4ee3\u7801\u65f6&#xff0c;\u81ea\u52a8\u89e6\u53d1\u4ee3\u7801\u7684\u6784\u5efa\u548c\u5355\u5143\u6d4b\u8bd5\u3002<\/li>\n<li>CD (Continuous Delivery\/Deployment, \u6301\u7eed\u4ea4\u4ed8\/\u90e8\u7f72)&#xff1a;\u5f53CI\u901a\u8fc7\u540e&#xff0c;\u81ea\u52a8\u5c06\u6a21\u578b\u6253\u5305\u3001\u8fdb\u884c\u96c6\u6210\u6d4b\u8bd5&#xff0c;\u5e76\u5c06\u5176\u90e8\u7f72\u5230\u9884\u751f\u4ea7\u6216\u751f\u4ea7\u73af\u5883\u3002<\/li>\n<li>CT (Continuous Training, \u6301\u7eed\u8bad\u7ec3)&#xff1a;\u8fd9\u662fMLOps\u7279\u6709\u7684\u6982\u5ff5\u3002\u5f53\u68c0\u6d4b\u5230\u6709\u5927\u91cf\u65b0\u6570\u636e\u6807\u6ce8\u5b8c\u6210&#xff0c;\u6216\u8005\u7ebf\u4e0a\u6a21\u578b\u7684\u6027\u80fd\u51fa\u73b0\u663e\u8457\u4e0b\u964d\u65f6&#xff0c;\u81ea\u52a8\u89e6\u53d1\u4e00\u4e2a\u5b8c\u6574\u7684\u6a21\u578b\u518d\u8bad\u7ec3\u3001\u8bc4\u4f30\u548c\u9a8c\u8bc1\u7684\u6d41\u6c34\u7ebf\u3002\u5982\u679c\u65b0\u6a21\u578b\u7684\u6027\u80fd\u4f18\u4e8e\u65e7\u6a21\u578b&#xff0c;\u5219\u81ea\u52a8\u8fdb\u5165CD\u6d41\u7a0b\u3002<\/li>\n<\/ul>\n<h5>14.4.4 MLOps\u7684\u613f\u666f&#xff1a;\u6784\u5efa\u53ef\u590d\u73b0\u3001\u53ef\u4fe1\u8d56\u3001\u53ef\u6f14\u8fdb\u7684AI\u7cfb\u7edf<\/h5>\n<p>MLOps\u7684\u7ec8\u6781\u76ee\u6807&#xff0c;\u662f\u6784\u5efa\u4e00\u4e2a\u81ea\u52a8\u5316\u7684\u3001\u95ed\u73af\u7684AI\u7cfb\u7edf\u3002\u5728\u8fd9\u4e2a\u7cfb\u7edf\u4e2d&#xff0c;\u4ece\u6570\u636e\u51c6\u5907\u3001\u6a21\u578b\u8bad\u7ec3\u3001\u90e8\u7f72\u4e0a\u7ebf&#xff0c;\u5230\u6027\u80fd\u76d1\u63a7\u3001\u81ea\u52a8\u518d\u8bad\u7ec3&#xff0c;\u5f62\u6210\u4e00\u4e2a\u751f\u751f\u4e0d\u606f\u7684\u5faa\u73af\u3002\u5b83\u65e8\u5728\u5c06AI\u5e94\u7528\u7684\u5f00\u53d1&#xff0c;\u4ece\u4e00\u79cd\u9ad8\u5ea6\u4f9d\u8d56\u4eba\u5de5\u548c\u7ecf\u9a8c\u7684\u201c\u624b\u5de5\u4f5c\u574a\u201d\u6a21\u5f0f&#xff0c;\u8f6c\u53d8\u4e3a\u4e00\u79cd\u6807\u51c6\u5316\u7684\u3001\u81ea\u52a8\u5316\u7684\u3001\u53ef\u4fe1\u8d56\u7684\u201c\u73b0\u4ee3\u5316\u5de5\u4e1a\u751f\u4ea7\u201d\u6a21\u5f0f\u3002<\/p>\n<hr \/>\n<p>\u5c0f\u7ed3&#xff1a;\u4ece\u5de5\u5320\u5230\u5efa\u7b51\u5e08\u7684\u8715\u53d8<\/p>\n<p>\u5728\u672c\u7ae0\u4e2d&#xff0c;\u6211\u4eec\u5b8c\u6210\u4e86\u4ece\u201c\u7b97\u6cd5\u201d\u5230\u201c\u5de5\u7a0b\u201d\u7684\u60ca\u9669\u4e00\u8dc3\u3002\u6211\u4eec\u63a2\u8ba8\u4e86\u6a21\u578b\u8f7b\u91cf\u5316\u7684\u201c\u7626\u8eab\u4e4b\u672f\u201d&#xff0c;\u5b66\u4e60\u4e86\u6a21\u578b\u90e8\u7f72\u7684\u201c\u5b89\u5bb6\u4e4b\u9053\u201d&#xff0c;\u5b9e\u8df5\u4e86API\u5c01\u88c5\u7684\u201c\u6c9f\u901a\u4e4b\u6865\u201d&#xff0c;\u5e76\u6700\u7ec8\u4ef0\u671b\u4e86MLOps\u8fd9\u5ea7\u7cfb\u7edf\u5316\u7ba1\u7406\u7684\u201c\u5b8f\u4f1f\u84dd\u56fe\u201d\u3002\u8fd9\u56db\u5927\u4e3b\u9898&#xff0c;\u5171\u540c\u6784\u6210\u4e86AI\u5de5\u7a0b\u5316\u7684\u575a\u5b9e\u57fa\u77f3\u3002<\/p>\n<p>\u8fd9\u6bb5\u65c5\u7a0b&#xff0c;\u4e5f\u8981\u6c42\u6211\u4eec\u5b8c\u6210\u4e00\u6b21\u601d\u7ef4\u4e0a\u7684\u6df1\u523b\u8f6c\u53d8\u3002\u5982\u679c\u8bf4\u524d\u5341\u4e09\u7ae0&#xff0c;\u6211\u4eec\u66f4\u50cf\u4e00\u4f4d\u4e13\u6ce8\u4e8e\u96d5\u7422\u6280\u827a\u7684\u5de5\u5320&#xff0c;\u8ffd\u6c42\u7684\u662f\u7b97\u6cd5\u7684\u7cbe\u5999\u4e0e\u6a21\u578b\u7684\u7cbe\u51c6&#xff1b;\u90a3\u4e48\u672c\u7ae0&#xff0c;\u6211\u4eec\u5219\u5f00\u59cb\u626e\u6f14\u4e00\u4f4d\u9ad8\u77bb\u8fdc\u77a9\u7684\u5efa\u7b51\u5e08&#xff0c;\u8ffd\u6c42\u7684\u662f\u6574\u4e2a\u7cfb\u7edf\u7684\u7a33\u5065\u3001\u9ad8\u6548\u3001\u53ef\u6269\u5c55\u4e0e\u53ef\u7ef4\u62a4\u3002<\/p>\n<p>\u8bf7\u5c06\u8fd9\u4efd\u5de5\u7a0b\u5316\u7684\u601d\u7ef4&#xff0c;\u878d\u5165\u672a\u6765\u7684\u6bcf\u4e00\u4e2aAI\u9879\u76ee\u4e2d\u3002\u56e0\u4e3a&#xff0c;\u53ea\u6709\u90a3\u4e9b\u80fd\u591f\u7a7f\u8d8a\u5b9e\u9a8c\u5ba4\u7684\u98ce\u66b4&#xff0c;\u7a33\u5065\u5730\u8fd0\u884c\u5728\u771f\u5b9e\u4e16\u754c\u4e2d&#xff0c;\u5e76\u80fd\u4e0d\u65ad\u81ea\u6211\u6f14\u8fdb\u7684\u6a21\u578b&#xff0c;\u624d\u62e5\u6709\u771f\u6b63\u6301\u4e45\u800c\u5f3a\u5927\u7684\u751f\u547d\u529b\u3002\u8fd9&#xff0c;\u4fbf\u662f\u667a\u6167\u843d\u5730\u7684\u6700\u7ec8\u5f62\u6001\u3002<\/p>\n<hr \/>\n<h3>\u7b2c\u4e94\u90e8\u5206&#xff1a;\u5c55\u671b\u7bc7 \u2014\u2014 \u63a2\u7d22\u672a\u6765\u7684\u8fb9\u754c<\/h3>\n<hr \/>\n<h3>\u7b2c\u5341\u4e94\u7ae0&#xff1a;\u524d\u6cbf\u4e13\u9898\u4e0e\u672a\u6765\u8d8b\u52bf<\/h3>\n<p>\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u770b\u89c1\u672a\u6765<\/p>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u5f53\u60a8\u7ffb\u5f00\u8fd9\u6700\u540e\u4e00\u7ae0\u65f6&#xff0c;\u6211\u4eec\u5df2\u5171\u540c\u5b8c\u6210\u4e86\u4e00\u6bb5\u975e\u51e1\u7684\u65c5\u7a0b\u3002\u6211\u4eec\u4ece\u6700\u57fa\u7840\u7684\u795e\u7ecf\u5143\u51fa\u53d1&#xff0c;\u4eb2\u624b\u6784\u5efa\u4e86\u611f\u77e5\u4e16\u754c\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u3001\u7406\u89e3\u5e8f\u5217\u7684\u5faa\u73af\u795e\u7ecf\u7f51\u7edc&#xff0c;\u63a2\u7d22\u4e86\u6ce8\u610f\u529b\u673a\u5236\u7684\u5965\u79d8&#xff0c;\u5e76\u89c1\u8bc1\u4e86\u5b83\u4eec\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u3001\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7b49\u9886\u57df\u7684\u5f3a\u5927\u5a01\u529b\u3002\u6211\u4eec\u4e0d\u4ec5\u5b66\u4f1a\u4e86\u201c\u70bc\u4e39\u201d\u7684\u6280\u827a&#xff0c;\u66f4\u638c\u63e1\u4e86\u201c\u6d4e\u4e16\u201d\u7684\u5de5\u7a0b\u4e4b\u9053\u3002\u53ef\u4ee5\u8bf4&#xff0c;\u6211\u4eec\u5df2\u7ecf\u5728\u6df1\u5ea6\u5b66\u4e60\u8fd9\u7247\u575a\u5b9e\u7684\u5927\u9646\u4e0a&#xff0c;\u5efa\u7acb\u8d77\u4e86\u4e00\u5ea7\u5c5e\u4e8e\u81ea\u5df1\u7684\u3001\u89c6\u91ce\u5f00\u9614\u7684\u77e5\u8bc6\u5821\u5792\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u771f\u6b63\u7684\u63a2\u7d22\u8005&#xff0c;\u4ece\u4e0d\u6ee1\u8db3\u4e8e\u5df2\u77e5\u7684\u7586\u57df\u3002<\/p>\n<p>\u6211\u4eec\u811a\u4e0b\u7684\u8fd9\u7247\u5927\u9646&#xff0c;\u5176\u8fb9\u754c\u4ecd\u5728\u4ee5\u4e00\u79cd\u8fd1\u4e4e\u7206\u70b8\u6027\u7684\u901f\u5ea6&#xff0c;\u5411\u7740\u672a\u77e5\u7684\u8fdc\u65b9\u6269\u5f20\u3002\u65b0\u7684\u601d\u60f3\u3001\u65b0\u7684\u67b6\u6784\u3001\u65b0\u7684\u6311\u6218&#xff0c;\u5982\u96e8\u540e\u6625\u7b0b\u822c\u4e0d\u65ad\u6d8c\u73b0&#xff0c;\u5171\u540c\u5851\u9020\u7740\u4eba\u5de5\u667a\u80fd\u7684\u672a\u6765\u5f62\u6001\u3002\u56e0\u6b64&#xff0c;\u5728\u6211\u4eec\u7ed3\u675f\u8fd9\u6b21\u7cfb\u7edf\u7684\u5b66\u4e60\u4e4b\u65c5\u524d&#xff0c;\u6709\u5fc5\u8981\u767b\u4e0a\u8fd9\u5ea7\u77e5\u8bc6\u5821\u5792\u7684\u6700\u9ad8\u5904&#xff0c;\u5c06\u76ee\u5149\u6295\u5411\u90a3\u7247\u6ce2\u6f9c\u58ee\u9614\u3001\u6b63\u5728\u751f\u6210\u4e2d\u7684\u65b0\u4e16\u754c\u3002<\/p>\n<p>\u672c\u7ae0&#xff0c;\u4fbf\u662f\u8fd9\u5ea7\u4e13\u4e3a\u60a8\u642d\u5efa\u7684\u201c\u77ad\u671b\u5854\u201d\u3002\u6211\u4eec\u5c06\u4e00\u540c\u773a\u671b\u4e94\u4e2a\u6b63\u5728\u6df1\u523b\u5f71\u54cdAI\u53d1\u5c55\u7684\u3001\u6fc0\u52a8\u4eba\u5fc3\u7684\u524d\u6cbf\u65b9\u5411&#xff1a;<\/p>\n<ul>\n<li>\u56fe\u795e\u7ecf\u7f51\u7edc&#xff0c;\u5b83\u8ba9AI\u7684\u89e6\u89d2\u5f97\u4ee5\u4f38\u5411\u7531\u201c\u5173\u7cfb\u201d\u6784\u6210\u7684\u590d\u6742\u4e16\u754c\u3002<\/li>\n<li>\u8054\u90a6\u5b66\u4e60&#xff0c;\u5b83\u4e3a\u7834\u89e3\u6570\u636e\u9690\u79c1\u4e0e\u534f\u4f5c\u7684\u56f0\u5c40&#xff0c;\u63d0\u4f9b\u4e86\u8054\u90a6\u5236\u7684\u667a\u6167\u3002<\/li>\n<li>\u53ef\u89e3\u91ca\u6027AI&#xff0c;\u5b83\u81f4\u529b\u4e8e\u70b9\u4eae\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u9ed1\u7bb1\u201d&#xff0c;\u8ba9\u6211\u4eec\u5f97\u4ee5\u4fe1\u4efb\u5e76\u5ba1\u89c6AI\u7684\u51b3\u7b56\u3002<\/li>\n<li>\u591a\u6a21\u6001\u5b66\u4e60&#xff0c;\u5b83\u6b63\u5f15\u9886AI\u5411\u7740\u4eba\u7c7b\u7684\u611f\u77e5\u65b9\u5f0f\u9760\u62e2&#xff0c;\u878d\u5408\u4e07\u7269\u4fe1\u606f\u3002<\/li>\n<li>AI\u4f26\u7406&#xff0c;\u5b83\u8d85\u8d8a\u4e86\u6280\u672f\u672c\u8eab&#xff0c;\u62f7\u95ee\u7740\u6211\u4eec\u4f5c\u4e3a\u521b\u9020\u8005\u7684\u521d\u5fc3\u4e0e\u8d23\u4efb\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u4e94\u6247\u7a97&#xff0c;\u5206\u522b\u671d\u5411\u4e86\u6570\u636e\u7ed3\u6784\u3001\u534f\u4f5c\u65b9\u5f0f\u3001\u6a21\u578b\u900f\u660e\u5ea6\u3001\u4fe1\u606f\u878d\u5408\u7ef4\u5ea6\u4ee5\u53ca\u6280\u672f\u80cc\u540e\u7684\u4eba\u6587\u5173\u6000\u7684\u672a\u6765\u3002\u5b66\u4e60\u5b83\u4eec&#xff0c;\u4e0d\u4ec5\u662f\u5bf9\u73b0\u6709\u77e5\u8bc6\u4f53\u7cfb\u7684\u8865\u5145\u4e0e\u62d3\u5c55&#xff0c;\u66f4\u662f\u4e3a\u4e86\u6fc0\u53d1\u60a8\u5bf9\u672a\u6765\u7684\u597d\u5947\u4e0e\u601d\u8003&#xff0c;\u4e3a\u60a8\u4e0b\u4e00\u6bb5\u66f4\u7cbe\u5f69\u7684\u63a2\u7d22\u4e4b\u65c5&#xff0c;\u9884\u5148\u70b9\u4eae\u4e00\u76cf\u8fdc\u822a\u7684\u706f\u5854\u3002<\/p>\n<p>\u6765\u5427&#xff0c;\u8ba9\u6211\u4eec\u4e00\u8d77&#xff0c;\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e0a&#xff0c;\u53bb\u770b\u89c1\u90a3\u6b63\u5728\u53d1\u751f\u7684\u672a\u6765\u3002<\/p>\n<h4>15.1 \u56fe\u795e\u7ecf\u7f51\u7edc&#xff08;GNN&#xff09;&#xff1a;\u5f53\u795e\u7ecf\u7f51\u7edc\u9047\u89c1\u201c\u5173\u7cfb\u201d<\/h4>\n<p>\u5728\u6b64\u4e4b\u524d&#xff0c;\u6211\u4eec\u6240\u5904\u7406\u7684\u6570\u636e&#xff0c;\u65e0\u8bba\u662f\u56fe\u50cf\u8fd8\u662f\u6587\u672c&#xff0c;\u90fd\u6709\u4e00\u4e2a\u5171\u540c\u7684\u7279\u70b9&#xff1a;\u5b83\u4eec\u662f\u9ad8\u5ea6\u7ed3\u6784\u5316\u7684\u3002\u4e00\u5f20\u56fe\u50cf&#xff0c;\u53ef\u4ee5\u88ab\u770b\u4f5c\u4e00\u4e2a\u89c4\u6574\u7684\u50cf\u7d20\u7f51\u683c&#xff1b;\u4e00\u6bb5\u6587\u672c&#xff0c;\u53ef\u4ee5\u88ab\u89c6\u4e3a\u4e00\u4e2a\u7ebf\u6027\u7684\u8bcd\u8bed\u5e8f\u5217\u3002\u5728\u6570\u5b66\u4e0a&#xff0c;\u6211\u4eec\u79f0\u8fd9\u7c7b\u6570\u636e\u4e3a\u6b27\u51e0\u91cc\u5f97\u6570\u636e\u3002\u7136\u800c&#xff0c;\u73b0\u5b9e\u4e16\u754c\u8fdc\u6bd4\u8fd9\u8981\u590d\u6742\u548c\u201c\u4e0d\u89c4\u5219\u201d\u3002<\/p>\n<h5>15.1.1 \u8d85\u8d8a\u6b27\u51e0\u91cc\u5f97&#xff1a;\u4e3a\u4f55\u9700\u8981GNN&#xff1f;<\/h5>\n<ul>\n<li>\n<p>\u4f20\u7edf\u6a21\u578b\u7684\u5c40\u9650 \u6211\u4eec\u5f15\u4ee5\u4e3a\u50b2\u7684CNN\u548cRNN&#xff0c;\u5728\u9762\u5bf9\u8bb8\u591a\u73b0\u5b9e\u4e16\u754c\u7684\u95ee\u9898\u65f6&#xff0c;\u4f1a\u663e\u5f97\u529b\u4e0d\u4ece\u5fc3\u3002<\/p>\n<ul>\n<li>CNN\u88ab\u8bbe\u8ba1\u7528\u6765\u5904\u7406\u50cf\u56fe\u50cf\u8fd9\u6837\u7684\u7f51\u683c\u7ed3\u6784\u6570\u636e\u3002\u5b83\u7684\u5377\u79ef\u6838\u4f9d\u8d56\u4e8e\u56fa\u5b9a\u7684\u3001\u5c40\u90e8\u7684\u90bb\u5c45\u5173\u7cfb&#xff08;\u5982\u4e00\u4e2a\u50cf\u7d20\u5468\u56f4\u76848\u4e2a\u50cf\u7d20&#xff09;\u3002<\/li>\n<li>RNN\u5219\u88ab\u8bbe\u8ba1\u7528\u6765\u5904\u7406\u5e8f\u5217\u6570\u636e&#xff0c;\u5176\u6838\u5fc3\u662f\u6355\u6349\u7ebf\u6027\u7684\u3001\u6709\u5e8f\u7684\u4f9d\u8d56\u5173\u7cfb\u3002 \u4f46\u5982\u679c\u6570\u636e\u672c\u8eab\u65e2\u4e0d\u662f\u89c4\u6574\u7684\u7f51\u683c&#xff0c;\u4e5f\u4e0d\u662f\u7ebf\u6027\u7684\u5e8f\u5217&#xff0c;\u90a3\u8be5\u600e\u4e48\u529e&#xff1f;<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u73b0\u5b9e\u4e16\u754c\u7684\u201c\u56fe\u201d\u7ed3\u6784 \u8bf7\u73af\u987e\u6211\u4eec\u6240\u5904\u7684\u4e16\u754c&#xff0c;\u60a8\u4f1a\u53d1\u73b0\u5b83\u672c\u8d28\u4e0a\u662f\u7531\u5b9e\u4f53\u4ee5\u53ca\u5b9e\u4f53\u4e4b\u95f4\u7684\u8054\u7cfb\u6784\u6210\u7684\u201c\u56fe\u201d&#xff1a;<\/p>\n<ul>\n<li>\u793e\u4ea4\u7f51\u7edc&#xff1a;\u6bcf\u4e2a\u4eba\u662f\u4e00\u4e2a\u8282\u70b9&#xff08;Node&#xff09;&#xff0c;\u4eba\u4e0e\u4eba\u4e4b\u95f4\u7684\u597d\u53cb\u5173\u7cfb\u662f\u4e00\u6761\u8fb9&#xff08;Edge&#xff09;\u3002<\/li>\n<li>\u5206\u5b50\u7ed3\u6784&#xff1a;\u6bcf\u4e2a\u539f\u5b50\u662f\u4e00\u4e2a\u8282\u70b9&#xff0c;\u5316\u5b66\u952e\u662f\u8fde\u63a5\u5b83\u4eec\u7684\u8fb9\u3002<\/li>\n<li>\u4ea4\u901a\u8def\u7f51&#xff1a;\u6bcf\u4e2a\u8def\u53e3\u662f\u4e00\u4e2a\u8282\u70b9&#xff0c;\u9053\u8def\u662f\u8fde\u63a5\u5b83\u4eec\u7684\u8fb9\u3002<\/li>\n<li>\u77e5\u8bc6\u56fe\u8c31&#xff1a;\u6bcf\u4e2a\u6982\u5ff5&#xff08;\u5982\u201c\u7231\u56e0\u65af\u5766\u201d\u3001\u201c\u76f8\u5bf9\u8bba\u201d&#xff09;\u662f\u4e00\u4e2a\u8282\u70b9&#xff0c;\u5b83\u4eec\u4e4b\u95f4\u7684\u5173\u7cfb&#xff08;\u5982\u201c\u63d0\u51fa\u8005\u201d&#xff09;\u662f\u4e00\u6761\u8fb9\u3002 \u8fd9\u4e9b\u7531\u8282\u70b9\u548c\u8fb9\u6784\u6210\u7684\u56fe&#xff08;Graph&#xff09;\u7ed3\u6784\u6570\u636e&#xff0c;\u5c31\u662f\u5178\u578b\u7684\u975e\u6b27\u51e0\u91cc\u5f97\u6570\u636e\u3002\u5b83\u4eec\u7684\u90bb\u5c45\u5173\u7cfb\u4e0d\u56fa\u5b9a&#xff0c;\u8282\u70b9\u7684\u8fde\u63a5\u6570\u53ef\u4ee5\u5343\u5dee\u4e07\u522b\u3002\u4e3a\u4e86\u8ba9\u6df1\u5ea6\u5b66\u4e60\u80fd\u591f\u5904\u7406\u8fd9\u79cd\u666e\u904d\u5b58\u5728\u7684\u5173\u7cfb\u578b\u6570\u636e&#xff0c;**\u56fe\u795e\u7ecf\u7f51\u7edc&#xff08;Graph Neural Network, GNN&#xff09;**\u5e94\u8fd0\u800c\u751f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>15.1.2 GNN\u7684\u6838\u5fc3\u601d\u60f3&#xff1a;\u6d88\u606f\u4f20\u9012<\/h5>\n<p>GNN\u5982\u4f55\u5728\u4e00\u4e2a\u4e0d\u89c4\u5219\u7684\u56fe\u4e0a\u8fdb\u884c\u5b66\u4e60&#xff1f;\u5176\u6838\u5fc3\u601d\u60f3\u975e\u5e38\u7b26\u5408\u76f4\u89c9&#xff1a;\u4e00\u4e2a\u8282\u70b9\u7684\u7279\u6027&#xff0c;\u5e94\u8be5\u7531\u5b83\u5468\u56f4\u90bb\u5c45\u7684\u7279\u6027\u6765\u5171\u540c\u5b9a\u4e49\u3002 \u6b63\u5982\u201c\u8fd1\u6731\u8005\u8d64&#xff0c;\u8fd1\u58a8\u8005\u9ed1\u201d&#xff0c;\u8981\u4e86\u89e3\u4e00\u4e2a\u4eba&#xff0c;\u5148\u770b\u770b\u4ed6\u7684\u670b\u53cb\u3002GNN\u5c06\u8fd9\u4e2a\u6734\u7d20\u7684\u9053\u7406&#xff0c;\u53d8\u6210\u4e86\u4e00\u5957\u4f18\u96c5\u7684\u6570\u5b66\u8303\u5f0f\u2014\u2014\u6d88\u606f\u4f20\u9012&#xff08;Message Passing&#xff09;\u3002<\/p>\n<ul>\n<li>\n<p>\u8282\u70b9\u5982\u4f55\u5b66\u4e60 \u60f3\u8c61\u4e00\u4e0b&#xff0c;\u56fe\u4e2d\u7684\u6bcf\u4e2a\u8282\u70b9\u90fd\u62e5\u6709\u4e00\u4e2a\u521d\u59cb\u7684\u7279\u5f81\u5411\u91cf&#xff08;\u4f8b\u5982&#xff0c;\u5728\u793e\u4ea4\u7f51\u7edc\u4e2d&#xff0c;\u53ef\u4ee5\u662f\u4e00\u4e2a\u7528\u6237\u7684\u5e74\u9f84\u3001\u6027\u522b\u3001\u5174\u8da3\u6807\u7b7e\u7b49&#xff09;\u3002GNN\u7684\u76ee\u6807&#xff0c;\u662f\u901a\u8fc7\u5b66\u4e60&#xff0c;\u5c06\u8fd9\u4e2a\u521d\u59cb\u7279\u5f81&#xff0c;\u66f4\u65b0\u4e3a\u4e00\u4e2a\u66f4\u5177\u4fe1\u606f\u91cf\u3001\u8574\u542b\u4e86\u5176\u5728\u56fe\u4e2d\u6240\u5904\u7ed3\u6784\u4fe1\u606f\u7684\u201c\u9ad8\u7ea7\u201d\u7279\u5f81\u5411\u91cf\u3002<\/p>\n<\/li>\n<li>\n<p>\u6d88\u606f\u4f20\u9012\u8303\u5f0f \u8fd9\u4e2a\u66f4\u65b0\u8fc7\u7a0b\u662f\u8fed\u4ee3\u8fdb\u884c\u7684\u3002\u5728\u6bcf\u4e00\u8f6e&#xff08;\u6216\u6bcf\u4e00\u5c42&#xff09;\u6d88\u606f\u4f20\u9012\u4e2d&#xff0c;\u56fe\u4e2d\u7684\u6bcf\u4e00\u4e2a\u8282\u70b9\u90fd\u4f1a\u540c\u6b65\u6267\u884c\u4ee5\u4e0b\u4e09\u6b65\u64cd\u4f5c&#xff1a;<\/p>\n<li>\u6536\u96c6&#xff08;Gather&#xff09;&#xff1a;\u6bcf\u4e2a\u8282\u70b9\u50cf\u4e00\u4e2a\u201c\u63a5\u6536\u5668\u201d&#xff0c;\u4ece\u5b83\u7684\u6240\u6709\u76f4\u63a5\u76f8\u8fde\u7684\u90bb\u5c45\u8282\u70b9\u90a3\u91cc&#xff0c;\u6536\u96c6\u5b83\u4eec\u7684\u5f53\u524d\u7279\u5f81\u5411\u91cf&#xff08;\u8fd9\u4e9b\u5411\u91cf\u53ef\u4ee5\u88ab\u770b\u4f5c\u662f\u90bb\u5c45\u53d1\u6765\u7684\u201c\u6d88\u606f\u201d&#xff09;\u3002<\/li>\n<li>\u805a\u5408&#xff08;Aggregate&#xff09;&#xff1a;\u8282\u70b9\u5c06\u6536\u96c6\u5230\u7684\u6240\u6709\u90bb\u5c45\u7684\u201c\u6d88\u606f\u201d&#xff08;\u7279\u5f81\u5411\u91cf&#xff09;&#xff0c;\u7528\u4e00\u79cd\u7f6e\u6362\u4e0d\u53d8\u7684\u65b9\u5f0f&#xff08;\u5373\u4e0e\u90bb\u5c45\u7684\u987a\u5e8f\u65e0\u5173&#xff09;\u805a\u5408\u8d77\u6765\u3002\u6700\u5e38\u89c1\u7684\u805a\u5408\u51fd\u6570\u662f\u6c42\u548c\u3001\u6c42\u5e73\u5747\u6216\u6c42\u6700\u5927\u503c\u3002<\/li>\n<li>\u66f4\u65b0&#xff08;Update&#xff09;&#xff1a;\u8282\u70b9\u5c06\u805a\u5408\u540e\u7684\u90bb\u5c45\u4fe1\u606f\u4e0e\u8282\u70b9\u81ea\u8eab\u7684\u4e0a\u4e00\u8f6e\u65e7\u4fe1\u606f\u7ed3\u5408\u8d77\u6765&#xff0c;\u7136\u540e\u5c06\u8fd9\u4e2a\u7ec4\u5408\u540e\u7684\u7ed3\u679c&#xff0c;\u5582\u7ed9\u4e00\u4e2a\u53ef\u5b66\u4e60\u7684\u795e\u7ecf\u7f51\u7edc&#xff08;\u901a\u5e38\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u5168\u8fde\u63a5\u5c42&#xff09;&#xff0c;\u8ba1\u7b97\u51fa\u8be5\u8282\u70b9\u5728\u672c\u8f6e\u7684\u3001\u5168\u65b0\u7684\u7279\u5f81\u5411\u91cf\u3002<\/li>\n<\/li>\n<li>\n<p>\u76f4\u89c2\u7406\u89e3 \u8fd9\u4e2a\u8fc7\u7a0b\u975e\u5e38\u7cbe\u5999&#xff1a;<\/p>\n<ul>\n<li>\u7ecf\u8fc7\u7b2c\u4e00\u8f6e\u6d88\u606f\u4f20\u9012&#xff0c;\u6bcf\u4e2a\u8282\u70b9\u90fd\u878d\u5408\u4e86\u5176\u4e00\u8df3\u90bb\u5c45\u7684\u4fe1\u606f\u3002<\/li>\n<li>\u7ecf\u8fc7\u7b2c\u4e8c\u8f6e\u6d88\u606f\u4f20\u9012&#xff0c;\u4fe1\u606f\u4f1a\u8fdb\u4e00\u6b65\u4f20\u64ad&#xff0c;\u6bcf\u4e2a\u8282\u70b9\u5b9e\u9645\u4e0a\u5df2\u7ecf\u878d\u5408\u4e86\u5176\u4e8c\u8df3\u90bb\u5c45\u7684\u4fe1\u606f\u3002 \u901a\u8fc7\u5806\u53e0\u591a\u5c42GNN&#xff08;\u5373\u8fdb\u884c\u591a\u8f6e\u6d88\u606f\u4f20\u9012&#xff09;&#xff0c;\u6bcf\u4e2a\u8282\u70b9\u7684\u6700\u7ec8\u7279\u5f81\u5411\u91cf&#xff0c;\u5c31\u5982\u540c\u4e00\u4e2a\u4ee5\u5b83\u4e3a\u4e2d\u5fc3\u3001\u4e0d\u65ad\u5411\u5916\u6269\u5c55\u7684\u201c\u6d9f\u6f2a\u201d&#xff0c;\u8574\u542b\u4e86\u5176\u5728\u56fe\u4e2d\u7684\u5c40\u90e8\u4e43\u81f3\u5168\u5c40\u7684\u7ed3\u6784\u4fe1\u606f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>15.1.3 GNN\u7684\u5e94\u7528\u573a\u666f\u4e0e\u5c55\u671b<\/h5>\n<p>\u88c5\u5907\u4e86\u6d88\u606f\u4f20\u9012\u80fd\u529b\u7684GNN&#xff0c;\u4e3a\u5904\u7406\u56fe\u6570\u636e\u63d0\u4f9b\u4e86\u5f3a\u5927\u7684\u6b66\u5668&#xff0c;\u5e76\u5df2\u5728\u4f17\u591a\u9886\u57df\u5c55\u73b0\u51fa\u5de8\u5927\u6f5c\u529b\u3002<\/p>\n<ul>\n<li>\u8282\u70b9\u5206\u7c7b&#xff08;Node Classification&#xff09;&#xff1a;\u9884\u6d4b\u56fe\u4e2d\u67d0\u4e2a\u8282\u70b9\u7684\u5c5e\u6027\u3002\u4f8b\u5982&#xff0c;\u5728\u793e\u4ea4\u7f51\u7edc\u4e2d&#xff0c;\u6839\u636e\u7528\u6237\u7684\u793e\u4ea4\u5173\u7cfb&#xff0c;\u9884\u6d4b\u5176\u6d88\u8d39\u5174\u8da3\u6216\u653f\u6cbb\u503e\u5411\u3002<\/li>\n<li>\u56fe\u5206\u7c7b&#xff08;Graph Classification&#xff09;&#xff1a;\u9884\u6d4b\u6574\u4e2a\u56fe\u7684\u5c5e\u6027\u3002\u4f8b\u5982&#xff0c;\u5728\u836f\u7269\u53d1\u73b0\u4e2d&#xff0c;\u5c06\u5206\u5b50\u770b\u4f5c\u4e00\u4e2a\u56fe&#xff0c;\u9884\u6d4b\u8be5\u5206\u5b50\u662f\u5426\u5177\u6709\u6297\u764c\u6d3b\u6027\u3002<\/li>\n<li>\u94fe\u63a5\u9884\u6d4b&#xff08;Link Prediction&#xff09;&#xff1a;\u9884\u6d4b\u4e24\u4e2a\u8282\u70b9\u4e4b\u95f4\u662f\u5426\u5b58\u5728\u8fde\u63a5\u3002\u4f8b\u5982&#xff0c;\u5728\u63a8\u8350\u7cfb\u7edf\u4e2d&#xff0c;\u5411\u7528\u6237\u63a8\u8350\u53ef\u80fd\u8ba4\u8bc6\u7684\u670b\u53cb\u6216\u53ef\u80fd\u611f\u5174\u8da3\u7684\u5546\u54c1\u3002<\/li>\n<\/ul>\n<p>\u5c55\u671b\u672a\u6765&#xff0c;GNN\u9886\u57df\u7684\u53d1\u5c55\u65b9\u5174\u672a\u827e\u3002\u7814\u7a76\u8005\u4eec\u6b63\u5728\u63a2\u7d22\u5982\u4f55\u5c06\u5176\u4e0e\u6ce8\u610f\u529b\u673a\u5236\u3001Transformer\u7b49\u66f4\u5f3a\u5927\u7684\u67b6\u6784\u76f8\u7ed3\u5408&#xff0c;\u4ee5\u5904\u7406\u66f4\u590d\u6742\u7684\u52a8\u6001\u56fe\u3001\u5f02\u6784\u56fe&#xff0c;\u5e76\u6784\u5efa\u80fd\u591f\u7406\u89e3\u66f4\u590d\u6742\u5173\u7cfb\u63a8\u7406\u7684AI\u7cfb\u7edf\u3002GNN\u6b63\u5728\u6210\u4e3a\u7ee7CNN\u548cRNN\u4e4b\u540e&#xff0c;\u6df1\u5ea6\u5b66\u4e60\u5de5\u5177\u7bb1\u4e2d\u53c8\u4e00\u4e2a\u4e0d\u53ef\u6216\u7f3a\u7684\u6838\u5fc3\u7ec4\u4ef6\u3002<\/p>\n<h4>15.2 \u8054\u90a6\u5b66\u4e60&#xff08;Federated Learning&#xff09;&#xff1a;\u6570\u636e\u5b64\u5c9b\u4e0a\u7684\u201c\u8054\u90a6\u201d\u667a\u6167<\/h4>\n<p>\u5982\u679c\u8bf4GNN\u89e3\u51b3\u4e86\u6570\u636e\u7ed3\u6784\u4e0a\u7684\u6311\u6218&#xff0c;\u90a3\u4e48\u8054\u90a6\u5b66\u4e60\u5219\u81f4\u529b\u4e8e\u89e3\u51b3\u6570\u636e\u534f\u4f5c\u65b9\u5f0f\u4e0a\u7684\u4e00\u4e2a\u6839\u672c\u6027\u96be\u9898&#xff1a;\u5728\u6570\u636e\u65e0\u6cd5\u96c6\u4e2d\u7684\u60c5\u51b5\u4e0b&#xff0c;\u6211\u4eec\u5982\u4f55\u8fdb\u884c\u6709\u6548\u7684\u673a\u5668\u5b66\u4e60&#xff1f;<\/p>\n<h5>15.2.1 \u9690\u79c1\u7684\u56f0\u5883&#xff1a;\u6570\u636e\u65e0\u6cd5\u96c6\u4e2d\u600e\u4e48\u529e&#xff1f;<\/h5>\n<ul>\n<li>\n<p>\u4f20\u7edf\u6a21\u5f0f\u7684\u74f6\u9888 \u6211\u4eec\u672c\u4e66\u4e2d\u6240\u6709\u9879\u76ee\u7684\u9ed8\u8ba4\u524d\u63d0&#xff0c;\u90fd\u662f\u6570\u636e\u53ef\u4ee5\u88ab\u96c6\u4e2d\u5230\u4e00\u4e2a\u5730\u65b9&#xff08;\u5982\u4e00\u53f0\u670d\u52a1\u5668\u6216\u4e00\u4e2a\u6570\u636e\u4e2d\u5fc3&#xff09;\u8fdb\u884c\u7edf\u4e00\u7684\u8bad\u7ec3\u3002\u8fd9\u662f\u4f20\u7edf\u7684\u3001\u4e2d\u5fc3\u5316\u7684\u673a\u5668\u5b66\u4e60\u6a21\u5f0f\u3002<\/p>\n<\/li>\n<li>\n<p>\u9690\u79c1\u4e0e\u6cd5\u89c4\u7684\u6311\u6218 \u7136\u800c&#xff0c;\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d&#xff0c;\u8fd9\u4e2a\u524d\u63d0\u6b63\u53d8\u5f97\u8d8a\u6765\u8d8a\u5962\u4f88\u3002<\/p>\n<ul>\n<li>\u6570\u636e\u9690\u79c1&#xff1a;\u7528\u6237\u7684\u4e2a\u4eba\u6570\u636e&#xff0c;\u5c24\u5176\u662f\u533b\u7597\u8bb0\u5f55\u3001\u91d1\u878d\u4ea4\u6613\u3001\u804a\u5929\u5185\u5bb9\u7b49&#xff0c;\u662f\u9ad8\u5ea6\u654f\u611f\u7684\u3002\u5c06\u5176\u4e0a\u4f20\u5230\u4e00\u4e2a\u4e2d\u592e\u670d\u52a1\u5668&#xff0c;\u5b58\u5728\u5de8\u5927\u7684\u9690\u79c1\u6cc4\u9732\u98ce\u9669\u3002<\/li>\n<li>\u6cd5\u5f8b\u6cd5\u89c4&#xff1a;\u5168\u7403\u8303\u56f4\u5185&#xff0c;\u6570\u636e\u4fdd\u62a4\u6cd5\u89c4&#xff08;\u5982\u6b27\u76df\u7684GDPR\u3001\u4e2d\u56fd\u7684\u300a\u4e2a\u4eba\u4fe1\u606f\u4fdd\u62a4\u6cd5\u300b&#xff09;\u65e5\u76ca\u4e25\u683c&#xff0c;\u5bf9\u6570\u636e\u7684\u8de8\u5883\u6d41\u52a8\u548c\u96c6\u4e2d\u5b58\u50a8\u65bd\u52a0\u4e86\u4e25\u683c\u7684\u9650\u5236\u3002<\/li>\n<li>\u5546\u4e1a\u58c1\u5792&#xff1a;\u4e0d\u540c\u516c\u53f8\u4e4b\u95f4&#xff0c;\u51fa\u4e8e\u5546\u4e1a\u7ade\u4e89\u7684\u8003\u8651&#xff0c;\u4e5f\u5f80\u5f80\u4e0d\u613f\u610f\u5171\u4eab\u81ea\u5df1\u7684\u6838\u5fc3\u6570\u636e\u3002 \u8fd9\u4e9b\u56e0\u7d20&#xff0c;\u5171\u540c\u5bfc\u81f4\u4e86\u6570\u636e\u88ab\u56f0\u5728\u5404\u81ea\u4ea7\u751f\u7684\u5730\u65b9&#xff0c;\u5f62\u6210\u4e86\u4e00\u5ea7\u5ea7\u65e0\u6cd5\u8fde\u901a\u7684**\u201c\u6570\u636e\u5b64\u5c9b\u201d**\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>15.2.2 \u8054\u90a6\u5b66\u4e60\u7684\u667a\u6167&#xff1a;\u201c\u6a21\u578b\u79fb\u52a8&#xff0c;\u6570\u636e\u4e0d\u52a8\u201d<\/h5>\n<p>\u9762\u5bf9\u6570\u636e\u5b64\u5c9b\u7684\u56f0\u5c40&#xff0c;\u8054\u90a6\u5b66\u4e60&#xff08;Federated Learning, FL&#xff09;\u63d0\u51fa\u4e86\u4e00\u79cd\u6781\u5177\u521b\u89c1\u7684\u3001\u98a0\u8986\u6027\u7684\u5206\u5e03\u5f0f\u5b66\u4e60\u8303\u5f0f\u3002\u5176\u6838\u5fc3\u601d\u60f3\u7b80\u6d01\u800c\u6df1\u523b&#xff1a;\u201c\u6570\u636e\u4e0d\u52a8&#xff0c;\u6a21\u578b\u79fb\u52a8\u201d\u3002<\/p>\n<p>\u65e2\u7136\u6570\u636e\u4e0d\u80fd\u88ab\u6c47\u96c6\u5230\u6a21\u578b\u8fd9\u91cc&#xff0c;\u90a3\u5c31\u8ba9\u6a21\u578b\u201c\u767b\u95e8\u62dc\u8bbf\u201d&#xff0c;\u53bb\u5230\u6570\u636e\u6240\u5728\u7684\u5730\u65b9\u8fdb\u884c\u5b66\u4e60\u3002<\/p>\n<ul>\n<li>\u5de5\u4f5c\u6d41\u7a0b\u00a0\u4e00\u4e2a\u5178\u578b\u7684\u8054\u90a6\u5b66\u4e60\u8fc7\u7a0b&#xff0c;\u5982\u540c\u4e00\u6b21\u4e25\u8c28\u7684\u201c\u5de1\u56de\u6559\u5b66\u201d&#xff1a;\n<li>\u5206\u53d1&#xff08;Dispatch&#xff09;&#xff1a;\u4e2d\u592e\u670d\u52a1\u5668&#xff08;\u534f\u8c03\u65b9&#xff09;\u5c06\u4e00\u4e2a\u521d\u59cb\u7684\u3001\u672a\u7ecf\u8bad\u7ec3\u7684\u5168\u5c40\u6a21\u578b&#xff08;Global Model&#xff09;&#xff0c;\u901a\u8fc7\u7f51\u7edc\u5206\u53d1\u7ed9\u6240\u6709\u53c2\u4e0e\u534f\u4f5c\u7684\u5ba2\u6237\u7aef&#xff08;\u4f8b\u5982&#xff0c;\u6570\u767e\u4e07\u90e8\u624b\u673a&#xff0c;\u6216\u6570\u5341\u5bb6\u533b\u9662&#xff09;\u3002<\/li>\n<li>\u672c\u5730\u8bad\u7ec3&#xff08;Local Training&#xff09;&#xff1a;\u6bcf\u4e00\u4e2a\u5ba2\u6237\u7aef\u5728\u63a5\u6536\u5230\u6a21\u578b\u540e&#xff0c;\u4f7f\u7528\u81ea\u5df1\u672c\u5730\u7684\u3001\u4ece\u4e0d\u5916\u4f20\u7684\u6570\u636e&#xff0c;\u5bf9\u6a21\u578b\u8fdb\u884c\u51e0\u8f6e\u8bad\u7ec3\u3002\u8fd9\u4e2a\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u5c06\u672c\u5730\u6570\u636e\u7684\u201c\u77e5\u8bc6\u201d&#xff0c;\u6ce8\u5165\u5230\u4e86\u6a21\u578b\u4e2d&#xff0c;\u5f62\u6210\u4e86\u4e00\u4e2a\u672c\u5730\u66f4\u65b0\u540e\u7684\u6a21\u578b\u3002<\/li>\n<li>\u5b89\u5168\u805a\u5408&#xff08;Secure Aggregation&#xff09;&#xff1a;\u8bad\u7ec3\u5b8c\u6210\u540e&#xff0c;\u5ba2\u6237\u7aef\u5e76\u4e0d\u4f1a\u4e0a\u4f20\u81ea\u5df1\u7684\u6570\u636e&#xff0c;\u800c\u662f\u53ea\u5c06\u6a21\u578b\u7684\u66f4\u65b0\u4fe1\u606f&#xff08;\u4f8b\u5982&#xff0c;\u6743\u91cd\u7684\u53d8\u5316\u91cf\u6216\u68af\u5ea6&#xff09;\u8fdb\u884c\u52a0\u5bc6\u6216\u4f7f\u7528\u5176\u4ed6\u9690\u79c1\u4fdd\u62a4\u6280\u672f\u5904\u7406\u540e&#xff0c;\u5b89\u5168\u5730\u4e0a\u4f20\u7ed9\u4e2d\u592e\u670d\u52a1\u5668\u3002<\/li>\n<li>\u66f4\u65b0\u5168\u5c40\u6a21\u578b&#xff08;Global Update&#xff09;&#xff1a;\u4e2d\u592e\u670d\u52a1\u5668\u7b49\u5f85\u5e76\u6536\u96c6\u6765\u81ea\u591a\u4e2a\u5ba2\u6237\u7aef\u7684\u6a21\u578b\u66f4\u65b0\u3002\u7136\u540e&#xff0c;\u5b83\u4f7f\u7528\u4e00\u79cd\u805a\u5408\u7b97\u6cd5&#xff08;\u6700\u7ecf\u5178\u7684\u662f\u8054\u90a6\u5e73\u5747&#xff0c;Federated Averaging&#xff09;&#xff0c;\u5c06\u6240\u6709\u8fd9\u4e9b\u66f4\u65b0\u91cf\u8fdb\u884c\u52a0\u6743\u5e73\u5747&#xff0c;\u7528\u6765\u66f4\u65b0\u5168\u5c40\u6a21\u578b\u3002\u8fd9\u4e2a\u805a\u5408\u540e\u7684\u5168\u5c40\u6a21\u578b&#xff0c;\u5c31\u878d\u5408\u4e86\u591a\u4e2a\u6570\u636e\u5b64\u5c9b\u7684\u5171\u540c\u667a\u6167\u3002<\/li>\n<li>\u8fd9\u4e2a\u201c\u5206\u53d1-\u8bad\u7ec3-\u805a\u5408\u201d\u7684\u5faa\u73af\u4f1a\u8fed\u4ee3\u8fdb\u884c\u591a\u8f6e&#xff0c;\u6bcf\u4e00\u8f6e&#xff0c;\u5168\u5c40\u6a21\u578b\u90fd\u4f1a\u5438\u6536\u66f4\u591a\u53c2\u4e0e\u65b9\u7684\u77e5\u8bc6&#xff0c;\u6700\u7ec8\u6536\u655b\u4e3a\u4e00\u4e2a\u6027\u80fd\u5f3a\u5927\u7684\u6a21\u578b&#xff0c;\u800c\u5168\u7a0b\u6ca1\u6709\u4efb\u4f55\u539f\u59cb\u6570\u636e\u79bb\u5f00\u8fc7\u672c\u5730\u8bbe\u5907\u3002<\/li>\n<\/li>\n<\/ul>\n<h5>15.2.3 \u5e94\u7528\u4e0e\u6311\u6218<\/h5>\n<ul>\n<li>\n<p>\u5e94\u7528\u573a\u666f \u8054\u90a6\u5b66\u4e60\u4e3a\u8bb8\u591a\u8fc7\u53bb\u56e0\u9690\u79c1\u95ee\u9898\u800c\u96be\u4ee5\u5f00\u5c55\u7684AI\u5e94\u7528&#xff0c;\u6253\u5f00\u4e86\u5927\u95e8&#xff1a;<\/p>\n<ul>\n<li>\u667a\u80fd\u624b\u673a&#xff1a;\u624b\u673a\u8f93\u5165\u6cd5\u53ef\u4ee5\u5229\u7528\u6210\u5343\u4e0a\u4e07\u7528\u6237\u7684\u672c\u5730\u8f93\u5165\u4e60\u60ef\u6765\u4f18\u5316\u9884\u6d4b\u6a21\u578b&#xff0c;\u800c\u65e0\u9700\u4e0a\u4f20\u7528\u6237\u7684\u804a\u5929\u8bb0\u5f55\u3002<\/li>\n<li>\u533b\u7597\u5065\u5eb7&#xff1a;\u591a\u5bb6\u533b\u9662\u53ef\u4ee5\u5728\u4e0d\u5171\u4eab\u4efb\u4f55\u75c5\u4eba\u75c5\u5386\u7684\u524d\u63d0\u4e0b&#xff0c;\u8054\u5408\u8bad\u7ec3\u4e00\u4e2a\u66f4\u7cbe\u51c6\u7684\u75be\u75c5\u8bca\u65ad\u6a21\u578b\u3002<\/li>\n<li>\u91d1\u878d\u670d\u52a1&#xff1a;\u591a\u5bb6\u94f6\u884c\u53ef\u4ee5\u8054\u5408\u8bad\u7ec3\u4e00\u4e2a\u66f4\u5f3a\u5927\u7684\u53cd\u6b3a\u8bc8\u6a21\u578b&#xff0c;\u800c\u65e0\u9700\u66b4\u9732\u5404\u81ea\u7684\u5ba2\u6237\u4ea4\u6613\u6570\u636e\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u6311\u6218 \u8054\u90a6\u5b66\u4e60\u867d\u7136\u524d\u666f\u5e7f\u9614&#xff0c;\u4f46\u4ecd\u9762\u4e34\u8bf8\u591a\u6311\u6218&#xff0c;\u5982&#xff1a;<\/p>\n<ul>\n<li>\u901a\u4fe1\u5f00\u9500&#xff1a;\u5728\u5ba2\u6237\u7aef\u548c\u670d\u52a1\u5668\u4e4b\u95f4\u53cd\u590d\u4f20\u8f93\u6a21\u578b\u53c2\u6570&#xff0c;\u53ef\u80fd\u4f1a\u975e\u5e38\u8017\u8d39\u5e26\u5bbd\u3002<\/li>\n<li>\u6570\u636e\u5f02\u6784\u6027&#xff1a;\u4e0d\u540c\u5ba2\u6237\u7aef\u7684\u6570\u636e\u5206\u5e03\u53ef\u80fd\u5dee\u5f02\u5de8\u5927&#xff08;Non-IID\u95ee\u9898&#xff09;&#xff0c;\u8fd9\u7ed9\u6a21\u578b\u6536\u655b\u5e26\u6765\u4e86\u56f0\u96be\u3002<\/li>\n<li>\u7cfb\u7edf\u5f02\u6784\u6027&#xff1a;\u4e0d\u540c\u5ba2\u6237\u7aef\u7684\u8ba1\u7b97\u80fd\u529b\u3001\u7f51\u7edc\u72b6\u51b5\u5343\u5dee\u4e07\u522b&#xff0c;\u5982\u4f55\u9ad8\u6548\u534f\u8c03\u662f\u4e00\u4e2a\u96be\u9898\u3002<\/li>\n<li>\u66f4\u9ad8\u7ea7\u7684\u9690\u79c1\u5b89\u5168&#xff1a;\u5373\u4f7f\u4e0d\u4e0a\u4f20\u6570\u636e&#xff0c;\u6076\u610f\u653b\u51fb\u8005\u4ecd\u6709\u53ef\u80fd\u4ece\u6a21\u578b\u7684\u66f4\u65b0\u91cf\u4e2d&#xff0c;\u53cd\u63a8\u51fa\u90e8\u5206\u539f\u59cb\u6570\u636e\u7684\u4fe1\u606f&#xff08;\u6a21\u578b\u53cd\u6f14\u653b\u51fb&#xff09;\u3002\u56e0\u6b64&#xff0c;\u8fd8\u9700\u8981\u7ed3\u5408\u5dee\u5206\u9690\u79c1\u3001\u540c\u6001\u52a0\u5bc6\u7b49\u66f4\u5f3a\u7684\u9690\u79c1\u6280\u672f\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>\u8054\u90a6\u5b66\u4e60\u4ee3\u8868\u4e86\u4e00\u79cd\u5168\u65b0\u7684\u3001\u53bb\u4e2d\u5fc3\u5316\u7684AI\u534f\u4f5c\u8303\u5f0f\u3002\u5b83\u8bd5\u56fe\u5728\u201c\u6570\u636e\u4ef7\u503c\u201d\u4e0e\u201c\u9690\u79c1\u4fdd\u62a4\u201d\u8fd9\u4e24\u4e2a\u770b\u4f3c\u77db\u76fe\u7684\u76ee\u6807\u4e4b\u95f4&#xff0c;\u67b6\u8d77\u4e00\u5ea7\u7cbe\u5de7\u7684\u6865\u6881&#xff0c;\u5f15\u9886\u6211\u4eec\u8fdb\u5165\u4e00\u4e2a\u66f4\u5b89\u5168\u3001\u66f4\u516c\u5e73\u7684AI\u534f\u4f5c\u65b0\u65f6\u4ee3\u3002<\/p>\n<h4>15.3 \u53ef\u89e3\u91ca\u6027AI&#xff08;XAI&#xff09;&#xff1a;\u6253\u5f00\u795e\u7ecf\u7f51\u7edc\u7684\u201c\u9ed1\u7bb1\u201d<\/h4>\n<p>\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b&#xff0c;\u5c24\u5176\u662f\u90a3\u4e9b\u6df1\u9083\u800c\u590d\u6742\u7684\u7f51\u7edc&#xff0c;\u5e38\u5e38\u88ab\u4eba\u4eec\u8bdf\u75c5\u4e3a\u4e00\u4e2a\u201c\u9ed1\u7bb1\u201d\u3002\u6211\u4eec\u77e5\u9053\u8f93\u5165\u662f\u4ec0\u4e48&#xff0c;\u4e5f\u770b\u5230\u4e86\u8f93\u51fa\u662f\u4ec0\u4e48&#xff0c;\u4f46\u5bf9\u4e8e\u5176\u5185\u90e8\u6570\u4ee5\u4ebf\u8ba1\u7684\u53c2\u6570\u662f\u5982\u4f55\u76f8\u4e92\u4f5c\u7528&#xff0c;\u6700\u7ec8\u63a8\u5bfc\u51fa\u8fd9\u4e2a\u7ed3\u679c\u7684&#xff0c;\u6211\u4eec\u5374\u77e5\u4e4b\u751a\u5c11\u3002\u8fd9\u79cd\u4e0d\u900f\u660e\u6027&#xff0c;\u5728\u63a2\u7d22\u6027\u7814\u7a76\u4e2d\u6216\u8bb8\u53ef\u4ee5\u63a5\u53d7&#xff0c;\u4f46\u5728\u8bb8\u591a\u9ad8\u98ce\u9669\u7684\u73b0\u5b9e\u573a\u666f\u4e2d&#xff0c;\u5374\u662f\u4e00\u4e2a\u81f4\u547d\u7684\u7f3a\u9677\u3002<\/p>\n<h5>15.3.1 \u201c\u9ed1\u7bb1\u201d\u4e4b\u95ee&#xff1a;\u6211\u4eec\u80fd\u4fe1\u4efb\u4e00\u4e2a\u65e0\u6cd5\u7406\u89e3\u7684\u51b3\u7b56\u5417&#xff1f;<\/h5>\n<ul>\n<li>\n<p>\u95ee\u9898\u7684\u63d0\u51fa \u60f3\u8c61\u4ee5\u4e0b\u573a\u666f&#xff1a;<\/p>\n<ul>\n<li>\u4e00\u4e2aAI\u7cfb\u7edf\u62d2\u7edd\u4e86\u60a8\u7684\u8d37\u6b3e\u7533\u8bf7&#xff0c;\u4f46\u65e0\u6cd5\u7ed9\u51fa\u4efb\u4f55\u7406\u7531\u3002<\/li>\n<li>\u4e00\u4e2a\u533b\u7597AI\u8bca\u65ad\u60a8\u60a3\u6709\u67d0\u79cd\u7f55\u89c1\u75be\u75c5&#xff0c;\u4f46\u533b\u751f\u65e0\u6cd5\u7406\u89e3\u5176\u8bca\u65ad\u4f9d\u636e\u3002<\/li>\n<li>\u4e00\u4e2a\u81ea\u52a8\u9a7e\u9a76\u6c7d\u8f66\u5728\u4e8b\u6545\u4e2d\u505a\u51fa\u4e86\u4e00\u4e2a\u81f4\u547d\u7684\u8f6c\u5411\u51b3\u7b56&#xff0c;\u4f46\u8c03\u67e5\u4eba\u5458\u65e0\u6cd5\u590d\u76d8\u5176\u201c\u51b3\u7b56\u903b\u8f91\u201d\u3002 \u5728\u8fd9\u4e9b\u573a\u666f\u4e0b&#xff0c;\u4e00\u4e2a\u6027\u80fd\u518d\u9ad8\u4f46\u65e0\u6cd5\u89e3\u91ca\u7684\u201c\u9ed1\u7bb1\u201d\u51b3\u7b56&#xff0c;\u662f\u5b8c\u5168\u4e0d\u53ef\u63a5\u53d7\u7684\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4e3a\u4f55\u9700\u8981\u53ef\u89e3\u91ca\u6027&#xff08;Explainable AI, XAI&#xff09; \u6211\u4eec\u8ffd\u6c42\u53ef\u89e3\u91ca\u6027&#xff0c;\u5176\u76ee\u7684\u8fdc\u8d85\u6ee1\u8db3\u4eba\u7c7b\u7684\u597d\u5947\u5fc3\u3002<\/p>\n<ul>\n<li>\u4fe1\u4efb\u4e0e\u8c03\u8bd5&#xff1a;\u7406\u89e3\u6a21\u578b\u201c\u4e3a\u4f55\u201d\u505a\u51fa\u67d0\u4e2a\u5224\u65ad&#xff0c;\u662f\u5efa\u7acb\u4fe1\u4efb\u3001\u53d1\u73b0\u6a21\u578b\u7f3a\u9677\u5e76\u8fdb\u884c\u6709\u6548\u8c03\u8bd5\u7684\u524d\u63d0\u3002<\/li>\n<li>\u516c\u5e73\u4e0e\u5408\u89c4&#xff1a;XAI\u53ef\u4ee5\u5e2e\u52a9\u6211\u4eec\u68c0\u6d4b\u548c\u7ea0\u6b63\u6a21\u578b\u4e2d\u53ef\u80fd\u5b58\u5728\u7684\u504f\u89c1&#xff0c;\u786e\u4fdd\u5176\u51b3\u7b56\u7684\u516c\u5e73\u6027&#xff0c;\u5e76\u6ee1\u8db3\u76d1\u7ba1\u673a\u6784\u7684\u8981\u6c42\u3002<\/li>\n<li>\u77e5\u8bc6\u53d1\u73b0&#xff1a;\u5728\u79d1\u5b66\u7814\u7a76\u4e2d&#xff0c;\u89e3\u91ca\u6a21\u578b\u6709\u65f6\u80fd\u5e2e\u52a9\u6211\u4eec\u4ece\u6570\u636e\u4e2d\u53d1\u73b0\u65b0\u7684\u3001\u672a\u77e5\u7684\u89c4\u5f8b\u548c\u77e5\u8bc6\u3002<\/li>\n<li>\u8ffd\u8d23\u4e0e\u5b89\u5168&#xff1a;\u5f53AI\u7cfb\u7edf\u72af\u9519\u65f6&#xff0c;\u6e05\u6670\u7684\u89e3\u91ca\u662f\u8fdb\u884c\u8d23\u4efb\u8ba4\u5b9a\u7684\u57fa\u7840&#xff0c;\u4e5f\u662f\u62b5\u5fa1\u6076\u610f\u653b\u51fb\u3001\u63d0\u5347\u7cfb\u7edf\u9c81\u68d2\u6027\u7684\u5173\u952e\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>15.3.2 XAI\u7684\u4e3b\u6d41\u65b9\u6cd5<\/h5>\n<p>\u6253\u5f00\u201c\u9ed1\u7bb1\u201d\u7684\u5c1d\u8bd5&#xff0c;\u4e3b\u8981\u6cbf\u7740\u4e24\u6761\u6280\u672f\u8def\u5f84\u5c55\u5f00\u3002<\/p>\n<ul>\n<li>\n<p>\u4e8b\u540e\u89e3\u91ca&#xff08;Post-hoc Explanations&#xff09; \u8fd9\u7c7b\u65b9\u6cd5\u5982\u540c\u201c\u4fa6\u63a2\u201d&#xff0c;\u5b83\u4eec\u5728\u4e0d\u6539\u53d8\u6a21\u578b\u672c\u8eab\u7ed3\u6784\u7684\u524d\u63d0\u4e0b&#xff0c;\u5bf9\u4e00\u4e2a\u5df2\u7ecf\u8bad\u7ec3\u597d\u7684\u201c\u9ed1\u7bb1\u201d\u6a21\u578b\u8fdb\u884c\u5916\u90e8\u63a2\u67e5\u548c\u5206\u6790\u3002<\/p>\n<ul>\n<li>\u7279\u5f81\u5f52\u56e0&#xff08;Feature Attribution&#xff09;&#xff1a;\u8fd9\u7c7b\u65b9\u6cd5\u65e8\u5728\u56de\u7b54\u201c\u54ea\u4e9b\u8f93\u5165\u7279\u5f81\u5bf9\u8fd9\u4e2a\u9884\u6d4b\u7ed3\u679c\u6700\u91cd\u8981&#xff1f;\u201d\n<ul>\n<li>LIME (Local Interpretable Model-agnostic Explanations)&#xff1a;\u5b83\u7684\u601d\u60f3\u975e\u5e38\u5de7\u5999\u3002\u5b83\u8ba4\u4e3a&#xff0c;\u867d\u7136\u4e00\u4e2a\u590d\u6742\u7684\u6a21\u578b\u5728\u5168\u5c40\u4e0a\u96be\u4ee5\u7406\u89e3&#xff0c;\u4f46\u5728\u4efb\u4f55\u4e00\u4e2a\u5177\u4f53\u7684\u9884\u6d4b\u70b9\u9644\u8fd1&#xff0c;\u6211\u4eec\u603b\u53ef\u4ee5\u7528\u4e00\u4e2a\u7b80\u5355\u7684\u3001\u53ef\u89e3\u91ca\u7684\u201c\u4ee3\u7406\u6a21\u578b\u201d&#xff08;\u5982\u7ebf\u6027\u6a21\u578b&#xff09;\u6765\u8fd1\u4f3c\u5b83\u7684\u5c40\u90e8\u884c\u4e3a\u3002\u901a\u8fc7\u89e3\u91ca\u8fd9\u4e2a\u7b80\u5355\u7684\u4ee3\u7406\u6a21\u578b&#xff0c;\u6765\u7406\u89e3\u590d\u6742\u6a21\u578b\u5728\u8fd9\u4e00\u7279\u5b9a\u70b9\u7684\u51b3\u7b56\u3002<\/li>\n<li>SHAP (SHapley Additive exPlanations)&#xff1a;\u5b83\u501f\u9274\u4e86\u5408\u4f5c\u535a\u5f08\u8bba\u4e2d\u7684\u201c\u590f\u666e\u5229\u503c\u201d\u601d\u60f3&#xff0c;\u5c06\u201c\u9884\u6d4b\u7ed3\u679c\u201d\u770b\u4f5c\u662f\u6240\u6709\u8f93\u5165\u7279\u5f81\u201c\u5408\u4f5c\u201d\u4ea7\u751f\u7684\u603b\u6536\u76ca\u3002SHAP\u80fd\u591f\u516c\u5e73\u5730\u3001\u6709\u7406\u8bba\u4fdd\u8bc1\u5730\u8ba1\u7b97\u51fa\u6bcf\u4e2a\u7279\u5f81\u5bf9\u6700\u7ec8\u9884\u6d4b\u7ed3\u679c\u7684\u8d21\u732e\u503c\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u57fa\u4e8e\u68af\u5ea6\u7684\u53ef\u89c6\u5316&#xff1a;\u8fd9\u7c7b\u65b9\u6cd5\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u5c24\u4e3a\u5e38\u7528&#xff0c;\u65e8\u5728\u56de\u7b54\u201c\u6a21\u578b\u5728\u505a\u51b3\u7b56\u65f6&#xff0c;\u5230\u5e95\u5728\u770b\u56fe\u50cf\u7684\u54ea\u4e2a\u90e8\u5206&#xff1f;\u201d\n<ul>\n<li>\u663e\u8457\u56fe&#xff08;Saliency Maps&#xff09;&#xff1a;\u901a\u8fc7\u8ba1\u7b97\u6a21\u578b\u8f93\u51fa\u76f8\u5bf9\u4e8e\u8f93\u5165\u56fe\u50cf\u7684\u68af\u5ea6&#xff0c;\u53ef\u4ee5\u5f97\u5230\u4e00\u5f20\u201c\u70ed\u529b\u56fe\u201d&#xff0c;\u56fe\u4e2d\u4eae\u5ea6\u8d8a\u9ad8\u7684\u533a\u57df&#xff0c;\u8868\u793a\u5bf9\u6a21\u578b\u51b3\u7b56\u5f71\u54cd\u8d8a\u5927\u7684\u50cf\u7d20\u3002<\/li>\n<li>Grad-CAM (Gradient-weighted Class Activation Mapping)&#xff1a;\u5b83\u662f\u4e00\u79cd\u66f4\u5148\u8fdb\u7684\u6280\u672f&#xff0c;\u80fd\u591f\u751f\u6210\u66f4\u6e05\u6670\u3001\u66f4\u805a\u7126\u4e8e\u7269\u4f53\u7684\u70ed\u529b\u56fe&#xff0c;\u76f4\u89c2\u5730\u5c55\u793a\u51faCNN\u505a\u51fa\u5206\u7c7b\u51b3\u7b56\u65f6\u6240\u4f9d\u8d56\u7684\u5173\u952e\u533a\u57df\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u672c\u8d28\u53ef\u89e3\u91ca\u6a21\u578b&#xff08;Inherently Interpretable Models&#xff09; \u53e6\u4e00\u6761\u8def\u5f84&#xff0c;\u5219\u662f\u201c\u5efa\u7b51\u5e08\u201d\u7684\u601d\u8def&#xff1a;\u6211\u4eec\u80fd\u5426\u76f4\u63a5\u8bbe\u8ba1\u548c\u4f7f\u7528\u4e00\u4e9b\u672c\u8eab\u7ed3\u6784\u5c31\u76f8\u5bf9\u900f\u660e\u3001\u51b3\u7b56\u8fc7\u7a0b\u4e00\u76ee\u4e86\u7136\u7684\u6a21\u578b&#xff1f;\u4f8b\u5982&#xff0c;\u4f20\u7edf\u7684\u51b3\u7b56\u6811\u3001\u7ebf\u6027\u56de\u5f52\u6a21\u578b\u3002\u5728\u6df1\u5ea6\u5b66\u4e60\u9886\u57df&#xff0c;\u8fd9\u610f\u5473\u7740\u6709\u9009\u62e9\u5730\u4f7f\u7528\u90a3\u4e9b\u5177\u6709\u5185\u5728\u53ef\u89e3\u91ca\u6027\u7684\u7ed3\u6784&#xff0c;\u6bd4\u5982\u5c06\u6ce8\u610f\u529b\u673a\u5236\u663e\u5f0f\u5730\u5e94\u7528\u5e76\u5c06\u5176\u6743\u91cd\u4f5c\u4e3a\u89e3\u91ca\u4f9d\u636e\u3002\u7136\u800c&#xff0c;\u8fd9\u901a\u5e38\u9700\u8981\u5728\u201c\u6a21\u578b\u6027\u80fd\u201d\u4e0e\u201c\u53ef\u89e3\u91ca\u6027\u201d\u4e4b\u95f4\u505a\u51fa\u6743\u8861\u3002<\/p>\n<\/li>\n<\/ul>\n<h5>15.3.3 \u4ece\u201c\u89e3\u91ca\u201d\u5230\u201c\u8d1f\u8d23\u201d<\/h5>\n<p>XAI\u7684\u7ec8\u6781\u76ee\u6807&#xff0c;\u5e76\u975e\u4ec5\u4ec5\u662f\u4e3a\u6bcf\u4e00\u4e2a\u51b3\u7b56\u9644\u4e0a\u4e00\u4efd\u201c\u8bf4\u660e\u4e66\u201d\u3002\u5b83\u662f\u901a\u5f80\u6784\u5efa\u66f4\u516c\u5e73\u3001\u66f4\u7a33\u5065\u3001\u66f4\u503c\u5f97\u4fe1\u8d56\u7684AI\u7cfb\u7edf\u7684\u5fc5\u7ecf\u4e4b\u8def\u3002\u53ef\u89e3\u91ca\u6027&#xff0c;\u662f\u5b9e\u73b0\u8d1f\u8d23\u4efb\u7684AI&#xff08;Responsible AI&#xff09;\u7684\u6280\u672f\u57fa\u77f3\u3002<\/p>\n<h4>15.4 \u591a\u6a21\u6001\u5b66\u4e60&#xff1a;1 &#043; 1 &gt; 2 \u7684\u878d\u5408\u827a\u672f<\/h4>\n<p>\u4eba\u7c7b\u7684\u611f\u77e5\u7cfb\u7edf&#xff0c;\u662f\u4e00\u4e2a\u5929\u751f\u7684\u591a\u6a21\u6001\u878d\u5408\u5927\u5e08\u3002\u6211\u4eec\u542c\u58f0\u8fa8\u4f4d&#xff0c;\u5bdf\u8a00\u89c2\u8272&#xff0c;\u5c06\u56fe\u50cf\u3001\u58f0\u97f3\u3001\u8bed\u8a00\u7b49\u591a\u79cd\u4fe1\u606f\u65e0\u7f1d\u5730\u878d\u5408\u5728\u4e00\u8d77&#xff0c;\u5f62\u6210\u5bf9\u4e16\u754c\u5b8c\u6574\u800c\u7acb\u4f53\u7684\u8ba4\u77e5\u3002\u800c\u591a\u6a21\u6001\u5b66\u4e60&#xff0c;\u6b63\u662f\u8981\u6559\u4f1aAI\u8fd9\u9879\u201c1&#043;1&gt;2\u201d\u7684\u878d\u5408\u827a\u672f\u3002<\/p>\n<h5>15.4.1 \u4e16\u754c\u662f\u591a\u6a21\u6001\u7684<\/h5>\n<ul>\n<li>\u4eba\u7c7b\u7684\u611f\u77e5&#xff1a;\u5f53\u60a8\u89c2\u770b\u4e00\u573a\u7535\u5f71\u65f6&#xff0c;\u60a8\u540c\u65f6\u5728\u5904\u7406\u6f14\u5458\u7684\u89c6\u89c9\u8868\u60c5\u3001\u542c\u89c9\u7684\u8bed\u8c03\u548c\u80cc\u666f\u97f3\u4e50&#xff0c;\u4ee5\u53ca\u8bed\u8a00\u7684\u53f0\u8bcd\u5185\u5bb9\u3002\u8fd9\u4e09\u79cd\u6a21\u6001\u7684\u4fe1\u606f\u76f8\u4e92\u8865\u5145\u3001\u76f8\u4e92\u5370\u8bc1&#xff0c;\u5171\u540c\u6784\u6210\u4e86\u60a8\u5b8c\u6574\u7684\u60c5\u611f\u4f53\u9a8c\u3002<\/li>\n<li>AI\u7684\u6311\u6218&#xff1a;\u4f20\u7edf\u7684AI\u6a21\u578b\u901a\u5e38\u662f\u5355\u6a21\u6001\u7684&#xff0c;\u8981\u4e48\u5904\u7406\u56fe\u50cf&#xff0c;\u8981\u4e48\u5904\u7406\u6587\u672c\u3002\u591a\u6a21\u6001\u5b66\u4e60&#xff08;Multimodal Learning&#xff09;\u7684\u76ee\u6807&#xff0c;\u5c31\u662f\u8ba9AI\u6a21\u578b\u80fd\u591f\u50cf\u4eba\u4e00\u6837&#xff0c;\u540c\u65f6\u5904\u7406\u548c\u7406\u89e3\u6765\u81ea\u4e0d\u540c\u6a21\u6001&#xff08;Modality&#xff09;\u7684\u6570\u636e&#xff0c;\u5e76\u7406\u89e3\u5b83\u4eec\u4e4b\u95f4\u7684\u590d\u6742\u5173\u7cfb\u3002<\/li>\n<\/ul>\n<h5>15.4.2 \u591a\u6a21\u6001\u5b66\u4e60\u7684\u6838\u5fc3\u4efb\u52a1\u4e0e\u6280\u672f<\/h5>\n<p>\u591a\u6a21\u6001\u5b66\u4e60\u7684\u7814\u7a76&#xff0c;\u56f4\u7ed5\u7740\u51e0\u4e2a\u6838\u5fc3\u95ee\u9898\u5c55\u5f00\u3002<\/p>\n<ul>\n<li>\u878d\u5408&#xff08;Fusion&#xff09;&#xff1a;\u8fd9\u662f\u591a\u6a21\u6001\u5b66\u4e60\u6700\u6838\u5fc3\u7684\u6280\u672f\u6311\u6218\u3002\u5373&#xff0c;\u5982\u4f55\u5728\u6a21\u578b\u7684\u4e0d\u540c\u9636\u6bb5&#xff0c;\u5c06\u6765\u81ea\u4e0d\u540c\u6a21\u6001\u7684\u7279\u5f81\u8868\u793a\u6709\u6548\u5730\u7ed3\u5408\u8d77\u6765\u3002\n<ul>\n<li>\u65e9\u671f\u878d\u5408&#xff1a;\u5728\u8f93\u5165\u5c42\u5c31\u5c06\u4e0d\u540c\u6a21\u6001\u7684\u6570\u636e\u62fc\u63a5\u8d77\u6765\u3002<\/li>\n<li>\u665a\u671f\u878d\u5408&#xff1a;\u4e3a\u6bcf\u4e2a\u6a21\u6001\u5355\u72ec\u8bad\u7ec3\u4e00\u4e2a\u6a21\u578b&#xff0c;\u5728\u6700\u540e\u51b3\u7b56\u5c42\u518d\u5c06\u5176\u7ed3\u679c\u878d\u5408\u3002<\/li>\n<li>\u4e2d\u671f\u878d\u5408&#xff08;\u6df7\u5408\u878d\u5408&#xff09;&#xff1a;\u8fd9\u662f\u5f53\u524d\u7684\u4e3b\u6d41&#xff0c;\u901a\u8fc7\u590d\u6742\u7684\u4ea4\u4e92\u5c42&#xff08;\u5982\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236&#xff09;\u5728\u6a21\u578b\u7684\u4e2d\u95f4\u5c42\u6b21\u8fdb\u884c\u6df1\u5ea6\u3001\u975e\u7ebf\u6027\u7684\u4fe1\u606f\u878d\u5408\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u5bf9\u9f50&#xff08;Alignment&#xff09;&#xff1a;\u5728\u4e0d\u540c\u6a21\u6001\u7684\u6570\u636e\u4e4b\u95f4&#xff0c;\u663e\u5f0f\u5730\u5efa\u7acb\u8d77\u7ec6\u7c92\u5ea6\u7684\u8054\u7cfb\u3002\u4f8b\u5982&#xff0c;\u5728\u4e00\u4efd\u70f9\u996a\u6559\u7a0b\u89c6\u9891\u4e2d&#xff0c;\u5c06\u8bed\u97f3\u6307\u4ee4\u201c\u5207\u788e\u6d0b\u8471\u201d&#xff0c;\u4e0e\u89c6\u9891\u753b\u9762\u4e2d\u6b63\u5728\u5207\u6d0b\u8471\u7684\u90a3\u4e2a\u7247\u6bb5\u7cbe\u786e\u5730\u5bf9\u5e94\u8d77\u6765\u3002<\/li>\n<li>\u8de8\u6a21\u6001\u751f\u6210&#xff08;Translation\/Generation&#xff09;&#xff1a;\u8fd9\u662f\u591a\u6a21\u6001\u5b66\u4e60\u6700\u4ee4\u4eba\u60ca\u8273\u7684\u5e94\u7528\u65b9\u5411&#xff0c;\u5373\u4ece\u4e00\u79cd\u6a21\u6001\u751f\u6210\u53e6\u4e00\u79cd\u6a21\u6001\u7684\u5185\u5bb9\u3002\n<ul>\n<li>\u770b\u56fe\u8bf4\u8bdd&#xff08;Image Captioning&#xff09;&#xff1a;\u8f93\u5165\u4e00\u5f20\u56fe\u7247&#xff0c;\u6a21\u578b\u751f\u6210\u4e00\u6bb5\u63cf\u8ff0\u8be5\u56fe\u7247\u7684\u6587\u5b57\u3002<\/li>\n<li>\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210&#xff08;Text-to-Image Generation&#xff09;&#xff1a;\u8fd9\u5728\u8fd1\u5e74\u6765\u53d6\u5f97\u4e86\u7a81\u7834\u6027\u8fdb\u5c55\u3002\u6a21\u578b&#xff08;\u5982DALL-E, Midjourney, Stable Diffusion&#xff09;\u53ef\u4ee5\u6839\u636e\u4e00\u6bb5\u5929\u9a6c\u884c\u7a7a\u7684\u6587\u5b57\u63cf\u8ff0&#xff0c;\u751f\u6210\u4e00\u5f20\u9ad8\u8d28\u91cf\u3001\u5bcc\u6709\u60f3\u8c61\u529b\u7684\u56fe\u50cf\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5>15.4.3 \u5e94\u7528\u4e0e\u672a\u6765<\/h5>\n<ul>\n<li>\u5e94\u7528&#xff1a;\n<ul>\n<li>\u591a\u6a21\u6001\u60c5\u611f\u5206\u6790&#xff1a;\u7ed3\u5408\u7528\u6237\u7684\u9762\u90e8\u8868\u60c5\u3001\u8bed\u97f3\u8bed\u8c03\u548c\u8bc4\u8bba\u6587\u672c&#xff0c;\u505a\u51fa\u66f4\u7cbe\u51c6\u7684\u60c5\u611f\u5224\u65ad\u3002<\/li>\n<li>\u81ea\u52a8\u9a7e\u9a76&#xff1a;\u878d\u5408\u6444\u50cf\u5934&#xff08;\u89c6\u89c9&#xff09;\u3001\u6fc0\u5149\u96f7\u8fbe&#xff08;LiDAR&#xff0c;3D\u70b9\u4e91&#xff09;\u3001\u6beb\u7c73\u6ce2\u96f7\u8fbe\u7b49\u591a\u79cd\u4f20\u611f\u5668\u7684\u6570\u636e&#xff0c;\u505a\u51fa\u66f4\u5b89\u5168\u3001\u66f4\u9c81\u68d2\u7684\u9a7e\u9a76\u51b3\u7b56\u3002<\/li>\n<li>\u66f4\u5f3a\u5927\u7684\u667a\u80fd\u641c\u7d22&#xff1a;\u5b9e\u73b0\u201c\u4ee5\u56fe\u641c\u89c6\u9891\u201d\u3001\u201c\u7528\u6587\u5b57\u63cf\u8ff0\u641c\u7d22\u4e00\u6bb5\u97f3\u9891\u201d\u7b49\u66f4\u81ea\u7136\u3001\u66f4\u5f3a\u5927\u7684\u4ea4\u4e92\u65b9\u5f0f\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u672a\u6765&#xff1a;\u6784\u5efa\u4e00\u4e2a\u80fd\u591f\u7406\u89e3\u548c\u751f\u6210\u4efb\u610f\u6a21\u6001\u4fe1\u606f\u7684\u3001\u7edf\u4e00\u7684**\u201c\u4e16\u754c\u6a21\u578b&#xff08;World Model&#xff09;\u201d**&#xff0c;\u662f\u8bb8\u591a\u9876\u5c16AI\u5b9e\u9a8c\u5ba4\u7684\u7ec8\u6781\u76ee\u6807\u4e4b\u4e00\u3002\u8fd9\u6837\u7684\u6a21\u578b&#xff0c;\u5c06\u80fd\u66f4\u5168\u9762\u3001\u66f4\u6df1\u5165\u5730\u7406\u89e3\u6211\u4eec\u8fd9\u4e2a\u590d\u6742\u591a\u5f69\u7684\u4e16\u754c\u3002<\/li>\n<\/ul>\n<h4>15.5 AI\u4f26\u7406\u4e0e\u793e\u4f1a\u8d23\u4efb&#xff1a;\u4f5c\u4e3a\u6280\u672f\u521b\u9020\u8005\u7684\u201c\u7b2c\u4e00\u6027\u539f\u7406\u201d<\/h4>\n<p>\u6211\u4eec\u7ec8\u4e8e\u6765\u5230\u4e86\u672c\u4e66\u7684\u6700\u540e\u4e00\u8282&#xff0c;\u4e5f\u662f\u6700\u91cd\u8981\u7684\u4e00\u8282\u3002\u5b83\u4e0d\u6d89\u53ca\u4efb\u4f55\u5177\u4f53\u7684\u7b97\u6cd5\u6216\u4ee3\u7801&#xff0c;\u4f46\u5b83\u5173\u4e4e\u6211\u4eec\u672a\u6765\u6240\u5199\u7684\u6bcf\u4e00\u884c\u4ee3\u7801\u7684\u610f\u4e49\u548c\u65b9\u5411\u3002\u8fd9&#xff0c;\u5c31\u662fAI\u4f26\u7406\u4e0e\u6211\u4eec\u4f5c\u4e3a\u6280\u672f\u521b\u9020\u8005&#xff0c;\u80a9\u4e0a\u6240\u5fc5\u987b\u627f\u8f7d\u7684\u793e\u4f1a\u8d23\u4efb\u3002<\/p>\n<h5>15.5.1 \u6280\u672f\u7684\u4e2d\u7acb\u4e0e\u4eba\u7684\u8d23\u4efb<\/h5>\n<p>\u53e4\u5e0c\u814a\u795e\u8bdd\u4e2d&#xff0c;\u666e\u7f57\u7c73\u4fee\u65af\u5c06\u706b\u79cd\u5e26\u5230\u4eba\u95f4&#xff0c;\u706b\u65e2\u80fd\u5e26\u6765\u6e29\u6696\u4e0e\u5149\u660e&#xff0c;\u4e5f\u80fd\u71c3\u8d77\u711a\u6bc1\u4e00\u5207\u7684\u70c8\u7130\u3002AI&#xff0c;\u6b63\u662f\u6211\u4eec\u8fd9\u4e2a\u65f6\u4ee3\u7684\u201c\u666e\u7f57\u7c73\u4fee\u65af\u4e4b\u706b\u201d\u3002\u5b83\u662f\u4e00\u9879\u65e0\u6bd4\u5f3a\u5927\u7684\u8d4b\u80fd\u6280\u672f&#xff0c;\u4f46\u8fd9\u4efd\u529b\u91cf\u540c\u6837\u53ef\u80fd\u88ab\u8bef\u7528&#xff0c;\u6216\u5e26\u6765\u6211\u4eec\u59cb\u6599\u672a\u53ca\u7684\u8d1f\u9762\u793e\u4f1a\u540e\u679c\u3002<\/p>\n<p>\u4eba\u4eec\u5e38\u8bf4\u201c\u6280\u672f\u662f\u4e2d\u7acb\u7684\u201d\u3002\u8fd9\u53e5\u8bdd\u5728\u67d0\u79cd\u7a0b\u5ea6\u4e0a\u662f\u5bf9\u7684&#xff0c;\u4f46\u5b83\u6781\u6613\u4ea7\u751f\u8bef\u5bfc\u3002\u56e0\u4e3a\u6280\u672f\u662f\u7531\u4eba\u521b\u9020\u3001\u7531\u4eba\u8bbe\u8ba1\u3001\u7531\u4eba\u90e8\u7f72\u3001\u7531\u4eba\u4f7f\u7528\u7684\u3002\u5728\u6574\u4e2a\u94fe\u6761\u4e2d&#xff0c;\u4eba\u7684\u4ef7\u503c\u89c2\u3001\u610f\u56fe\u548c\u9009\u62e9&#xff0c;\u65e0\u65f6\u65e0\u523b\u4e0d\u5728\u5851\u9020\u7740\u6280\u672f\u7684\u5f62\u6001\u548c\u5f71\u54cd\u3002\u56e0\u6b64&#xff0c;\u4f5c\u4e3a\u6280\u672f\u7684\u521b\u9020\u8005&#xff0c;\u6211\u4eec\u7edd\u4e0d\u80fd\u4ee5\u201c\u6280\u672f\u4e2d\u7acb\u201d\u4e3a\u501f\u53e3&#xff0c;\u6765\u9003\u907f\u6211\u4eec\u5e94\u5c3d\u7684\u4f26\u7406\u8d23\u4efb\u3002<\/p>\n<h5>15.5.2 \u5fc5\u987b\u6b63\u89c6\u7684\u6838\u5fc3\u4f26\u7406\u8bae\u9898<\/h5>\n<p>\u5728\u5f00\u53d1\u548c\u5e94\u7528AI\u6280\u672f\u7684\u8fc7\u7a0b\u4e2d&#xff0c;\u6211\u4eec\u5fc5\u987b\u65f6\u523b\u8b66\u60d5\u5e76\u5ba1\u614e\u601d\u8003\u4ee5\u4e0b\u51e0\u4e2a\u6838\u5fc3\u8bae\u9898&#xff1a;<\/p>\n<ul>\n<li>\u504f\u89c1\u4e0e\u516c\u5e73&#xff08;Bias and Fairness&#xff09;&#xff1a;AI\u6a21\u578b\u662f\u4ece\u6570\u636e\u4e2d\u5b66\u4e60\u7684\u3002\u5982\u679c\u6211\u4eec\u7684\u8bad\u7ec3\u6570\u636e\u672c\u8eab\u5c31\u5305\u542b\u4e86\u73b0\u5b9e\u4e16\u754c\u4e2d\u5b58\u5728\u7684\u793e\u4f1a\u504f\u89c1&#xff08;\u5982\u6027\u522b\u6b67\u89c6\u3001\u79cd\u65cf\u504f\u89c1&#xff09;&#xff0c;\u90a3\u4e48\u6a21\u578b\u4e0d\u4ec5\u4f1a\u5fe0\u5b9e\u5730\u201c\u5b66\u4f1a\u201d\u8fd9\u4e9b\u504f\u89c1&#xff0c;\u751a\u81f3\u4f1a\u5c06\u5176\u653e\u5927\u3002\u8fd9\u4f1a\u5bfc\u81f4AI\u7cfb\u7edf\u5bf9\u7279\u5b9a\u4eba\u7fa4\u505a\u51fa\u4e0d\u516c\u5e73\u7684\u3001\u6b67\u89c6\u6027\u7684\u51b3\u7b56\u3002\u5982\u4f55\u5ea6\u91cf\u3001\u53d1\u73b0\u548c\u7f13\u89e3\u6a21\u578b\u504f\u89c1&#xff0c;\u662f\u786e\u4fddAI\u516c\u5e73\u6027\u7684\u6838\u5fc3\u6311\u6218\u3002<\/li>\n<li>\u9690\u79c1&#xff08;Privacy&#xff09;&#xff1a;\u6211\u4eec\u5728\u8054\u90a6\u5b66\u4e60\u4e2d\u5df2\u7ecf\u6df1\u5165\u63a2\u8ba8\u4e86\u8fd9\u4e00\u70b9\u3002\u5982\u4f55\u5728\u5145\u5206\u5229\u7528\u6570\u636e\u4ef7\u503c\u7684\u540c\u65f6&#xff0c;\u4ee5\u6700\u4e25\u683c\u7684\u6807\u51c6\u4fdd\u62a4\u7528\u6237\u7684\u4e2a\u4eba\u9690\u79c1&#xff0c;\u662f\u6240\u6709AI\u5e94\u7528\u5fc5\u987b\u9075\u5b88\u7684\u5e95\u7ebf\u3002<\/li>\n<li>\u900f\u660e\u5ea6\u4e0e\u95ee\u8d23\u5236&#xff08;Transparency and Accountability&#xff09;&#xff1a;\u5f53\u4e00\u4e2aAI\u7cfb\u7edf&#xff08;\u5982\u81ea\u52a8\u9a7e\u9a76\u6c7d\u8f66&#xff09;\u72af\u4e0b\u9519\u8bef\u5e76\u9020\u6210\u635f\u5931\u65f6&#xff0c;\u8c01\u5e94\u8be5\u6765\u8d1f\u8d23&#xff1f;\u662f\u7528\u6237\u3001\u662f\u5f00\u53d1\u8005\u3001\u662f\u5236\u9020\u5546&#xff0c;\u8fd8\u662f\u7b97\u6cd5\u672c\u8eab&#xff1f;\u8981\u56de\u7b54\u8fd9\u4e2a\u95ee\u9898&#xff0c;\u5c31\u9700\u8981\u6e05\u6670\u7684\u95ee\u8d23\u673a\u5236\u548c\u5fc5\u8981\u7684\u6280\u672f\u900f\u660e\u5ea6&#xff08;XAI\u662f\u5176\u4e2d\u7684\u4e00\u73af&#xff09;&#xff0c;\u4ee5\u786e\u4fdd\u51b3\u7b56\u94fe\u6761\u7684\u53ef\u8ffd\u6eaf\u6027\u3002<\/li>\n<li>\u5b89\u5168\u4e0e\u9c81\u68d2\u6027&#xff08;Safety and Robustness&#xff09;&#xff1a;\u7814\u7a76\u8868\u660e&#xff0c;\u8bb8\u591a\u5148\u8fdb\u7684AI\u6a21\u578b\u90fd\u975e\u5e38\u201c\u8106\u5f31\u201d&#xff0c;\u5b83\u4eec\u53ef\u80fd\u4f1a\u88ab\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u3001\u4eba\u773c\u96be\u4ee5\u5bdf\u89c9\u7684\u5fae\u5c0f\u6270\u52a8&#xff08;\u5373\u5bf9\u6297\u6027\u653b\u51fb&#xff09;\u8f7b\u6613\u5730\u6b3a\u9a97&#xff0c;\u505a\u51fa\u5b8c\u5168\u9519\u8bef\u7684\u5224\u65ad\u3002\u5982\u4f55\u6784\u5efa\u5728\u6076\u610f\u653b\u51fb\u548c\u672a\u77e5\u73af\u5883\u4e0b\u4f9d\u7136\u7a33\u5065\u3001\u53ef\u9760\u7684AI\u7cfb\u7edf&#xff0c;\u662f\u5b89\u5168\u9886\u57df\u7684\u91cd\u5927\u6311\u6218\u3002<\/li>\n<li>\u5bf9\u5c31\u4e1a\u4e0e\u793e\u4f1a\u7ed3\u6784\u7684\u5f71\u54cd&#xff1a;AI\u81ea\u52a8\u5316\u6d6a\u6f6e\u5bf9\u672a\u6765\u5de5\u4f5c\u5c97\u4f4d\u3001\u6280\u80fd\u9700\u6c42\u3001\u8d22\u5bcc\u5206\u914d\u4e43\u81f3\u6574\u4e2a\u793e\u4f1a\u7ed3\u6784\u7684\u6df1\u8fdc\u5f71\u54cd&#xff0c;\u662f\u4e00\u4e2a\u65e0\u6cd5\u56de\u907f\u7684\u5b8f\u5927\u8bae\u9898\u3002\u8fd9\u9700\u8981\u6280\u672f\u4e13\u5bb6\u3001\u7ecf\u6d4e\u5b66\u5bb6\u3001\u793e\u4f1a\u5b66\u5bb6\u548c\u653f\u7b56\u5236\u5b9a\u8005\u5171\u540c\u8fdb\u884c\u8de8\u5b66\u79d1\u7684\u3001\u6709\u8fdc\u89c1\u7684\u5ba1\u614e\u601d\u8003\u548c\u79ef\u6781\u5f15\u5bfc\u3002<\/li>\n<\/ul>\n<h5>15.5.3 \u6211\u4eec\u7684\u8a93\u8a00&#xff1a;\u8fc8\u5411\u8d1f\u8d23\u4efb\u7684\u521b\u65b0<\/h5>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u4f5c\u4e3a\u5373\u5c06\u6216\u5df2\u7ecf\u6295\u8eab\u4e8eAI\u6d6a\u6f6e\u7684\u65b0\u4e00\u4ee3\u6280\u672f\u521b\u9020\u8005&#xff0c;\u6211\u4eec\u624b\u4e2d\u638c\u63e1\u7740\u5851\u9020\u672a\u6765\u7684\u5f3a\u5927\u529b\u91cf\u3002\u8fd9\u4efd\u529b\u91cf&#xff0c;\u8d4b\u4e88\u6211\u4eec\u7684\u4e0d\u4ec5\u662f\u673a\u9047&#xff0c;\u66f4\u662f\u6c89\u7538\u7538\u7684\u8d23\u4efb\u3002<\/p>\n<p>\u6211\u4eec\u5fc5\u987b\u5c06\u4f26\u7406\u601d\u8003&#xff0c;\u4f5c\u4e3a\u6211\u4eec\u6280\u672f\u5f00\u53d1\u7684**\u201c\u7b2c\u4e00\u6027\u539f\u7406\u201d**\u3002\u5b83\u4e0d\u5e94\u8be5\u662f\u5728\u9879\u76ee\u540e\u671f\u624d\u88ab\u60f3\u8d77\u7684\u201c\u68c0\u67e5\u9879\u201d&#xff0c;\u800c\u5e94\u662f\u8d2f\u7a7f\u4e8e\u4ece\u95ee\u9898\u5b9a\u4e49\u3001\u6570\u636e\u6536\u96c6\u3001\u6a21\u578b\u8bbe\u8ba1\u3001\u5230\u90e8\u7f72\u5e94\u7528\u5168\u8fc7\u7a0b\u7684\u3001\u5185\u5316\u4e8e\u5fc3\u3001\u5916\u5316\u4e8e\u884c\u7684\u57fa\u672c\u51c6\u5219\u3002<\/p>\n<p>\u5728\u8ffd\u6c42\u66f4\u9ad8\u7cbe\u5ea6\u3001\u66f4\u5feb\u901f\u5ea6\u3001\u66f4\u5f3a\u6027\u80fd\u7684\u540c\u65f6&#xff0c;\u8bf7\u8ba9\u6211\u4eec\u65f6\u523b\u5fc3\u6000\u5bf9\u4eba\u3001\u5bf9\u793e\u4f1a\u3001\u5bf9\u672a\u6765\u7684\u656c\u754f\u4e0e\u8d23\u4efb\u3002\u8ba9\u6211\u4eec\u5171\u540c\u81f4\u529b\u4e8e\u6784\u5efa\u90a3\u4e9b\u4e0d\u4ec5\u201c\u667a\u80fd\u201d&#xff0c;\u800c\u4e14\u201c\u667a\u6167\u201d&#xff1b;\u4e0d\u4ec5\u201c\u5f3a\u5927\u201d&#xff0c;\u800c\u4e14\u201c\u5584\u826f\u201d\u7684AI\u7cfb\u7edf\u3002<\/p>\n<hr \/>\n<p>\u7ec8\u7ae0&#xff1a;\u667a\u6167\u65e0\u57a0&#xff0c;\u63a2\u7d22\u4e0d\u606f<\/p>\n<p>\u63a9\u5377\u6c89\u601d&#xff0c;\u6211\u4eec\u5171\u540c\u7684\u65c5\u7a0b\u81f3\u6b64\u5df2\u8fd1\u7ec8\u70b9\u3002\u4ece\u7b2c\u4e00\u884c\u4ee3\u7801\u5230\u6700\u540e\u4e00\u7ae0\u7684\u4f26\u7406\u6c89\u601d&#xff0c;\u6211\u4eec\u4e00\u540c\u8d70\u8fc7\u4e86\u4e00\u6761\u7cfb\u7edf\u800c\u6df1\u5165\u7684\u6df1\u5ea6\u5b66\u4e60\u4e4b\u8def\u3002\u6211\u4eec\u4e0d\u4ec5\u5b66\u4e60\u4e86\u201c\u662f\u4ec0\u4e48\u201d\u548c\u201c\u600e\u4e48\u505a\u201d&#xff0c;\u66f4\u5728\u4e0d\u65ad\u63a2\u95ee\u201c\u4e3a\u4ec0\u4e48\u201d\u4ee5\u53ca\u201c\u5e94\u8be5\u505a\u4ec0\u4e48\u201d\u3002<\/p>\n<p>\u6211\u4eec\u638c\u63e1\u4e86\u7528Python\u548c\u4e3b\u6d41\u6846\u67b6\u6784\u5efa\u3001\u8bad\u7ec3\u548c\u90e8\u7f72\u5f3a\u5927AI\u6a21\u578b\u7684\u5168\u5957\u6280\u80fd\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u6211\u4eec\u5efa\u7acb\u8d77\u4e86\u4e00\u5957\u601d\u8003\u95ee\u9898\u7684\u6846\u67b6&#xff1a;\u9762\u5bf9\u4e00\u4e2a\u65b0\u95ee\u9898&#xff0c;\u6211\u4eec\u77e5\u9053\u5982\u4f55\u9009\u62e9\u5408\u9002\u7684\u6a21\u578b\u67b6\u6784&#xff0c;\u5982\u4f55\u51c6\u5907\u6570\u636e&#xff0c;\u5982\u4f55\u8bca\u65ad\u548c\u4f18\u5316\u8bad\u7ec3\u8fc7\u7a0b&#xff0c;\u4ee5\u53ca\u5982\u4f55\u4ee5\u5de5\u7a0b\u5316\u7684\u601d\u7ef4\u5c06\u5176\u8f6c\u5316\u4e3a\u73b0\u5b9e\u4ef7\u503c\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u77e5\u8bc6\u7684\u6d77\u6d0b\u6d69\u701a\u65e0\u57a0&#xff0c;\u672c\u4e66\u6240\u80fd\u5448\u73b0\u7684&#xff0c;\u7ec8\u7a76\u53ea\u662f\u5176\u4e2d\u7684\u4e00\u7247\u98ce\u666f\u3002\u5b83\u4e3a\u60a8\u63d0\u4f9b\u4e86\u4e00\u5f20\u8be6\u5c3d\u7684\u5730\u56fe\u3001\u4e00\u4e2a\u53ef\u9760\u7684\u6307\u5357\u9488\u548c\u4e00\u8eab\u8fdc\u822a\u7684\u672c\u9886&#xff0c;\u4f46\u771f\u6b63\u7684\u63a2\u7d22&#xff0c;\u4ece\u60a8\u5408\u4e0a\u672c\u4e66\u7684\u90a3\u4e00\u523b&#xff0c;\u624d\u521a\u521a\u5f00\u59cb\u3002<\/p>\n<p>AI\u7684\u6d6a\u6f6e\u4ecd\u5728\u5954\u6d8c\u5411\u524d&#xff0c;\u65b0\u7684\u8bba\u6587\u3001\u65b0\u7684\u4ee3\u7801\u3001\u65b0\u7684\u601d\u60f3\u6bcf\u5929\u90fd\u5728\u8bde\u751f\u3002\u8bf7\u4fdd\u6301\u8fd9\u4efd\u5728\u5b66\u4e60\u65c5\u7a0b\u4e2d\u57f9\u517b\u8d77\u6765\u7684\u597d\u5947\u5fc3\u4e0e\u6c42\u77e5\u6b32&#xff0c;\u8ba9\u5b83\u6210\u4e3a\u60a8\u672a\u6765\u4e0d\u65ad\u524d\u884c\u7684\u4e0d\u7aed\u52a8\u529b\u3002\u53bb\u9605\u8bfb\u6700\u65b0\u7684\u7814\u7a76&#xff0c;\u53bb\u590d\u73b0\u6709\u8da3\u7684\u9879\u76ee&#xff0c;\u53bb\u53c2\u4e0e\u5f00\u6e90\u7684\u793e\u533a&#xff0c;\u53bb\u601d\u8003\u90a3\u4e9b\u5c1a\u672a\u88ab\u89e3\u51b3\u7684\u96be\u9898\u3002<\/p>\n<p>\u6700\u91cd\u8981\u7684\u662f&#xff0c;\u5e26\u7740\u60a8\u5728\u672c\u4e66\u4e2d\u5b66\u5230\u7684\u6240\u6709\u77e5\u8bc6\u3001\u6280\u80fd\u4e0e\u601d\u8003&#xff0c;\u52c7\u6562\u5730\u53bb\u63a2\u7d22\u3001\u53bb\u521b\u9020&#xff0c;\u53bb\u89e3\u51b3\u90a3\u4e9b\u5bf9\u60a8\u3001\u5bf9\u60a8\u6240\u5728\u7684\u793e\u533a\u3001\u4e43\u81f3\u5bf9\u6574\u4e2a\u4e16\u754c\u6709\u610f\u4e49\u7684\u95ee\u9898\u3002<\/p>\n<p>\u667a\u6167\u65e0\u57a0&#xff0c;\u63a2\u7d22\u4e0d\u606f\u3002\u613f\u60a8\u5728\u4eba\u5de5\u667a\u80fd\u7684\u5e7f\u9614\u661f\u8fb0\u5927\u6d77\u4e2d&#xff0c;\u627e\u5230\u5c5e\u4e8e\u81ea\u5df1\u7684\u3001\u90a3\u9897\u6700\u95ea\u4eae\u7684\u661f\u3002<\/p>\n<p>\u518d\u4f1a&#xff0c;\u672a\u6765\u7684\u521b\u9020\u8005&#xff01;<\/p>\n<hr \/>\n<h3>\u9644\u5f55&#xff1a;\u884c\u8005\u7684\u201c\u5b9d\u5e93\u201d\u4e0e\u201c\u8def\u4e66\u201d<\/h3>\n<ul>\n<li>A&#xff1a;\u5e38\u7528\u6570\u636e\u96c6\u4e0e\u8d44\u6e90\u94fe\u63a5\u3002<\/li>\n<li>B&#xff1a;\u6570\u5b66\u7b26\u53f7\u8868\u3002<\/li>\n<li>C&#xff1a;\u5e38\u89c1\u95ee\u9898\u4e0e\u6392\u9519\u6307\u5357\u3002<\/li>\n<li>D&#xff1a;\u8fdb\u4e00\u6b65\u9605\u8bfb\u5efa\u8bae\u3002<\/li>\n<\/ul>\n<p>\u6574\u88c5\u518d\u51fa\u53d1<\/p>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005&#xff0c;\u5f53\u60a8\u62b5\u8fbe\u8fd9\u91cc&#xff0c;\u610f\u5473\u7740\u6211\u4eec\u5171\u540c\u7684\u6df1\u5ea6\u5b66\u4e60\u63a2\u7d22\u4e4b\u65c5\u5df2\u753b\u4e0a\u4e00\u4e2a\u5706\u6ee1\u7684\u53e5\u53f7\u3002\u7136\u800c&#xff0c;\u4efb\u4f55\u5b66\u4e60\u7684\u7ec8\u70b9&#xff0c;\u90fd\u5e94\u662f\u53e6\u4e00\u6bb5\u66f4\u5e7f\u9614\u5f81\u7a0b\u7684\u8d77\u70b9\u3002\u4e3a\u4e86\u8ba9\u60a8\u5728\u672a\u6765\u7684\u9053\u8def\u4e0a\u884c\u5f97\u66f4\u7a33\u3001\u8d70\u5f97\u66f4\u8fdc&#xff0c;\u6211\u4eec\u4e3a\u60a8\u7cbe\u5fc3\u51c6\u5907\u4e86\u8fd9\u4efd\u9644\u5f55\u3002<\/p>\n<p>\u8bf7\u4e0d\u8981\u5c06\u5b83\u770b\u4f5c\u662f\u77e5\u8bc6\u7684\u4f59\u70ec&#xff0c;\u800c\u5e94\u89c6\u5176\u4e3a\u4e00\u4e2a\u6574\u88c5\u5f85\u53d1\u7684\u201c\u519b\u68b0\u5e93\u201d\u4e0e\u201c\u7ed9\u517b\u7ad9\u201d\u3002\u5728\u8fd9\u91cc&#xff0c;\u6211\u4eec\u4e3a\u60a8\u5907\u4e0b\u4e86\u5b9e\u8df5\u6240\u9700\u7684\u201c\u7cae\u8349\u201d&#xff08;\u6570\u636e\u96c6&#xff09;&#xff0c;\u67e5\u9605\u6982\u5ff5\u7684\u201c\u5bc6\u7801\u672c\u201d&#xff08;\u6570\u5b66\u7b26\u53f7&#xff09;&#xff0c;\u5e94\u5bf9\u56f0\u96be\u7684\u201c\u6025\u6551\u5305\u201d&#xff08;FAQ&#xff09;&#xff0c;\u4ee5\u53ca\u6307\u5411\u65b0\u5927\u9646\u7684\u201c\u822a\u6d77\u56fe\u201d&#xff08;\u9605\u8bfb\u5efa\u8bae&#xff09;\u3002<\/p>\n<p>\u613f\u8fd9\u4efd\u201c\u884c\u56ca\u201d\u80fd\u4f34\u60a8\u5de6\u53f3&#xff0c;\u5728\u60a8\u63a2\u7d22\u4eba\u5de5\u667a\u80fd\u7684\u661f\u8fb0\u5927\u6d77\u65f6&#xff0c;\u4e3a\u60a8\u63d0\u4f9b\u6700\u575a\u5b9e\u7684\u652f\u6301\u4e0e\u6700\u53ca\u65f6\u7684\u5e2e\u52a9\u3002\u73b0\u5728&#xff0c;\u8ba9\u6211\u4eec\u4e00\u540c\u6e05\u70b9\u884c\u56ca&#xff0c;\u51c6\u5907\u8fce\u63a5\u65b0\u7684\u6311\u6218\u3002<\/p>\n<hr \/>\n<h4>A&#xff1a;\u5e38\u7528\u6570\u636e\u96c6\u4e0e\u8d44\u6e90\u94fe\u63a5<\/h4>\n<p>\u201c\u7eb8\u4e0a\u5f97\u6765\u7ec8\u89c9\u6d45&#xff0c;\u7edd\u77e5\u6b64\u4e8b\u8981\u8eac\u884c\u3002\u201d \u6df1\u5ea6\u5b66\u4e60\u662f\u4e00\u95e8\u9ad8\u5ea6\u4f9d\u8d56\u5b9e\u8df5\u7684\u5b66\u79d1&#xff0c;\u800c\u9ad8\u8d28\u91cf\u7684\u6570\u636e\u96c6&#xff0c;\u6b63\u662f\u5b9e\u8df5\u7684\u201c\u7cae\u8349\u201d\u3002\u4ee5\u4e0b\u6211\u4eec\u6574\u7406\u4e86\u4e00\u4e9b\u5728\u5b66\u672f\u754c\u548c\u5de5\u4e1a\u754c\u6700\u5e38\u7528\u3001\u6700\u7ecf\u5178\u7684\u516c\u5171\u6570\u636e\u96c6&#xff0c;\u4ee5\u53ca\u4e00\u7cfb\u5217\u5b9d\u8d35\u7684\u5b66\u4e60\u8d44\u6e90&#xff0c;\u5e0c\u671b\u80fd\u4e3a\u60a8\u7684\u63a2\u7d22\u4e4b\u65c5\u63d0\u4f9b\u5145\u8db3\u7684\u52a8\u529b\u3002<\/p>\n<p>A.1 \u7ecf\u5178\u6570\u636e\u96c6<\/p>\n<p>\u8ba1\u7b97\u673a\u89c6\u89c9&#xff08;CV&#xff09;<\/p>\n<ul>\n<li>MNIST \/ Fashion-MNIST&#xff1a;\u624b\u5199\u6570\u5b57\u548c\u670d\u88c5\u7c7b\u522b\u7684\u56fe\u50cf\u6570\u636e\u96c6\u3002\u5b83\u4eec\u662f\u56fe\u50cf\u5206\u7c7b\u9886\u57df\u6700\u521d\u7684\u201cHello, World\u201d&#xff0c;\u975e\u5e38\u9002\u5408\u7528\u4e8e\u5feb\u901f\u9a8c\u8bc1\u548c\u6d4b\u8bd5\u65b0\u7684\u6a21\u578b\u60f3\u6cd5\u3002<\/li>\n<li>CIFAR-10 \/ CIFAR-100&#xff1a;\u5305\u542b\u4e8610\u4e2a\u6216100\u4e2a\u7c7b\u522b\u768460000\u5f2032&#215;32\u5f69\u8272\u56fe\u50cf\u3002\u76f8\u6bd4MNIST&#xff0c;\u5b83\u4eec\u66f4\u5177\u6311\u6218\u6027&#xff0c;\u662f\u8bc4\u4f30\u5c0f\u578b\u56fe\u50cf\u5206\u7c7b\u6a21\u578b\u6027\u80fd\u7684\u5e38\u7528\u57fa\u51c6\u3002<\/li>\n<li>ImageNet&#xff1a;\u4e00\u4e2a\u62e5\u6709\u8d85\u8fc71400\u4e07\u5f20\u56fe\u50cf\u3001\u8d85\u8fc72\u4e07\u4e2a\u7c7b\u522b\u7684\u5927\u89c4\u6a21\u56fe\u50cf\u6570\u636e\u96c6\u3002ImageNet\u6311\u6218\u8d5b&#xff08;ILSVRC&#xff09;\u6781\u5927\u5730\u63a8\u52a8\u4e86\u6df1\u5ea6\u5b66\u4e60\u548c\u8ba1\u7b97\u673a\u89c6\u89c9\u7684\u53d1\u5c55&#xff0c;\u50ac\u751f\u4e86AlexNet\u3001VGG\u3001ResNet\u7b49\u4e00\u7cfb\u5217\u91cc\u7a0b\u7891\u5f0f\u7684\u7f51\u7edc\u67b6\u6784\u3002<\/li>\n<li>COCO (Common Objects in Context)&#xff1a;\u4e00\u4e2a\u5927\u89c4\u6a21\u7684\u3001\u65e8\u5728\u63a8\u52a8\u76ee\u6807\u68c0\u6d4b\u3001\u56fe\u50cf\u5206\u5272\u548c\u56fe\u50cf\u63cf\u8ff0\u7b49\u66f4\u590d\u6742\u89c6\u89c9\u4efb\u52a1\u7684\u6570\u636e\u96c6\u3002\u5b83\u7684\u7279\u70b9\u662f\u56fe\u50cf\u573a\u666f\u590d\u6742&#xff0c;\u5355\u4e2a\u56fe\u50cf\u4e2d\u5305\u542b\u591a\u4e2a\u7269\u4f53\u3002<\/li>\n<\/ul>\n<p>\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;<\/p>\n<ul>\n<li>IMDb Movie Reviews&#xff1a;\u4e00\u4e2a\u5305\u542b\u4e8650000\u6761\u7535\u5f71\u8bc4\u8bba\u7684\u6587\u672c\u6570\u636e\u96c6&#xff0c;\u88ab\u6807\u8bb0\u4e3a\u6b63\u9762\u6216\u8d1f\u9762\u3002\u5b83\u662f\u5b66\u4e60\u548c\u5b9e\u8df5\u6587\u672c\u60c5\u611f\u5206\u6790\u4efb\u52a1\u7684\u7ecf\u5178\u9009\u62e9\u3002<\/li>\n<li>SQuAD (Stanford Question Answering Dataset)&#xff1a;\u4e00\u4e2a\u5927\u89c4\u6a21\u7684\u673a\u5668\u9605\u8bfb\u7406\u89e3\u6570\u636e\u96c6\u3002\u5176\u5f62\u5f0f\u4e3a&#xff08;\u6587\u7ae0&#xff0c;\u95ee\u9898&#xff0c;\u7b54\u6848&#xff09;\u4e09\u5143\u7ec4&#xff0c;\u8981\u6c42\u6a21\u578b\u6839\u636e\u7ed9\u5b9a\u7684\u6587\u7ae0&#xff0c;\u56de\u7b54\u76f8\u5173\u95ee\u9898&#xff0c;\u662f\u8861\u91cf\u95ee\u7b54\u7cfb\u7edf\u80fd\u529b\u7684\u91cd\u8981\u57fa\u51c6\u3002<\/li>\n<li>WMT (Workshop on Machine Translation)&#xff1a;\u673a\u5668\u7ffb\u8bd1\u9886\u57df\u7684\u5e74\u5ea6\u8bc4\u6d4b&#xff0c;\u63d0\u4f9b\u4e86\u591a\u79cd\u8bed\u8a00\u5bf9\u4e4b\u95f4\u7684\u5927\u89c4\u6a21\u9ad8\u8d28\u91cf\u5e73\u884c\u8bed\u6599\u5e93&#xff0c;\u662f\u8bad\u7ec3\u548c\u8bc4\u4f30\u673a\u5668\u7ffb\u8bd1\u6a21\u578b\u7684\u6807\u51c6\u6570\u636e\u96c6\u3002<\/li>\n<li>GLUE \/ SuperGLUE&#xff1a;\u4e00\u7cfb\u5217NLP\u4efb\u52a1\u7684\u96c6\u5408&#xff0c;\u65e8\u5728\u5168\u9762\u8bc4\u4f30\u4e00\u4e2a\u8bed\u8a00\u6a21\u578b\u5728\u8bed\u4e49\u7406\u89e3\u3001\u63a8\u7406\u3001\u60c5\u611f\u5206\u6790\u7b49\u591a\u79cd\u80fd\u529b\u4e0a\u7684\u7efc\u5408\u8868\u73b0\u3002\u5b83\u4eec\u5982\u540c\u8bed\u8a00\u6a21\u578b\u9886\u57df\u7684\u201c\u9ad8\u8003\u201d&#xff0c;\u662f\u8861\u91cf\u6a21\u578b\u6cdb\u5316\u80fd\u529b\u7684\u91cd\u8981\u6807\u5c3a\u3002<\/li>\n<\/ul>\n<p>\u5176\u4ed6\u9886\u57df<\/p>\n<ul>\n<li>MovieLens&#xff1a;\u7531GroupLens\u7814\u7a76\u5c0f\u7ec4\u53d1\u5e03\u7684\u4e00\u7cfb\u5217\u7535\u5f71\u8bc4\u5206\u6570\u636e\u96c6&#xff0c;\u5305\u542b\u4e86\u4e0d\u540c\u89c4\u6a21\u7684\u7528\u6237\u5bf9\u7535\u5f71\u7684\u8bc4\u5206\u6570\u636e\u3002\u5b83\u662f\u63a8\u8350\u7cfb\u7edf\u7814\u7a76\u9886\u57df\u6700\u5e38\u7528\u3001\u6700\u7ecf\u5178\u7684\u5165\u95e8\u548c\u57fa\u51c6\u6570\u636e\u96c6\u3002<\/li>\n<li>UCI Machine Learning Repository&#xff1a;\u4e00\u4e2a\u6c47\u96c6\u4e86\u6570\u767e\u4e2a\u7ecf\u5178\u673a\u5668\u5b66\u4e60\u6570\u636e\u96c6\u7684\u5b9d\u5e93\u3002\u8fd9\u4e9b\u6570\u636e\u96c6\u8986\u76d6\u4e86\u5206\u7c7b\u3001\u56de\u5f52\u3001\u805a\u7c7b\u7b49\u591a\u79cd\u4efb\u52a1&#xff0c;\u89c4\u6a21\u901a\u5e38\u8f83\u5c0f&#xff0c;\u975e\u5e38\u9002\u5408\u7528\u4e8e\u5b66\u4e60\u548c\u5b9e\u8df5\u4f20\u7edf\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u3002<\/li>\n<\/ul>\n<p>A.2 \u5b66\u4e60\u4e0e\u4ea4\u6d41\u8d44\u6e90<\/p>\n<ul>\n<li>\u5728\u7ebf\u8bfe\u7a0b\u5e73\u53f0&#xff1a;\n<ul>\n<li>Coursera: Andrew Ng\u6559\u6388\u7684\u300aMachine Learning\u300b\u548c\u300aDeep Learning Specialization\u300b\u662f\u65e0\u6570\u4eba\u7684\u542f\u8499\u8bfe\u7a0b\u3002<\/li>\n<li>fast.ai: Jeremy Howard\u7684\u8bfe\u7a0b\u4ee5\u201c\u4ee3\u7801\u4f18\u5148\u201d\u7684\u5b9e\u8df5\u6d3e\u98ce\u683c\u8457\u79f0&#xff0c;\u5f3a\u8c03\u5feb\u901f\u4e0a\u624b\u548c\u89e3\u51b3\u5b9e\u9645\u95ee\u9898\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u5b66\u672f\u8bba\u6587\u4e0e\u9884\u5370\u672c&#xff1a;\n<ul>\n<li>arXiv.org: \u5eb7\u5948\u5c14\u5927\u5b66\u8fd0\u8425\u7684\u9884\u5370\u672c\u670d\u52a1\u5668&#xff0c;\u662f\u83b7\u53d6\u6700\u65b0AI\u7814\u7a76\u8bba\u6587\u7684\u7b2c\u4e00\u7ad9\u3002\u91cd\u70b9\u5173\u6ce8cs.LG\u00a0(\u673a\u5668\u5b66\u4e60),\u00a0cs.CV\u00a0(\u8ba1\u7b97\u673a\u89c6\u89c9),\u00a0cs.CL\u00a0(\u8ba1\u7b97\u8bed\u8a00\u5b66)\u7b49\u5206\u7c7b\u3002<\/li>\n<li>Papers with Code: \u4e00\u4e2a\u5c06\u5b66\u672f\u8bba\u6587\u4e0e\u5f00\u6e90\u4ee3\u7801\u5b9e\u73b0\u7d27\u5bc6\u7ed3\u5408\u7684\u7f51\u7ad9&#xff0c;\u662f\u590d\u73b0\u548c\u5b66\u4e60\u6700\u65b0\u7814\u7a76\u6210\u679c\u7684\u7edd\u4f73\u5e73\u53f0\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae&#xff1a;AI\u9886\u57df\u7684\u91cd\u5927\u7a81\u7834\u901a\u5e38\u5728\u4ee5\u4e0b\u9876\u7ea7\u4f1a\u8bae\u4e0a\u53d1\u5e03&#xff1a;\n<ul>\n<li>\u901a\u7528\u673a\u5668\u5b66\u4e60: NeurIPS, ICML, ICLR<\/li>\n<li>\u8ba1\u7b97\u673a\u89c6\u89c9: CVPR, ICCV, ECCV<\/li>\n<li>\u81ea\u7136\u8bed\u8a00\u5904\u7406: ACL, EMNLP, NAACL<\/li>\n<\/ul>\n<\/li>\n<li>\u6846\u67b6\u4e0e\u5de5\u5177\u5b98\u65b9\u6587\u6863&#xff1a;\u5b98\u65b9\u6587\u6863\u6c38\u8fdc\u662f\u5b66\u4e60\u4e00\u4e2a\u5de5\u5177\u6700\u51c6\u786e\u3001\u6700\u53ef\u9760\u7684\u53c2\u8003\u3002\n<ul>\n<li>TensorFlow:\u00a0<span>http:\/\/www.tensorflow.org<\/span><\/li>\n<li>PyTorch:\u00a0<span>http:\/\/pytorch.org<\/span><\/li>\n<li>Keras:\u00a0<span>http:\/\/keras.io<\/span><\/li>\n<li>Scikit-learn:\u00a0<span>http:\/\/scikit-learn.org<\/span><\/li>\n<li>Hugging Face:\u00a0<span>http:\/\/huggingface.co<\/span><\/li>\n<\/ul>\n<\/li>\n<li>\u6280\u672f\u535a\u5ba2\u4e0e\u793e\u533a&#xff1a;\n<ul>\n<li>Distill.pub: \u4e00\u4e2a\u81f4\u529b\u4e8e\u4ee5\u6e05\u6670\u3001\u4ea4\u4e92\u5f0f\u7684\u65b9\u5f0f\u9610\u8ff0\u673a\u5668\u5b66\u4e60\u7814\u7a76\u7684\u5b66\u672f\u671f\u520a\u3002<\/li>\n<li>The Gradient: \u4e00\u4e2a\u53d1\u5e03\u5bf9AI\u9886\u57df\u6df1\u5ea6\u601d\u8003\u548c\u8bc4\u8bba\u7684\u72ec\u7acb\u51fa\u7248\u7269\u3002<\/li>\n<li>Reddit: r\/MachineLearning, r\/deeplearning\u7b49\u677f\u5757\u662f\u83b7\u53d6\u6700\u65b0\u8d44\u8baf\u548c\u53c2\u4e0e\u4e13\u4e1a\u8ba8\u8bba\u7684\u6d3b\u8dc3\u793e\u533a\u3002<\/li>\n<li>Stack Overflow: \u89e3\u51b3\u7f16\u7a0b\u548c\u6280\u672f\u95ee\u9898\u65f6\u6700\u79bb\u4e0d\u5f00\u7684\u95ee\u7b54\u793e\u533a\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>B&#xff1a;\u6570\u5b66\u7b26\u53f7\u8868<\/h4>\n<p>\u5728\u9605\u8bfb\u5b66\u672f\u8bba\u6587\u6216\u66f4\u6df1\u5165\u7684\u7406\u8bba\u4e66\u7c4d\u65f6&#xff0c;\u4e00\u5957\u6807\u51c6\u5316\u7684\u6570\u5b66\u7b26\u53f7\u662f\u4e0d\u53ef\u6216\u7f3a\u7684\u4ea4\u6d41\u8bed\u8a00\u3002\u672c\u8868\u5217\u51fa\u4e86\u672c\u4e66\u53ca\u76f8\u5173\u9886\u57df\u4e2d\u6700\u5e38\u7528\u7684\u6570\u5b66\u7b26\u53f7&#xff0c;\u4ee5\u5907\u8bfb\u8005\u901f\u67e5\u3002<\/p>\n<table>\n<tr>\n<p>\u7b26\u53f7<\/p>\n<p>\u542b\u4e49<\/p>\n<p>\u793a\u4f8b<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>\u901a\u7528\u6570\u5b66<\/p>\n<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>x, y, z<\/p>\n<\/td>\n<td>\n<p>\u6807\u91cf (\u5c0f\u5199\u659c\u4f53)<\/p>\n<\/td>\n<td>\n<p>y &#061; wx &#043; b<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>x, y, z<\/p>\n<\/td>\n<td>\n<p>\u5411\u91cf (\u5c0f\u5199\u7c97\u4f53)<\/p>\n<\/td>\n<td>\n<p>y &#061; Wx &#043; b<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>X, Y, Z<\/p>\n<\/td>\n<td>\n<p>\u77e9\u9635 (\u5927\u5199\u7c97\u4f53)<\/p>\n<\/td>\n<td>\n<p>Z &#061; XW<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>&#x1d4b3;, &#x1d4b4;, &#x1d4b5;<\/p>\n<\/td>\n<td>\n<p>\u5f20\u91cf (\u5927\u5199\u82b1\u4f53)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u03a3<\/p>\n<\/td>\n<td>\n<p>\u6c42\u548c<\/p>\n<\/td>\n<td>\n<p>\u03a3\u1d62 x\u1d62<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u03a0<\/p>\n<\/td>\n<td>\n<p>\u6c42\u79ef<\/p>\n<\/td>\n<td>\n<p>\u03a0\u1d62 x\u1d62<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u222b<\/p>\n<\/td>\n<td>\n<p>\u79ef\u5206<\/p>\n<\/td>\n<td>\n<p>\u222b f(x) dx<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u2202<\/p>\n<\/td>\n<td>\n<p>\u504f\u5bfc\u6570<\/p>\n<\/td>\n<td>\n<p>\u2202L\/\u2202w<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u2207<\/p>\n<\/td>\n<td>\n<p>\u68af\u5ea6 (Nabla\u7b97\u5b50)<\/p>\n<\/td>\n<td>\n<p>\u2207J(\u03b8)<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u7ebf\u6027\u4ee3\u6570<\/p>\n<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>X^T<\/p>\n<\/td>\n<td>\n<p>\u77e9\u9635\/\u5411\u91cf\u8f6c\u7f6e<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>X\u207b\u00b9<\/p>\n<\/td>\n<td>\n<p>\u77e9\u9635\u7684\u9006<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>x \u00b7 y<\/p>\n<\/td>\n<td>\n<p>\u5411\u91cf\u70b9\u79ef<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>x \u2297 y<\/p>\n<\/td>\n<td>\n<p>\u5411\u91cf\u5916\u79ef<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>X \u2299 Y<\/p>\n<\/td>\n<td>\n<p>Hadamard\u79ef (\u5143\u7d20\u5bf9\u5e94\u76f8\u4e58)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td><\/td>\n<td>\n<p>x<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u6982\u7387\u8bba\u4e0e\u7edf\u8ba1<\/p>\n<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>P(A)<\/p>\n<\/td>\n<td>\n<p>\u4e8b\u4ef6A\u53d1\u751f\u7684\u6982\u7387<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>p(x)<\/p>\n<\/td>\n<td>\n<p>\u8fde\u7eed\u53d8\u91cf\u7684\u6982\u7387\u5bc6\u5ea6\u51fd\u6570 (PDF)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>E[X]<\/p>\n<\/td>\n<td>\n<p>\u968f\u673a\u53d8\u91cfX\u7684\u671f\u671b<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>Var(X)<\/p>\n<\/td>\n<td>\n<p>\u968f\u673a\u53d8\u91cfX\u7684\u65b9\u5dee<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>Cov(X, Y)<\/p>\n<\/td>\n<td>\n<p>\u968f\u673a\u53d8\u91cfX\u548cY\u7684\u534f\u65b9\u5dee<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>x ~ D<\/p>\n<\/td>\n<td>\n<p>\u968f\u673a\u53d8\u91cfx\u670d\u4ece\u5206\u5e03D<\/p>\n<\/td>\n<td>\n<p>x ~ N(\u03bc, \u03c3\u00b2)<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>N(\u03bc, \u03c3\u00b2)<\/p>\n<\/td>\n<td>\n<p>\u5747\u503c\u4e3a\u03bc&#xff0c;\u65b9\u5dee\u4e3a\u03c3\u00b2\u7684\u9ad8\u65af\u5206\u5e03<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u795e\u7ecf\u7f51\u7edc<\/p>\n<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>w\u1d62\u2c7c, b\u1d62<\/p>\n<\/td>\n<td>\n<p>\u6743\u91cd\u548c\u504f\u7f6e<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>z\u207d\u02e1\u207e<\/p>\n<\/td>\n<td>\n<p>\u7b2cl\u5c42\u7684\u7ebf\u6027\u8f93\u51fa (z-value)<\/p>\n<\/td>\n<td>\n<p>z\u207d\u02e1\u207e &#061; w\u207d\u02e1\u207ea\u207d\u02e1\u207b\u00b9\u207e &#043; b\u207d\u02e1\u207e<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>a\u207d\u02e1\u207e<\/p>\n<\/td>\n<td>\n<p>\u7b2cl\u5c42\u7684\u6fc0\u6d3b\u8f93\u51fa<\/p>\n<\/td>\n<td>\n<p>a\u207d\u02e1\u207e &#061; g(z\u207d\u02e1\u207e)<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>g(z), \u03c3(z)<\/p>\n<\/td>\n<td>\n<p>\u6fc0\u6d3b\u51fd\u6570, Sigmoid\u51fd\u6570<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>L(\u0177, y)<\/p>\n<\/td>\n<td>\n<p>\u635f\u5931\u51fd\u6570 (\u5355\u4e2a\u6837\u672c)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>J(\u03b8)<\/p>\n<\/td>\n<td>\n<p>\u4ee3\u4ef7\u51fd\u6570 (\u6574\u4e2a\u6570\u636e\u96c6\u7684\u5e73\u5747\u635f\u5931)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u0177<\/p>\n<\/td>\n<td>\n<p>\u6a21\u578b\u7684\u9884\u6d4b\u503c (y-hat)<\/p>\n<\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u03b8<\/p>\n<\/td>\n<td>\n<p>\u6a21\u578b\u7684\u53c2\u6570\u96c6\u5408<\/p>\n<\/td>\n<td>\n<p>\u03b8 &#061; {W, b}<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<hr \/>\n<h4>C&#xff1a;\u5e38\u89c1\u95ee\u9898\u4e0e\u6392\u9519\u6307\u5357<\/h4>\n<p>\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u5b9e\u8df5\u4e2d&#xff0c;\u9047\u5230\u5404\u79cd\u9519\u8bef\u548c\u95ee\u9898\u662f\u5728\u6240\u96be\u514d\u7684\u3002\u8fd9\u91cc\u6211\u4eec\u6574\u7406\u4e86\u4e00\u4e9b\u6700\u5e38\u89c1\u7684\u201c\u7591\u96be\u6742\u75c7\u201d\u53ca\u5176\u6392\u67e5\u601d\u8def&#xff0c;\u5e0c\u671b\u80fd\u6210\u4e3a\u60a8\u65c5\u9014\u4e2d\u7684\u201c\u6025\u6551\u5305\u201d\u3002<\/p>\n<p>C.1 \u6a21\u578b\u8bad\u7ec3\u76f8\u5173<\/p>\n<ul>\n<li>\n<p>\u95ee\u9898&#xff1a;\u201c\u6211\u7684\u6a21\u578b\u4e0d\u6536\u655b&#xff0c;\u635f\u5931&#xff08;Loss&#xff09;\u4e00\u76f4\u4e0d\u4e0b\u964d\u600e\u4e48\u529e&#xff1f;\u201d<\/p>\n<ul>\n<li>\u6392\u67e5\u6e05\u5355&#xff1a;\n<li>\u5b66\u4e60\u7387&#xff08;Learning Rate&#xff09;&#xff1a;\u8fd9\u662f\u6700\u5e38\u89c1\u7684\u201c\u5143\u51f6\u201d\u3002\u5b66\u4e60\u7387\u8fc7\u5927&#xff0c;\u4f1a\u5bfc\u81f4\u4f18\u5316\u8fc7\u7a0b\u5728\u6700\u4f18\u89e3\u9644\u8fd1\u5267\u70c8\u9707\u8361\u751a\u81f3\u53d1\u6563&#xff1b;\u5b66\u4e60\u7387\u8fc7\u5c0f&#xff0c;\u5219\u4f1a\u5bfc\u81f4\u6536\u655b\u6781\u5176\u7f13\u6162\u6216\u9677\u5165\u5c40\u90e8\u6700\u4f18\u3002\u53ef\u4ee5\u5c1d\u8bd5\u4ece\u4e00\u4e2a\u8f83\u5c0f\u7684\u503c&#xff08;\u59821e-4&#xff09;\u5f00\u59cb&#xff0c;\u9010\u6e10\u589e\u5927\u8fdb\u884c\u5b9e\u9a8c\u3002<\/li>\n<li>\u6570\u636e\u5f52\u4e00\u5316&#xff1a;\u786e\u4fdd\u8f93\u5165\u6570\u636e\u5df2\u7ecf\u8fdb\u884c\u4e86\u9002\u5f53\u7684\u5f52\u4e00\u5316&#xff08;\u5982\u7f29\u653e\u52300-1\u6216\u6807\u51c6\u5316\u4e3a\u5747\u503c\u4e3a0\u3001\u65b9\u5dee\u4e3a1&#xff09;\u3002\u8fd9\u5bf9\u4e8e\u52a0\u901f\u6536\u655b\u548c\u907f\u514d\u6570\u503c\u95ee\u9898\u81f3\u5173\u91cd\u8981\u3002<\/li>\n<li>\u635f\u5931\u51fd\u6570\u4e0e\u6a21\u578b\u8f93\u51fa&#xff1a;\u68c0\u67e5\u635f\u5931\u51fd\u6570\u662f\u5426\u4e0e\u6a21\u578b\u7684\u8f93\u51fa\u5c42\u6fc0\u6d3b\u51fd\u6570\u5339\u914d\u3002\u4f8b\u5982&#xff0c;\u4e8c\u5143\u5206\u7c7b\u901a\u5e38\u4f7f\u7528sigmoid\u6fc0\u6d3b\u548cbinary_crossentropy\u635f\u5931&#xff1b;\u591a\u5143\u5206\u7c7b\u4f7f\u7528softmax\u6fc0\u6d3b\u548ccategorical_crossentropy\u635f\u5931\u3002<\/li>\n<li>\u6a21\u578b\u7ed3\u6784&#xff1a;\u6a21\u578b\u662f\u5426\u8fc7\u4e8e\u7b80\u5355&#xff0c;\u4ee5\u81f3\u4e8e\u65e0\u6cd5\u6355\u6349\u6570\u636e\u7684\u590d\u6742\u6027&#xff08;\u6b20\u62df\u5408&#xff09;&#xff1f;\u53ef\u4ee5\u5c1d\u8bd5\u9002\u5f53\u589e\u52a0\u6a21\u578b\u7684\u6df1\u5ea6\u6216\u5bbd\u5ea6\u3002<\/li>\n<li>\u6570\u636e\u672c\u8eab&#xff1a;\u68c0\u67e5\u6570\u636e\u6807\u7b7e\u662f\u5426\u5b58\u5728\u9519\u8bef&#xff0c;\u6216\u8005\u6570\u636e\u548c\u6807\u7b7e\u662f\u5426\u6b63\u786e\u5bf9\u5e94\u3002\u53ef\u4ee5\u5c1d\u8bd5\u7528\u4e00\u4e2a\u6781\u5c0f\u7684\u5b50\u96c6&#xff08;\u598210\u4e2a\u6837\u672c&#xff09;\u8fdb\u884c\u8bad\u7ec3&#xff0c;\u770b\u6a21\u578b\u80fd\u5426\u5728\u8fd9\u4e2a\u5b50\u96c6\u4e0a\u8fc7\u62df\u5408&#xff0c;\u8fd9\u53ef\u4ee5\u9a8c\u8bc1\u6574\u4e2a\u8bad\u7ec3\u6d41\u7a0b\u662f\u5426\u80fd\u6b63\u5e38\u5de5\u4f5c\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u95ee\u9898&#xff1a;\u201c\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u4e0a\u8868\u73b0\u5f88\u597d&#xff0c;\u4f46\u5728\u9a8c\u8bc1\u96c6\u4e0a\u5f88\u5dee&#xff0c;\u8fd9\u662f\u4e3a\u4ec0\u4e48&#xff1f;\u201d<\/p>\n<ul>\n<li>\u89e3\u7b54&#xff1a;\u8fd9\u662f\u5178\u578b\u7684\u8fc7\u62df\u5408&#xff08;Overfitting&#xff09;\u3002\u6a21\u578b\u8fc7\u5ea6\u5730\u5b66\u4e60\u4e86\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u566a\u58f0\u548c\u7ec6\u8282&#xff0c;\u800c\u5931\u53bb\u4e86\u5bf9\u65b0\u6570\u636e\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/li>\n<li>\u89e3\u51b3\u65b9\u6848&#xff1a;\n<li>\u589e\u52a0\u6570\u636e&#xff1a;\u83b7\u53d6\u66f4\u591a\u7684\u8bad\u7ec3\u6570\u636e\u662f\u89e3\u51b3\u8fc7\u62df\u5408\u6700\u6709\u6548\u7684\u65b9\u6cd5\u3002\u5982\u679c\u65e0\u6cd5\u83b7\u53d6\u65b0\u6570\u636e&#xff0c;\u53ef\u4ee5\u91c7\u7528**\u6570\u636e\u589e\u5f3a&#xff08;Data Augmentation&#xff09;**\u6280\u672f\u6765\u4eba\u5de5\u6269\u5145\u6570\u636e\u96c6\u3002<\/li>\n<li>\u4f7f\u7528\u6b63\u5219\u5316&#xff1a;\n<ul>\n<li>L1\/L2\u6b63\u5219\u5316&#xff1a;\u5728\u635f\u5931\u51fd\u6570\u4e2d\u52a0\u5165\u5bf9\u6a21\u578b\u6743\u91cd\u7684\u60e9\u7f5a\u9879&#xff0c;\u9650\u5236\u6a21\u578b\u7684\u590d\u6742\u5ea6\u3002<\/li>\n<li>Dropout&#xff1a;\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d&#xff0c;\u968f\u673a\u5730\u201c\u4e22\u5f03\u201d\u4e00\u90e8\u5206\u795e\u7ecf\u5143&#xff0c;\u8feb\u4f7f\u7f51\u7edc\u5b66\u4e60\u5230\u66f4\u9c81\u68d2\u7684\u7279\u5f81\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u65e9\u505c&#xff08;Early Stopping&#xff09;&#xff1a;\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u76d1\u63a7\u9a8c\u8bc1\u96c6\u7684\u6027\u80fd&#xff0c;\u5f53\u9a8c\u8bc1\u96c6\u6027\u80fd\u4e0d\u518d\u63d0\u5347\u65f6&#xff0c;\u5c31\u63d0\u524d\u7ec8\u6b62\u8bad\u7ec3\u3002<\/li>\n<li>\u7b80\u5316\u6a21\u578b&#xff1a;\u4f7f\u7528\u4e00\u4e2a\u66f4\u5c0f\u3001\u66f4\u7b80\u5355\u7684\u7f51\u7edc\u7ed3\u6784\u3002<\/li>\n<li>\u8fc1\u79fb\u5b66\u4e60&#xff1a;\u4f7f\u7528\u5728\u4e00\u4e2a\u5927\u578b\u6570\u636e\u96c6&#xff08;\u5982ImageNet&#xff09;\u4e0a\u9884\u8bad\u7ec3\u597d\u7684\u6a21\u578b&#xff0c;\u5728\u60a8\u7684\u5c0f\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u5fae\u8c03\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u95ee\u9898&#xff1a;\u201cNaN&#xff08;Not a Number&#xff09;\u51fa\u73b0\u5728\u6211\u7684\u635f\u5931\u4e2d&#xff0c;\u8fd9\u662f\u600e\u4e48\u56de\u4e8b&#xff1f;\u201d<\/p>\n<ul>\n<li>\u89e3\u7b54&#xff1a;\u8fd9\u901a\u5e38\u662f\u7531\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u6570\u503c\u4e0d\u7a33\u5b9a\u5f15\u8d77\u7684\u3002<\/li>\n<li>\u6392\u67e5\u6e05\u5355&#xff1a;\n<li>\u5b66\u4e60\u7387\u8fc7\u5927&#xff1a;\u8fc7\u5927\u7684\u5b66\u4e60\u7387\u53ef\u80fd\u5bfc\u81f4\u68af\u5ea6\u66f4\u65b0\u8fc7\u731b&#xff0c;\u4f7f\u5f97\u6743\u91cd\u53c2\u6570\u53d8\u5f97\u6781\u5927\u6216\u6781\u5c0f&#xff0c;\u6700\u7ec8\u5bfc\u81f4\u6570\u503c\u6ea2\u51fa&#xff0c;\u4ea7\u751fNaN\u3002\u8fd9\u662f\u6700\u5e38\u89c1\u7684\u539f\u56e0\u3002<\/li>\n<li>\u6570\u5b66\u8fd0\u7b97\u9519\u8bef&#xff1a;\u68c0\u67e5\u4ee3\u7801\u4e2d\u662f\u5426\u5b58\u5728\u53ef\u80fd\u5bfc\u81f4\u6570\u503c\u95ee\u9898\u7684\u8fd0\u7b97&#xff0c;\u4f8b\u5982\u5bf9\u4e00\u4e2a\u8d1f\u6570\u6216\u96f6\u53d6\u5bf9\u6570&#xff08;log(0)&#xff09;&#xff0c;\u6216\u8005\u9664\u4ee5\u4e00\u4e2a\u53ef\u80fd\u4e3a\u96f6\u7684\u6570\u3002<\/li>\n<li>\u68af\u5ea6\u88c1\u526a&#xff08;Gradient Clipping&#xff09;&#xff1a;\u8fd9\u662f\u4e00\u4e2a\u6709\u6548\u7684\u5e94\u5bf9\u7b56\u7565\u3002\u5728\u4f18\u5316\u5668\u66f4\u65b0\u6743\u91cd\u4e4b\u524d&#xff0c;\u68c0\u67e5\u68af\u5ea6\u7684\u8303\u6570&#xff0c;\u5982\u679c\u8d85\u8fc7\u4e00\u4e2a\u9608\u503c&#xff0c;\u5c31\u5c06\u5176\u7f29\u653e\u5230\u4e00\u4e2a\u56fa\u5b9a\u7684\u8303\u56f4\u5185&#xff0c;\u9632\u6b62\u68af\u5ea6\u7206\u70b8\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>C.2 \u73af\u5883\u4e0e\u6570\u636e\u76f8\u5173<\/p>\n<ul>\n<li>\n<p>\u95ee\u9898&#xff1a;\u201c\u5982\u4f55\u9009\u62e9\u4f7f\u7528CPU\u8fd8\u662fGPU&#xff1f;\u201d<\/p>\n<ul>\n<li>\u89e3\u7b54&#xff1a;\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7684\u8bad\u7ec3&#xff0c;\u5c24\u5176\u662fCNN\u3001RNN\u7b49&#xff0c;\u6d89\u53ca\u6d77\u91cf\u7684\u77e9\u9635\u4e58\u6cd5\u548c\u5e76\u884c\u8ba1\u7b97\u3002GPU&#xff08;\u56fe\u5f62\u5904\u7406\u5668&#xff09;\u7684\u6838\u5fc3\u8bbe\u8ba1\u6b63\u662f\u4e3a\u4e86\u9ad8\u6548\u5730\u6267\u884c\u8fd9\u7c7b\u4efb\u52a1&#xff0c;\u56e0\u6b64\u5b83\u662f\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60\u8bad\u7ec3\u7684\u9996\u9009\u548c\u201c\u6807\u914d\u201d\u3002\u5bf9\u4e8e\u5927\u578b\u6a21\u578b\u7684\u8bad\u7ec3&#xff0c;CPU\u51e0\u4e4e\u662f\u4e0d\u53ef\u884c\u7684\u3002\u5728**\u63a8\u7406&#xff08;Inference&#xff09;**\u9636\u6bb5&#xff0c;\u6216\u8005\u5bf9\u4e8e\u4e00\u4e9b\u975e\u5e38\u5c0f\u578b\u7684\u6a21\u578b\u548c\u7b80\u5355\u7684\u4efb\u52a1&#xff0c;CPU\u4e5f\u53ef\u4ee5\u80dc\u4efb\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u95ee\u9898&#xff1a;\u201c\u5f20\u91cf\u7684\u5f62\u72b6&#xff08;Shape&#xff09;\u4e0d\u5339\u914d\u9519\u8bef\u5982\u4f55\u89e3\u51b3&#xff1f;\u201d<\/p>\n<ul>\n<li>\u89e3\u7b54&#xff1a;\u8fd9\u662f\u5728\u7f16\u7a0b\u5b9e\u8df5\u4e2d\u4f1a\u9047\u5230\u7684\u6700\u9ad8\u9891\u7684\u9519\u8bef&#xff0c;\u6ca1\u6709\u4e4b\u4e00\u3002\u5b83\u610f\u5473\u7740\u6570\u636e\u5728\u6d41\u7ecf\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u65f6&#xff0c;\u67d0\u4e00\u5c42\u7684\u8f93\u51fa\u5f62\u72b6&#xff0c;\u4e0e\u4e0b\u4e00\u5c42\u7684\u671f\u671b\u8f93\u5165\u5f62\u72b6\u4e0d\u5339\u914d\u3002<\/li>\n<li>\u89e3\u51b3\u65b9\u6848&#xff1a;\n<li>\u4ed4\u7ec6\u9605\u8bfb\u9519\u8bef\u4fe1\u606f&#xff1a;\u9519\u8bef\u4fe1\u606f\u901a\u5e38\u4f1a\u660e\u786e\u6307\u51fa\u662f\u54ea\u4e00\u5c42\u51fa\u73b0\u4e86\u95ee\u9898&#xff0c;\u4ee5\u53ca\u671f\u671b\u7684\u5f62\u72b6\u548c\u5b9e\u9645\u7684\u5f62\u72b6\u662f\u4ec0\u4e48\u3002<\/li>\n<li>\u5584\u7528print(tensor.shape)&#xff1a;\u5728\u60a8\u6000\u7591\u7684\u5c42\u524d\u540e&#xff0c;\u6253\u5370\u51fa\u5f20\u91cf\u7684\u5f62\u72b6&#xff0c;\u8fd9\u662f\u6700\u76f4\u63a5\u3001\u6700\u6709\u6548\u7684\u8c03\u8bd5\u65b9\u6cd5\u3002<\/li>\n<li>\u5229\u7528model.summary()&#xff1a;\u5bf9\u4e8eKeras\u7b49\u6846\u67b6&#xff0c;\u8fd9\u4e2a\u51fd\u6570\u53ef\u4ee5\u6e05\u6670\u5730\u5217\u51fa\u6a21\u578b\u6bcf\u4e00\u5c42\u7684\u8f93\u51fa\u5f62\u72b6&#xff0c;\u5e2e\u52a9\u60a8\u4ece\u5168\u5c40\u4e0a\u68c0\u67e5\u6570\u636e\u6d41\u7684\u7ef4\u5ea6\u53d8\u5316\u3002<\/li>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<h4>D&#xff1a;\u8fdb\u4e00\u6b65\u9605\u8bfb\u5efa\u8bae<\/h4>\n<p>\u672c\u4e66\u4e3a\u60a8\u6253\u4e0b\u4e86\u575a\u5b9e\u7684\u57fa\u7840&#xff0c;\u4f46\u5b66\u6d77\u65e0\u6daf\u3002\u4ee5\u4e0b\u4e66\u7c4d\u63a8\u8350&#xff0c;\u5c06\u4e3a\u60a8\u5728\u7279\u5b9a\u65b9\u5411\u7684\u6df1\u5165\u63a2\u7d22&#xff0c;\u6216\u7406\u8bba\u9ad8\u5ea6\u7684\u8fdb\u4e00\u6b65\u63d0\u5347&#xff0c;\u63d0\u4f9b\u53ef\u9760\u7684\u6307\u5f15\u3002<\/p>\n<p>D.1 \u7ecf\u5178\u8457\u4f5c<\/p>\n<ul>\n<li>\u300aDeep Learning\u300b\u00a0(Ian Goodfellow, Yoshua Bengio, Aaron Courville)\n<ul>\n<li>\u4fd7\u79f0\u201c\u82b1\u4e66\u201d&#xff0c;\u7531\u4e09\u4f4d\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u5de8\u64d8\u5408\u8457\u3002\u8fd9\u672c\u4e66\u4ee5\u5176\u7406\u8bba\u7684\u6df1\u5ea6\u3001\u8986\u76d6\u7684\u5e7f\u5ea6\u548c\u6570\u5b66\u7684\u4e25\u8c28\u6027\u8457\u79f0&#xff0c;\u662f\u4efb\u4f55\u5e0c\u671b\u6df1\u5165\u7406\u89e3\u6df1\u5ea6\u5b66\u4e60\u80cc\u540e\u539f\u7406\u7684\u8bfb\u8005\u7684\u5fc5\u8bfb\u201c\u5723\u7ecf\u201d\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u300aPattern Recognition and Machine Learning\u300b\u00a0(Christopher Bishop)\n<ul>\n<li>\u7b80\u79f0PRML\u3002\u8fd9\u672c\u4e66\u4ece\u8d1d\u53f6\u65af\u7684\u89c6\u89d2&#xff0c;\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e86\u673a\u5668\u5b66\u4e60\u7684\u5404\u79cd\u6a21\u578b\u548c\u65b9\u6cd5\u3002\u867d\u7136\u5b83\u5e76\u975e\u4e13\u8bb2\u6df1\u5ea6\u5b66\u4e60&#xff0c;\u4f46\u5176\u6df1\u523b\u7684\u6982\u7387\u601d\u60f3\u548c\u4e25\u8c28\u7684\u6570\u5b66\u63a8\u5bfc&#xff0c;\u5bf9\u4e8e\u57f9\u517b\u673a\u5668\u5b66\u4e60\u7684\u201c\u5185\u529f\u201d\u5927\u6709\u88e8\u76ca\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>D.2 \u7279\u5b9a\u9886\u57df\u6df1\u5165<\/p>\n<ul>\n<li>\u300aSpeech and Language Processing\u300b\u00a0(Dan Jurafsky, James H. Martin)\n<ul>\n<li>\u81ea\u7136\u8bed\u8a00\u5904\u7406&#xff08;NLP&#xff09;\u9886\u57df\u7684\u6743\u5a01\u6559\u79d1\u4e66&#xff0c;\u4fd7\u79f0\u201c\u9f99\u4e66\u201d\u3002\u5b83\u5168\u9762\u800c\u6df1\u5165\u5730\u8986\u76d6\u4e86\u4ece\u4f20\u7edf\u7edf\u8ba1\u65b9\u6cd5\u5230\u73b0\u4ee3\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u7684NLP\u6280\u672f\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u300aComputer Vision: Algorithms and Applications\u300b\u00a0(Richard Szeliski)\n<ul>\n<li>\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u7684\u7efc\u5408\u6027\u7ecf\u5178\u8457\u4f5c&#xff0c;\u5185\u5bb9\u6db5\u76d6\u4e86\u4ece\u56fe\u50cf\u5f62\u6210\u3001\u56fe\u50cf\u5904\u7406\u5230\u4e09\u7ef4\u91cd\u5efa\u3001\u56fe\u50cf\u8bc6\u522b\u7b49\u5e7f\u6cdb\u4e3b\u9898\u3002<\/li>\n<\/ul>\n<\/li>\n<li>\u300aReinforcement Learning: An Introduction\u300b\u00a0(Richard S. Sutton, Andrew G. Barto)\n<ul>\n<li>\u5f3a\u5316\u5b66\u4e60&#xff08;RL&#xff09;\u9886\u57df\u65e0\u53ef\u4e89\u8bae\u7684\u5960\u57fa\u4e4b\u4f5c\u548c\u201c\u5723\u7ecf\u201d\u3002\u4f5c\u8005\u662f\u8be5\u9886\u57df\u7684\u5f00\u521b\u8005&#xff0c;\u4e66\u4e2d\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e86RL\u7684\u6838\u5fc3\u6982\u5ff5\u3001\u7b97\u6cd5\u548c\u601d\u60f3\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>D.3 \u524d\u6cbf\u4e0e\u601d\u60f3<\/p>\n<p>\u771f\u6b63\u7684\u5b66\u4e60&#xff0c;\u662f\u6301\u7eed\u4e0d\u65ad\u5730\u8ffd\u8e2a\u9886\u57df\u7684\u8109\u640f\u3002\u9664\u4e86\u4e0a\u8ff0\u7ecf\u5178\u4e66\u7c4d&#xff0c;\u6211\u4eec\u5f3a\u70c8\u5efa\u8bae\u60a8&#xff1a;<\/p>\n<ul>\n<li>\u517b\u6210\u9605\u8bfb\u9876\u7ea7\u4f1a\u8bae\u8bba\u6587\u7684\u4e60\u60ef&#xff1a;\u5173\u6ce8NeurIPS, ICML, ICLR, CVPR, ACL\u7b49\u4f1a\u8bae\u7684\u6700\u65b0\u5f55\u7528\u8bba\u6587&#xff0c;\u4e86\u89e3\u6700\u524d\u6cbf\u7684\u7814\u7a76\u65b9\u5411\u3002<\/li>\n<li>\u5173\u6ce8\u4f18\u79c0\u7684\u4e2a\u4eba\u6216\u56e2\u961f\u535a\u5ba2&#xff1a;\u8bb8\u591a\u9876\u5c16\u7684\u7814\u7a76\u8005\u548c\u5de5\u7a0b\u5e08&#xff08;\u5982Andrej Karpathy, Lilian Weng&#xff09;\u90fd\u4f1a\u901a\u8fc7\u535a\u5ba2\u5206\u4eab\u4ed6\u4eec\u7684\u6df1\u523b\u6d1e\u89c1\u3002<\/li>\n<li>\u4fdd\u6301\u597d\u5947&#xff0c;\u7ec8\u8eab\u5b66\u4e60&#xff1a;AI\u662f\u4e00\u4e2a\u65e5\u65b0\u6708\u5f02\u7684\u9886\u57df\u3002\u4fdd\u6301\u4e00\u9897\u5f00\u653e\u3001\u597d\u5947\u3001\u4e50\u4e8e\u5b9e\u8df5\u7684\u5fc3&#xff0c;\u662f\u5728\u8fd9\u6761\u9053\u8def\u4e0a\u4e0d\u65ad\u524d\u884c\u7684\u552f\u4e00\u79d8\u8bc0\u3002\u00a0<\/li>\n<\/ul>\n<hr \/>\n<h3>\u540e\u8bb0<\/h3>\n<p>\u4eb2\u7231\u7684\u8bfb\u8005\u670b\u53cb\u4eec&#xff1a;<\/p>\n<p>\u6211\u4eec\u7ec8\u4e8e\u8d70\u5230\u4e86\u8fd9\u91cc\u3002<\/p>\n<p>\u6211\u4f9d\u7136\u8bb0\u5f97&#xff0c;\u5728\u672c\u4e66\u7684\u5f00\u7bc7&#xff0c;\u6211\u66fe\u5c06\u6df1\u5ea6\u5b66\u4e60\u6bd4\u4f5c\u4e00\u7247\u6d69\u701a\u7684\u661f\u8fb0\u5927\u6d77\u3002\u90a3\u65f6&#xff0c;\u6211\u4eec\u8fd8\u7ad9\u5728\u5cb8\u8fb9&#xff0c;\u5bf9\u8fdc\u65b9\u7684\u7480\u74a8\u5fc3\u6000\u656c\u754f\u4e0e\u5411\u5f80\u3002\u800c\u4eca&#xff0c;\u60a8\u5df2\u7136\u662f\u4e00\u4f4d\u624b\u6301\u661f\u56fe\u3001\u80fd\u9a7e\u9a6d\u821f\u696b\u7684\u719f\u7ec3\u822a\u6d77\u5bb6\u4e86\u3002<\/p>\n<p>\u6211\u4eec\u4e00\u540c\u8d70\u8fc7\u4e86\u5f88\u957f\u7684\u8def\u3002\u4ece\u6700\u57fa\u7840\u7684\u6570\u5b66\u201c\u5fc3\u6cd5\u201d\u4e0e\u7f16\u7a0b\u201c\u62db\u5f0f\u201d&#xff0c;\u5230\u4eb2\u624b\u642d\u5efa\u8d77\u795e\u7ecf\u7f51\u7edc\u8fd9\u5ea7\u5b8f\u4f1f\u7684\u201c\u6bbf\u5802\u201d&#xff1b;\u4ece\u6df1\u5165\u5377\u79ef\u4e0e\u5faa\u73af\u7f51\u7edc\u7684\u201c\u5e7d\u5f84\u201d&#xff0c;\u5230\u9a7e\u9a6d\u6ce8\u610f\u529b\u673a\u5236\u8fd9\u5339\u9a70\u9a8b\u5728\u8bed\u8a00\u7586\u573a\u7684\u201c\u9a8f\u9a6c\u201d&#xff1b;\u6211\u4eec\u4e0d\u4ec5\u5728\u5b9e\u6218\u4e2d\u4f53\u9a8c\u4e86\u521b\u9020\u7684\u4e50\u8da3&#xff0c;\u66f4\u5728\u5de5\u7a0b\u5316\u7684\u78e8\u783a\u4e2d&#xff0c;\u5b66\u4f1a\u4e86\u5982\u4f55\u5c06\u667a\u6167\u7684\u679c\u5b9e&#xff0c;\u7a33\u59a5\u5730\u9001\u8fbe\u4eba\u95f4\u3002\u6700\u540e&#xff0c;\u6211\u4eec\u4e00\u540c\u7ad9\u5728\u5c71\u5dc5&#xff0c;\u773a\u671b\u4e86\u90a3\u6b63\u5728\u751f\u6210\u4e2d\u7684\u3001\u66f4\u5e7f\u9614\u7684\u672a\u6765\u3002<\/p>\n<p>\u6211\u5c3d\u6211\u6240\u80fd&#xff0c;\u5c06\u8fd9\u95e8\u6df1\u5965\u7684\u5b66\u95ee&#xff0c;\u5316\u4f5c\u60a8\u6613\u4e8e\u7406\u89e3\u7684\u8bed\u8a00&#xff1b;\u5c06\u8fd9\u8d9f\u8270\u6df1\u7684\u65c5\u7a0b&#xff0c;\u94fa\u6210\u4e00\u6761\u6b21\u7b2c\u5206\u660e\u7684\u9053\u8def\u3002\u6211\u8bd5\u56fe\u4e3a\u60a8\u70b9\u4eae\u7684&#xff0c;\u4e0d\u4ec5\u662f\u77e5\u8bc6\u7684\u706b\u628a&#xff0c;\u66f4\u662f\u63a2\u7d22\u672a\u77e5\u7684\u52c7\u6c14\u4e0e\u72ec\u7acb\u601d\u8003\u7684\u4e60\u60ef\u3002<\/p>\n<p>\u56e0\u4e3a&#xff0c;\u8fd9\u672c\u4e66\u7684\u771f\u6b63\u4ef7\u503c&#xff0c;\u4e0d\u5728\u4e8e\u5b83\u6559\u4f1a\u4e86\u60a8\u4ec0\u4e48&#xff0c;\u800c\u5728\u4e8e\u5b83\u80fd\u5524\u9192\u60a8\u4ec0\u4e48\u3002<\/p>\n<p>\u5b83\u5524\u9192\u7684&#xff0c;\u662f\u60a8\u4e0e\u751f\u4ff1\u6765\u7684\u3001\u5bf9\u4e16\u754c\u8fd0\u4f5c\u89c4\u5f8b\u7684\u597d\u5947\u5fc3&#xff1b;\u662f\u60a8\u9762\u5bf9\u590d\u6742\u95ee\u9898\u65f6&#xff0c;\u5c06\u5176\u62c6\u89e3\u3001\u5206\u6790\u3001\u5e76\u6700\u7ec8\u89e3\u51b3\u7684\u903b\u8f91\u80fd\u529b&#xff1b;\u662f\u60a8\u5c06\u62bd\u8c61\u7684\u7406\u8bba&#xff0c;\u4e0e\u9c9c\u6d3b\u7684\u73b0\u5b9e\u4e16\u754c\u76f8\u8fde\u63a5\u7684\u6d1e\u5bdf\u529b\u3002\u6700\u91cd\u8981\u7684\u662f&#xff0c;\u5b83\u5524\u9192\u4e86\u60a8\u4f5c\u4e3a\u4e00\u540d\u201c\u521b\u9020\u8005\u201d\u7684\u81ea\u89c9\u3002<\/p>\n<p>\u662f\u7684&#xff0c;\u521b\u9020\u8005\u3002\u4ece\u60a8\u5199\u4e0b\u7b2c\u4e00\u884cimport tensorflow\u5f00\u59cb&#xff0c;\u60a8\u4fbf\u4e0d\u518d\u4ec5\u4ec5\u662f\u77e5\u8bc6\u7684\u6d88\u8d39\u8005&#xff0c;\u800c\u662f\u4e00\u4f4d\u6f5c\u5728\u7684\u3001\u80fd\u591f\u7528\u4ee3\u7801\u5851\u9020\u672a\u6765\u7684\u521b\u9020\u8005\u3002\u60a8\u6240\u638c\u63e1\u7684&#xff0c;\u5df2\u4e0d\u518d\u662f\u4e00\u95e8\u5355\u7eaf\u7684\u6280\u672f&#xff0c;\u800c\u662f\u4e00\u79cd\u5f3a\u5927\u7684\u3001\u80fd\u591f\u653e\u5927\u667a\u6167\u3001\u81ea\u52a8\u5316\u51b3\u7b56\u3001\u751a\u81f3\u521b\u9020\u51fa\u5168\u65b0\u4e8b\u7269\u7684\u201c\u529b\u91cf\u201d\u3002<\/p>\n<p>\u8fd9\u80a1\u529b\u91cf&#xff0c;\u5728\u60a8\u624b\u4e2d&#xff0c;\u53ef\u4ee5\u662f\u4e00\u4e2a\u5c0f\u5c0f\u7684\u5b87\u5b99\u3002<\/p>\n<p>\u60a8\u53ef\u4ee5\u7528\u5b83&#xff0c;\u53bb\u8bc6\u522b\u4e00\u7247\u53f6\u5b50\u7684\u8109\u7edc&#xff0c;\u8046\u542c\u4e00\u53ea\u9e1f\u513f\u7684\u6b4c\u5531&#xff1b;\u60a8\u53ef\u4ee5\u8ba9\u5b83&#xff0c;\u53bb\u8bfb\u61c2\u6d69\u5982\u70df\u6d77\u7684\u5178\u7c4d&#xff0c;\u7ffb\u8bd1\u9694\u9602\u5f7c\u6b64\u7684\u8bed\u8a00&#xff1b;\u60a8\u53ef\u4ee5\u6d3e\u5b83&#xff0c;\u53bb\u9884\u6d4b\u5929\u4f53\u7684\u8fd0\u884c&#xff0c;\u6d1e\u5bdf\u5e02\u573a\u7684\u98ce\u4e91&#xff1b;\u60a8\u751a\u81f3\u53ef\u4ee5\u6559\u5b83&#xff0c;\u53bb\u751f\u6210\u68b5\u9ad8\u7684\u661f\u7a7a&#xff0c;\u8c31\u5199\u5df4\u8d6b\u7684\u4e50\u7ae0\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u4efb\u4f55\u5f3a\u5927\u7684\u529b\u91cf&#xff0c;\u90fd\u4f34\u968f\u7740\u540c\u6837\u6df1\u91cd\u7684\u8d23\u4efb\u3002<\/p>\n<p>\u5f53\u60a8\u672a\u6765\u6784\u5efa\u7684AI\u7cfb\u7edf&#xff0c;\u5f00\u59cb\u5f71\u54cd\u4e00\u4e2a\u4eba\u7684\u5de5\u4f5c\u3001\u4e00\u6b21\u533b\u7597\u7684\u8bca\u65ad\u3001\u4e00\u7b14\u8d37\u6b3e\u7684\u5ba1\u6279\u65f6&#xff0c;\u8bf7\u60a8\u52a1\u5fc5\u8bb0\u8d77\u6211\u4eec\u5728\u6700\u540e\u4e00\u7ae0\u7684\u63a2\u8ba8\u3002\u8bf7\u8ba9\u201c\u516c\u5e73\u3001\u900f\u660e\u3001\u9690\u79c1\u3001\u5b89\u5168\u3001\u5411\u5584\u201d\u8fd9\u4e9b\u8bcd\u8bed&#xff0c;\u6210\u4e3a\u60a8\u4ee3\u7801\u4e16\u754c\u91cc\u6c38\u4e0d\u7184\u706d\u7684\u706f\u5854&#xff0c;\u6210\u4e3a\u60a8\u6240\u6709\u6280\u672f\u51b3\u7b56\u7684\u201c\u7b2c\u4e00\u6027\u539f\u7406\u201d\u3002<\/p>\n<p>\u8bf7\u6c38\u8fdc\u4fdd\u6301\u8c26\u900a\u3002\u5bf9\u77e5\u8bc6\u8c26\u900a&#xff0c;\u56e0\u4e3a\u5b66\u6d77\u65e0\u6daf&#xff1b;\u5bf9\u81ea\u7136\u8c26\u900a&#xff0c;\u56e0\u4e3a\u667a\u80fd\u7684\u5965\u79d8\u6211\u4eec\u4ecd\u77e5\u4e4b\u751a\u5c11&#xff1b;\u5bf9\u751f\u547d\u8c26\u900a&#xff0c;\u56e0\u4e3a\u6211\u4eec\u6240\u521b\u9020\u7684\u4e00\u5207&#xff0c;\u7ec8\u7a76\u662f\u4e3a\u4e86\u670d\u52a1\u4e8e\u4eba\u7c7b\u66f4\u7f8e\u597d\u7684\u751f\u6d3b\u3002<\/p>\n<p>\u8fd9\u672c\u4e66&#xff0c;\u5230\u8fd9\u91cc\u5c31\u8981\u7ed3\u675f\u4e86\u3002\u4f46\u60a8\u7684\u6545\u4e8b&#xff0c;\u624d\u521a\u521a\u5f00\u59cb\u3002<\/p>\n<p>\u5408\u4e0a\u4e66\u672c&#xff0c;\u4e16\u754c\u4fbf\u662f\u60a8\u66f4\u5e7f\u9614\u7684\u8bfe\u5802\u3002\u53bb\u9605\u8bfb\u3001\u53bb\u5b9e\u8df5\u3001\u53bb\u4ea4\u6d41\u3001\u53bb\u521b\u9020\u3001\u53bb\u72af\u9519\u3001\u53bb\u4fee\u6b63\u2026\u2026\u53bb\u5c06\u60a8\u5728\u8fd9\u91cc\u5b66\u5230\u7684\u4e00\u5207&#xff0c;\u5185\u5316\u4e3a\u60a8\u81ea\u5df1\u7684\u8840\u8089\u4e0e\u9aa8\u9abc&#xff0c;\u7136\u540e&#xff0c;\u53bb\u89e3\u51b3\u4e00\u4e2a\u60a8\u771f\u6b63\u5173\u5fc3\u7684\u3001\u54ea\u6015\u518d\u5fae\u5c0f\u4e0d\u8fc7\u7684\u95ee\u9898\u3002<\/p>\n<p>\u90a3\u4fbf\u662f\u8fd9\u8d9f\u65c5\u7a0b&#xff0c;\u6700\u7f8e\u7684\u5ef6\u7eed\u3002<\/p>\n<p>\u611f\u8c22\u60a8&#xff0c;\u6211\u4eb2\u7231\u7684\u8bfb\u8005\u670b\u53cb\u4eec&#xff0c;\u613f\u610f\u82b1\u8d39\u5b9d\u8d35\u7684\u65f6\u95f4&#xff0c;\u4e0e\u6211\u4eec\u4e00\u540c\u8d70\u8fc7\u8fd9\u6bb5\u5947\u5999\u7684\u65c5\u7a0b\u3002\u73b0\u5728&#xff0c;\u884c\u56ca\u5df2\u6ee1&#xff0c;\u524d\u8def\u5df2\u660e&#xff0c;\u53bb\u5427&#xff0c;\u53bb\u5f00\u542f\u5c5e\u4e8e\u60a8\u81ea\u5df1\u7684\u3001\u90a3\u5145\u6ee1\u65e0\u9650\u53ef\u80fd\u7684\u58ee\u4e3d\u822a\u7a0b\u3002<\/p>\n<p>\u6211\u4f1a\u5728\u65f6\u95f4\u7684\u8fd9\u4e00\u7aef&#xff0c;\u5e26\u7740\u5fae\u7b11&#xff0c;\u9759\u9759\u5730\u4e3a\u60a8\u795d\u798f\u3002<\/p>\n<p>\u518d\u4f1a&#xff01;<\/p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u76ee\u5f55<br \/>\n\u7b2c\u4e00\u90e8\u5206&#xff1a;\u57fa\u7840\u7bc7 \u2014\u2014 \u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3<br \/>\n\u7b2c1\u7ae0&#xff1a;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u4e4b\u65c5<br \/>\n1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb\u30021.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174\u30021.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u751f\u6001\u3001\u793e\u533a\u4e0e\u54f2\u5b66\u7684\u7edf\u4e00\u30021.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66\u30021.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90\u3002<br \/>\n\u7b2c<\/p>\n","protected":false},"author":2,"featured_media":65298,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[187,81,524,50,188],"topic":[],"class_list":["post-65300","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-javascript","tag-python","tag-react-js","tag-50","tag-188"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/65300.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u76ee\u5f55 \u7b2c\u4e00\u90e8\u5206&#xff1a;\u57fa\u7840\u7bc7 \u2014\u2014 \u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3 \u7b2c1\u7ae0&#xff1a;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u4e4b\u65c5 1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb\u30021.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174\u30021.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u751f\u6001\u3001\u793e\u533a\u4e0e\u54f2\u5b66\u7684\u7edf\u4e00\u30021.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66\u30021.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90\u3002 \u7b2c\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/65300.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-01-24T11:32:38+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/01\/20260124113235-6974add34cef2.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"105 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/65300.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/65300.html\",\"name\":\"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-01-24T11:32:38+00:00\",\"dateModified\":\"2026-01-24T11:32:38+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/65300.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/65300.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/65300.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/65300.html","og_locale":"zh_CN","og_type":"article","og_title":"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u76ee\u5f55 \u7b2c\u4e00\u90e8\u5206&#xff1a;\u57fa\u7840\u7bc7 \u2014\u2014 \u5960\u5b9a\u667a\u6167\u7684\u57fa\u77f3 \u7b2c1\u7ae0&#xff1a;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u4e4b\u65c5 1.1 \u4eba\u5de5\u667a\u80fd\u3001\u673a\u5668\u5b66\u4e60\u4e0e\u6df1\u5ea6\u5b66\u4e60&#xff1a;\u6b63\u672c\u6e05\u6e90&#xff0c;\u7406\u89e3\u4e09\u8005\u5173\u7cfb\u30021.2 \u6df1\u5ea6\u5b66\u4e60\u7684\u201c\u524d\u4e16\u4eca\u751f\u201d&#xff1a;\u4ece\u8d6b\u5e03\u7406\u8bba\u5230\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174\u30021.3 \u4e3a\u4f55\u9009\u62e9Python&#xff1f;\u2014\u2014 \u751f\u6001\u3001\u793e\u533a\u4e0e\u54f2\u5b66\u7684\u7edf\u4e00\u30021.4 \u672c\u4e66\u7684\u7ed3\u6784\u4e0e\u5b66\u4e60\u8def\u5f84\u56fe&#xff1a;\u5982\u4f55\u9ad8\u6548\u5730\u5229\u7528\u8fd9\u672c\u4e66\u30021.5 \u5fc3\u6cd5\u603b\u7eb2&#xff1a;\u4fdd\u6301\u597d\u5947\u3001\u52e4\u4e8e\u5b9e\u8df5\u3001\u62e5\u62b1\u5f00\u6e90\u3002 \u7b2c","og_url":"https:\/\/www.wsisp.com\/helps\/65300.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-01-24T11:32:38+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/01\/20260124113235-6974add34cef2.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"105 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/65300.html","url":"https:\/\/www.wsisp.com\/helps\/65300.html","name":"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-01-24T11:32:38+00:00","dateModified":"2026-01-24T11:32:38+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/65300.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/65300.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/65300.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"Python\u6df1\u5ea6\u5b66\u4e60\uff1a\u4ece\u5165\u95e8\u5230\u5b9e\u6218"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/65300","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=65300"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/65300\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/65298"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=65300"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=65300"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=65300"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=65300"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}