{"id":7690,"date":"2017-02-25T01:00:00","date_gmt":"2017-02-24T17:00:00","guid":{"rendered":"http:\/\/swarma.org\/%swarma-paper%\/"},"modified":"2018-12-06T00:38:35","modified_gmt":"2018-12-05T16:38:35","slug":"%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e8%83%8c%e5%90%8e%e7%9a%84%e7%bb%9f%e8%ae%a1%e7%89%a9%e7%90%86-%e6%9c%89%e5%85%b3rnn%e4%b8%8e%e8%ae%b0%e5%bf%86%e6%a8%a1%e5%9e%8b-%e7%94%84%e6%85%a7","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=7690","title":{"rendered":"\u6df1\u5ea6\u5b66\u4e60\u80cc\u540e\u7684\u7edf\u8ba1\u7269\u7406\u2014\u6709\u5173RNN\u4e0e\u8bb0\u5fc6\u6a21\u578b | \u7504\u6167\u73b2"},"content":{"rendered":"
\n
<\/section>\n
\n
\n
\n
\n
\n

<\/section>\n
<\/section>\n
\n
\n
\n
\n
\n

 <\/span>\u8fd9\u662f\u4e00\u4e2a\u7cfb\u5217\u7684\u8bfb\u4e66\u7b14\u8bb0\uff0c\u4e3b\u8981\u56f4\u7ed5\u7740\u5173\u4e8eRNN\uff0cRL\u548cmemory\u4ee5\u53ca\u4ed6\u4eec\u7684\u5404\u79cd\u7ebf\u6027\u7ec4\u5408\u548c\u975e\u7ebf\u6027\u53e0\u52a0\u57282016\u5e74\u7684\u4e00\u4e9b\u6709\u610f\u601d\u7684\u8fdb\u5c55\uff0c\u5f53\u7136\u66f4\u591a\u7684\u56f4\u7ed5\u7740Bengio, Hinton, LeCun, DeepMind, Facebook\u7b49\u65b9\u9762\u7684\u5b9e\u9a8c\u8fdb\u5c55\uff0c\u548cHopfield\u4ee5\u53ca\u8bf8\u591a\u7edf\u8ba1\u7269\u7406superstars\u5bf9\u8fd9\u65b9\u9762\u5de5\u4f5c\u7684\u8d21\u732e\u3002<\/span><\/span><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n

\"\u96c6\u667a\"<\/div>\n
\n
\n
\n
\n
\n
\n

\u601d\u60f3\u78b0\u649e<\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n

\u9488\u5bf9\u8bfb\u4e66\u7b14\u8bb0\u7684\u5185\u5bb9\uff0c\u7504\u6167\u73b2\u4e5f\u63d0\u51fa\u4e86\u4e00\u4e2a\u95ee\u9898\uff1a<\/span><\/p>\n

Hinton\u5728\u53bb\u5e74\u7684\u6587\u7ae0\u4e2d\u63d0\u5230memory\u6709\u4e09\u79cd\uff0clong, short\u548cspiking\uff0c\u90a3\u4e48\uff0c\u4eca\u5929\u63a8\u9001\u7684\u8fd9\u4e9b\u6a21\u578b\u4e2d\uff0c\u6709\u54ea\u4e2a\u589e\u52a0\u4e86memory\u7684\u5bb9\u91cf\uff1f\u5982\u679c\u6709\uff0c\u589e\u52a0\u7684\u54ea\u79cd\uff1f\u4e3a\u4ec0\u4e48\uff1f<\/span><\/strong><\/span><\/p>\n

\u5927\u5bb6\u53ef\u4ee5\u5e26\u7740\u95ee\u9898\u9605\u8bfb\u6587\u7ae0\uff0c\u6b22\u8fce\u5728\u7559\u8a00\u533a\u5199\u4e0b\u81ea\u5df1\u7684\u601d<\/strong>\u8003<\/strong>\uff0c\u6240\u6709\u7ed9\u51fa\u7b54\u6848\u5e76\u88ab\u91c7\u7eb3\u7684\u670b\u53cb<\/strong>\u5c06\u83b7\u5f97\u96c6\u667a\u7684\u7eaa\u5ff5T\u6064\u4e00\u4ef6\u54e6\uff08T\u6064\u4ece\u4e0d\u5bf9\u5916\u51fa\u552e\uff09~<\/span><\/p>\n

\n
\n


<\/p>\n

\n
\n

1\u3001<\/span><\/strong><\/inherit><\/p>\n

Curriculum Learning & RNN<\/span><\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n


<\/p>\n


<\/p>\n<\/section>\n<\/section>\n

 Curriculum Learning\u5176\u5b9e\u662fBengio\u57282009\u5e74\u7684ICML\u7684\u6587\u7ae0\u201cCurriculum Learning\u201d\u5c31\u63d0\u51fa\u7684\u4e00\u4e2a\u6982\u5ff5\uff0c\u6700\u8fd1\u51e0\u5e74\u4e00\u76f4\u88ab\u5f53\u6210\u4e00\u4e2a\u5904\u7406\u6570\u636e\u96c6\u7684trick\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u5404\u4e2a\u6a21\u578b\u4e2d\u53d1\u6325\u4f5c\u7528\uff0c\u5305\u62ec2016\u5e74RNN\u7684\u5404\u79cd\u6709\u8da3\u7ed3\u679c\u3002\u672c\u6b21\u63a8\u8350\u548c\u5206\u4eab\u4e3b\u8981\u662f\u9488\u5bf9\u4e0b\u9762\u51e0\u7bc7\u6587\u7ae0\uff1a<\/span>
<\/span><\/p>\n

\n

Knowledge Matters: Importance of Prior Information for Optimization \uff08Bengio\uff09<\/span><\/strong><\/span><\/p>\n

Synthesizing the preferred inputs for neurons in neural networks via deep generator networks (NIPS\uff0c2016)<\/span><\/strong><\/span><\/p>\n

Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space (Bengio<\/span><\/strong>)<\/span><\/span><\/p>\n<\/blockquote>\n

\u4ee5\u53ca\u4e0b\u9762\u8fd9\u7bc7\u5bf9\u4e0a\u9762\u6587\u7ae0\u7684\u5e94\u7528\u7684\u6587\u7ae0\uff1a<\/span><\/p>\n

Visualizing and Understanding Curriculum Learning for Long Short-Term Memory Networks \uff08By Carnegie Mellon University\uff09<\/span><\/p>\n

\u5728\u5206\u4eab\u6587\u7ae0\u4e4b\u524d\uff0c\u5148\u7b80\u5355\u8bf4\u4ec0\u4e48\u662fCurriculum Learning.<\/strong><\/span><\/p>\n


<\/strong><\/span><\/p>\n

\u6700\u5f00\u59cb\u63d0\u5230\u8fd9\u4e2a\u8bad\u7ec3\u65b9\u6cd5\u7684\u662f\u4e3a\u4e86\u89e3\u51b3\u975e\u51f8\u4f18\u5316\u95ee\u9898\u4e2d\u7684local minima\uff0c\u60f3\u6cd5\u5176\u5b9e\u4e5f\u5f88\u7b80\u5355\uff0c\u57fa\u4e8e\u4ed6\u7684\u4e00\u4e2a\u8ba4\u77e5\uff1atraining set\u4e2d\u7684\u6570\u636e\u5bf9\u4e8e\u76ee\u6807\u51fd\u6570\u7684\u8d21\u732e\u4e0d\u4e00\u6837\uff0c\u7b80\u5355\u8bf4\uff0c\u4f60\u53ef\u80fd\u53ea\u77e5\u9053\u4e00\u90e8\u5206\u7684\u6570\u636e\u7684\u6807\u7b7e\uff0c\u6216\u8005\u77e5\u9053\u67d0\u4e00\u4e2acluster\u7684\u7279\u70b9\uff08\u6bd4\u5982\u662fnoiseless\u7684\u6216\u8005\u975e\u5e38noisy\u90fd\u53ef\u4ee5\u6210\u4e3a\u7279\u70b9\uff09\uff0c\u800c\u672c\u6765\u8fd9\u4e9beasy samples\u662f\u5f88\u597d\u8bad\u7ec3\u7684\uff0c\u4f46\u5982\u679c\u5c06\u6240\u6709\u7684\u6570\u636e\u4e00\u8d77\u6765\u5206\u6210batch\u8bad\u7ec3\uff0c\u5c31\u4f1a\u8ba9\u8fd9\u4e9b\u5bb9\u6613\u8bad\u7ec3\u7684\u7279\u70b9\u53d8\u7684\u5f88\u5e73\u5eb8\u3002\u4e8e\u662f\uff0c\u4ed6\u4eec\u5c31\u628atraining set\u4e2d\u7684\u6570\u636e\u6309\u7167\u81ea\u5df1\u5b9a\u4e49\u7684\u96be\u5ea6\u7b49\u7ea7\u5206\u6210q\u7c7b\uff0c\u4e00\u7c7b\u4e00\u7c7b\u7684\u8bad\u7ec3\uff0c\u4e0a\u4e00\u7c7b\u8bad\u7ec3\u7684\u7ed3\u679c\u4f5c\u4e3a\u4e0b\u4e00\u7c7b\u8bad\u7ec3\u7684\u521d\u503c\uff0c\u5c31\u76f8\u5f53\u4e8e\u7528easy samples\u7684training \u7ed3\u679c\u6765\u6559\u4e0b\u4e00\u4e2a\u7b49\u7ea7\u7684samples\u4e00\u6837\uff0c\u4e5f\u56e0\u6b64\u5b9a\u4e49\u7684curriculum.<\/span><\/p>\n


<\/span><\/p>\n

\u5f88\u7b80\u5355\u7684\u53d8\u5316\uff0c\u4f46\u662f\u8bad\u7ec3\u7ed3\u679c\uff0c\u5374\u771f\u7684\u8df3\u8fc7\u4e86local minima\uff0c\u4f8b\u5982\uff0c\u4e0b\u56fe\u5c31\u662f\u7528wiki\u4e0a\u7684\u6587\u672c\u505a\u7684\u8bad\u7ec3\u7ed3\u679c\uff08\u84dd\u8272\u7684\u662f\u8868\u793a\u7528\u4e86curriculum\uff09\u3002<\/span><\/p>\n

\"\u96c6\u667a\" <\/span><\/p>\n

\u5176\u5b9e\u8fd9\u6837\u7684\u65b9\u6cd5\u662f\u7edf\u8ba1\u7269\u7406\u5b66\u5bb6\u5e38\u5e38\u4f7f\u7528\u7684\uff08\u7269\u7406\u7cfb\u7684\u5b69\u7eb8\u600e\u4e48\u4f1a\u4e0d\u6bcf\u6b21\u90fd\u5e26\u4e0a\u7269\u7406\u4e0a\u7684\u65b9\u6cd5\u5462\uff0c\u90a3\u4e0d\u662f\u6211\u7684\u98ce\u683c\u3002O(\u2229_\u2229)O\uff09 \u4f8b\u5982\u5728\u6587\u7ae0\u201cSolving the inverse Ising problem by mean-field methods in a clustered phase space with many states\u201d\u201cInference and learning in sparse systems with multiple states\u201d\u7b49\u89e3\u51b3\u7ec4\u5408\u4f18\u5316\u95ee\u9898\u7684\u65f6\u5019\uff0c\u90fd\u662f\u5148\u5c06data\u5206\u6210\u4e0d\u540c\u7684cluster\uff0c\u7136\u540e\uff0c\u5728\u6bcf\u4e2acluster\u4e2d\u9010\u4e00\u6c42\u89e3\u3002\u540c\u65f6\uff0c\u4ed6\u4eec\u8bc1\u5b9e\u4e86\uff0c\u7528\u8fd9\u6837\u7684\u65b9\u6cd5\u6765\u505a\u7ec4\u5408\u4f18\u5316\u662f\u53ef\u4ee5\u5f97\u5230\u6574\u4e2adataset\u4e00\u8d77\u5904\u7406\u65f6\u5f97\u4e0d\u5230\u7684\u7269\u7406\u73b0\u8c61\u3002<\/span><\/p>\n

 <\/span><\/p>\n

\u90a3\u8fd9\u6837\u6709\u610f\u4e49\u7684trick\uff0c\u5728RNN\u4e0a\u6709\u4ec0\u4e48\u4f53\u73b0\u5462\uff1f\u4e0b\u9762\u6211\u4eec\u4e3e\u7684\u4e09\u7bc7\u6587\u7ae0\u6709\u4e24\u7bc7\u662fBengio\u7ec4\u57282016\u5e74\u7684\u5de5\u4f5c\uff0c\u53e6\u5916\u4e00\u7bc7\u662f2016\u5e74nips\u4e0a\u7684\u6587\u7ae0\u3002<\/span><\/p>\n

<\/span><\/p>\n

\n
\n
\n


<\/p>\n

\n
\n

2\u3001<\/span><\/strong><\/inherit><\/p>\n

Knowledge Matters: Importance of Prior Information for Optimization \uff08Bengio\uff09<\/span><\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n

\u8fd9\u7bc7\u6587\u7ae0\u7684\u521d\u8877\u8bf4\u5f88\u591astate-of-art\u7684\u7b97\u6cd5\u90fd\u4e0d\u80fdgeneralize\uff0c\u4e8e\u662f\uff0c\u4ed6\u4eec\u501f\u7528\u4e86curriculum\u7684\u60f3\u6cd5\uff0c\u4e0d\u8fc7\uff0c\u8fd9\u6b21\u4e0d\u662f\u8981\u5c06\u6570\u636e\u96c6\u5206\u6210\u82e5\u5e72\u90e8\u5206\uff0c\u800c\u662f\u5e72\u8106\uff0c\u901a\u8fc7\u5f15\u5165intermediate concept\u5c06\u76ee\u6807\u5206\u6210\u4e24\u90e8\u5206\uff0c\u5206\u522b\u5bf9\u5e94\u4e00\u4e2a\u591a\u5c42\u7f51\u7edcp1nn\u548cp2nn\u3002\u8fd9\u6837\u7684\u65b9\u5f0f\u53ef\u4ee5\u5206\u6210with hints\u548cwithout hints\u4e24\u79cd\uff08\u5982\u4e0b\u56fe\uff09\uff1a\u5de6\u56fe\u662f\u8868\u793awith hints\uff0c\u8fd9\u662f\u8bf4\u7b2c\u4e00\u4e2a\u7f51\u7edcp1nn\u7684\u76ee\u6807\u51fd\u6570\u4e0d\u662f\u6700\u540e\u6211\u4eec\u8981\u5904\u7406\u7684\u76ee\u6807\u51fd\u6570\uff0c\u800c\u662f\u4e00\u4e2a\u548c\u6700\u540e\u7684\u76ee\u6807\u6709\u5173\u7684\u66f4\u7b80\u5355\u7684\u51fd\u6570\uff08\u4e5f\u5c31\u662f\u6240\u8c13\u7684intermediate concept\uff09\uff0c\u53f3\u56fe\u5219\u662fwithout hints\u7684\u7248\u672c\uff0c\u5c31\u662f\u8bf4\u8fd9\u4e24\u4e2a\u7f51\u7edcp1nn\u548cp2nn\u7684\u76ee\u6807\u51fd\u6570\u90fd\u662f\u6211\u4eec\u6700\u540e\u8981\u4f18\u5316\u7684\u90a3\u4e2a\u3002<\/span><\/p>\n

 <\/span><\/p>\n

\u4f46\u662f\uff0c\u5355\u5355\u8fd9\u6837\u5206\u6210\u4e24\u90e8\u5206\u8fd8\u4e0d\u591f\uff0c\u4ed6\u4eec\u8fd8\u8bf4\u8981\u5f15\u5165standardization layer\uff0c\u5176\u5b9e\u5c31\u662f\u8981\u5bf9hidden layer\u505a\u4e00\u4e2alayer normalization\uff08\u7c7b\u4f3c\u6211\u4eec\u5728\u672c\u79d1\u5b66\u6982\u7387\u8bba\u7684\u65f6\u5019\u90a3\u4e2a\u4e2d\u5fc3\u5316\u6210Gaussian\u5206\u5e03\u7684\u8fc7\u7a0b\uff09\uff0c\u5177\u4f53\u7684\u8ba1\u7b97\u8fc7\u7a0b\u5982\u4e0b\u3002<\/span><\/p>\n

\"\u96c6\u667a\" <\/span><\/p>\n

\u6587\u4e2d\u5206\u522b\u5bf9\u6bd4\u4e86with hints\u548cwithout hints\uff0cwith standardization layer\u548cwithout standardization\u7684\u533a\u522b\uff0c\u5f53\u7136\u90fd\u662fwith\u6bd4without\u7684\u8981\u597d\uff0c\u800c\u4e14\u6570\u636e\u96c6\u8d8a\u5927\uff08\u5c31\u662f\u8bf4\u660e\u91cc\u9762\u7684cluster\u53ef\u80fd\u8d8a\u591a\uff0c\u6570\u636e\u5bf9\u76ee\u6807\u51fd\u6570\u7684\u5f71\u54cd\u8d8a\u590d\u6742\u591a\u53d8\uff09\u6548\u679c\u8d8a\u597d\u3002\u4f8b\u5982\uff1a<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u4e0b\u56fe\u4e5f\u8bf4\u660e\u4e86standardization\u4f1a\u5728\u65e0\u8bba\u662ftest set\u8fd8\u662ftrain set\u4e0a\uff0c\u90fd\u8df3\u8fc7local minima\u3002\u800c\u4e14\uff0c\u6587\u7ae0\u4e2d\u8fd8\u5229\u7528\u7269\u7406\u4e2d\u7684power spectra\u6765\u89e3\u91ca\u4e86\u4e3a\u4ec0\u4e48standardization\u4ee5\u540e\u6548\u679c\u4f1a\u597d\uff1a\u56e0\u4e3a\u4f1a\u8ba9\u6570\u636e\u8fb9\u7684\u66f4\u52a0\u7684sparse\uff0c\u8fd9\u6837\u4e0d\u4ec5\u53ef\u4ee5\u5229\u7528\u77e9\u9635\u8ba1\u7b97\uff0c\u8ba9\u6574\u4e2a\u7684\u8bad\u7ec3\u8fc7\u7a0b\u53d8\u7684\u66f4\u52a0\u5bb9\u6613\u5b9e\u73b0\uff0c\u540c\u65f6\uff0c\u4e5f\u8ba9preferred samples\u66f4\u52a0work\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u5176\u5b9e\u8fd9\u6837\u7684 standardization\uff0c\u5728memory\u7684\u7814\u7a76\u4e2d\u4e5f\u4f1a\u4f7f\u7528\u3002\u7b49\u6211\u4eec\u63a5\u4e0b\u6765\u603b\u7ed3\u5230memory\u7684\u90e8\u5206\u7684\u65f6\u5019\uff0c\u518d\u7ee7\u7eedgo deep\u3002<\/span>
<\/p>\n

\n
\n
\n


<\/p>\n

\n
\n

3\u3001<\/span><\/strong><\/inherit><\/p>\n

<\/span>Synthesizing the preferred inputs for neurons in neural networks via deep generator networks (2016 NIPS)<\/span><\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n

\u8fd9\u7bc7\u6587\u7ae0\u7684\u4e3b\u8981\u6280\u672f\u53ef\u4ee5\u4ece\u4e0a\u4e00\u7bc7\u7684standardization\u4e2d\u83b7\u5f97\u7075\u611f\uff1a\u4e0d\u662f\u9700\u8981\u901a\u8fc7\u8fd9\u4e2a\u53bb\u4e2d\u5fc3\u5316\u6765\u8ba9\u90a3\u4e9b\u6709\u7279\u70b9\u7684samples\u66f4\u52a0work\u5417\uff1f\u90a3\u6211\u4f55\u4e0d\u5229\u7528\u4e00\u4e9beasy samples\u6765\u5408\u6210\u66f4\u597d\u7684preferred samples\uff08\u76f8\u5f53\u4e8e\u66f4\u65b0training set\uff09\u5462\uff1f\u4e8e\u662f\uff0c\u4ed6\u4eec\u5c31\u501f\u52a9\u4e86\u4e0b\u9762\u7684\u6a21\u578b\uff0c\u9996\u5148\u5f53\u7136\u4f9d\u7136\u662f\u5229\u7528curriculum learning\u7684\u60f3\u6cd5\u8bb2training set\u5206\u6210D1\uff0c\u2026\u2026\uff0cDn\u8fd9\u6837n\u4e2a\u90e8\u5206\uff0cD1\u662feasiest sampes\uff0c\u5229\u7528D1\u8bad\u7ec3\u4e00\u4e2a\u751f\u6210\u7f51\u7edc\u6765\u7ed9\u51faprior\uff0c\u501f\u52a9prior\u5408\u6210\u7684samples\u4f5c\u4e3a\u65b0\u7684D1’\uff0c\u8fd9\u4e2aD1’\u4f5c\u4e3a\u771f\u6b63\u7684\u7528\u4e8e\u8bad\u7ec3\u6211\u4eec\u60f3\u8bad\u7ec3\u7684\u7f51\u7edc\u7684training set, \u8bad\u7ec3\u7ed3\u679c\u4f5c\u4e3aD2\u7684\u521d\u503c\uff0c\u5bf9D2\u505a\u540c\u6837\u7684\u64cd\u4f5c\uff0c\u4f9d\u6b21\u7c7b\u63a8\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

 <\/span><\/p>\n

\u5728\u5229\u7528\u751f\u6210\u7f51\u7edc\u6765\u751f\u6210prior\u7684\u65f6\u5019\uff0c\u4e3b\u8981\u662f\u7528\u5230\u4e86acitivation maximization\uff0c\u8fd9\u4e2a\u65b9\u6cd5\u7684\u672c\u8d28\u5c31\u662f\u901a\u8fc7\u6c42activation function\u7684\u6700\u5927\u503c\u6765\u5bf9\u8f93\u5165\uff08\u4f8b\u5982\u4e00\u4e2aimage\uff09\u8fdb\u884c\u5408\u6210\uff0c\u6211\u4eec\u7684\u76ee\u6807\u51fd\u6570\u662f\uff1a<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u800c\u540e\u7528gradient ascent\u6765\u4f18\u5316\u8fd9\u4e2a\u51fd\u6570\uff0c\u5c31\u53ef\u4ee5\u4e86\u3002<\/span> <\/span><\/p>\n

\u5176\u5b9e\u8fd9\u4e2a\u8fc7\u7a0b\uff0c\u5c31\u5f88\u50cf\u4e00\u4e2aGAN\uff1a\u524d\u9762\u7684\u90a3\u4e2a\u751f\u6210\u7f51\u7edc\u5c31\u662fG\uff0c\u540e\u9762\u7684\u90a3\u4e2a\u6700\u7ec8\u8981\u8bad\u7ec3\u7684\u7f51\u7edc\u662fD\uff0c\u4e0d\u8fc7\uff0c\u548cGAN\u4e0d\u540c\u7684\u662f\uff0cGAN\u662f\u540c\u65f6\u8bad\u7ec3G\u548cD\u4e24\u90e8\u5206\uff0c\u7528D\u6765\u5224\u65adG\u751f\u6210\u7684\u7ed3\u679c\u597d\u4e0d\u597d\uff0c\u6700\u7ec8\u5f53D\u5df2\u7ecf\u5224\u65ad\u4e0d\u4e86\u7684\u65f6\u5019\u5c31\u505c\u6b62\uff08\u4e5f\u5c31\u662f\u8fbe\u5230\u4e86\u7eb3\u4ec0\u5747\u8861\u7684\u65f6\u5019\uff09\uff0c\u800c\u8fd9\u4e2a\u7f51\u7edc\u662f\u4e00\u5f00\u59cb\u5c31\u7ed9G\u90e8\u5206samples\uff0c\u800c\u540e\u7528G\u7684\u7ed3\u679c\u4f5c\u4e3aprior\u751f\u6210\u66f4\u591a\u7684preferred samples\uff0c\u4ee5\u6b64\u6765\u66f4\u65b0\u6574\u4e2atraining set\u3002\u4e3e\u6587\u4e2d\u7684\u4e00\u4e2a\u4f8b\u5b50\u6765\u8bf4\u660e\u8fd9\u6837\u7684\u7f51\u7edc\u53ef\u4ee5\u8fbe\u5230\u7684\u6548\u679c\uff1a<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n


<\/p>\n

<\/span><\/p>\n

\u4f46\u8fd9\u6837\u7684\u7f51\u7edc\u76f8\u6bd4\u4e8eGAN\u6765\u8bf4\u662f\u6709\u4f18\u52bf\u7684\uff0c\u6211\u4eec\u90fd\u77e5\u9053GAN\u4e2ddiscriminator\u4e5f\u5c31\u662f\u7f51\u7edcD\u5f88\u5f3a\uff0c\u56e0\u6b64\u4f1a\u5f15\u8d77\u5404\u79cd\u4e0d\u7a33\u5b9a\uff0c\u8bf8\u5982missing mode\u7b49\u7b49\uff0c\u4eba\u4eec\u7ecf\u5e38\u91c7\u7528\u52a0\u5165\u5404\u79cdregularization\u7684\u65b9\u5f0f\u6765\u8ba9GAN\u7a33\u5b9a\uff0c\u5176\u5b9e\u6709\u4e00\u4e2a\u5f88\u7b80\u5355\u7684trick\u5c31\u662f\u6211\u4eec\u4e0d\u8981\u5148\u8bad\u7ec3D\uff0c\u800c\u662f\u5148\u8bad\u7ec3G\uff0c\u591a\u8bad\u7ec3\u51e0\u4e2aiterations\u4ee5\u540e\uff0c\u518d\u5b8c\u6210\u4eceD\u5230G\u4e0d\u65ad\u5faa\u73af\u7684\u8fc7\u7a0b\uff0c\u90fd\u4f1a\u6bd4\u4e4b\u524d\u6548\u679c\u597d\u5f88\u591a\u3002\u90a3\u4e48\u8fd9\u7bc7\u6587\u7ae0\u4e2d\u7684\u7f51\u7edc\uff0c\u5c31\u662f\u5148\u8bad\u7ec3\u5bf9\u5e94G\u7684\u90a3\u4e2a\u751f\u6210\u7f51\u7edc\uff0c\u8fd9\u4e2a\u7f51\u7edc\u7a33\u5b9a\u4ee5\u540e\u518d\u8bad\u7ec3D\uff0c\u800c\u4e14\uff0c\u5582\u7ed9G\u7684input\u4e0d\u662f\u968f\u4fbf\u7684samples\uff0c\u662f\u6211\u4eec\u5148\u5229\u7528curriculum learning\u7684\u60f3\u6cd5\u5bf9training set\u5206\u5757\u4ee5\u540e\u7684\uff0c\u800c\u5582\u7ed9D\u7684\u662f\u66f4\u65b0\u7684training set\uff0c\u8fd9\u6837\u4e00\u6765\uff0c\u5c31\u660e\u663e\u4f1a\u6bd4\u539f\u59cb\u7684GAN\u66f4\u6709\u6548\u679c\u3002\u4e8b\u5b9e\u4e0a\uff0c\u57282016\u5e7412\u6708\u4efdNIPS\u5173\u4e8eGAN\u7684\u90a3\u4e2atutorial\u4e2d\uff0c\u5c31\u63d0\u5230\u4e86\u8fd9\u7bc7\u6587\u7ae0\u7684\u6a21\u578b\uff0c\u597d\u4e1c\u897f\u5927\u5bb6\u90fd\u4f1a\u6ce8\u610f\u5230\u54d2\uff01\uff01<\/span><\/p>\n

 <\/span><\/p>\n

\n
\n
\n
\n
\n

4\u3001<\/span><\/strong><\/inherit><\/p>\n

Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space (Bengio)<\/span><\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n<\/section>\n

\u8fd9\u7bc7\u6587\u7ae0\u5219\u5b8c\u5168\u662f\u4e3a\u4e86\u6539\u8fdb\u201csynthesizing preferred inputs\u201d\uff0c\u6539\u8fdb\u7684\u65b9\u5f0f\u5c31\u662f\u5149\u660e\u6b63\u5927\u7684\u79f0\u547c\u4e0a\u4e00\u4e2a\u7f51\u7edc\u4e2d\u4f5c\u4e3aD\u7684\u90a3\u4e2a\u5224\u522b\u7f51\u7edc\u4e3aconditional network\uff0c\u800c\u540e\uff0c\u5728\u4e4b\u524d\u7684\u751f\u6210\u7f51\u7edc\u4e0a\u52a0\u5165\u5404\u79cd\u6a21\u578b\uff0c\u800c\u4e0d\u518d\u662f\u7b80\u5355\u7684\u4e00\u4e2aupconvolutional\u4e86\u3002\u5f53\u7136\uff0c\u4ed6\u4eec\u8bf4\u4ed6\u4eec\u662f\u901a\u8fc7\u5f15\u5165additional prior in latent code\u6765\u63d0\u9ad8\u6837\u672c\u7684quality\u548cdiversity\u3002<\/span>
<\/p>\n<\/section>\n

 <\/span><\/p>\n

\u5982\u679csyhthesizing preferred inputs\u4e2d\u7684\u90a3\u4e2a\u6a21\u578b\u7528\u4e0b\u56feb\u6765\u8868\u793a\uff0c\u8fd9\u91cc\u4ed6\u4eec\u8bf4h\u8868\u793ahigh-level feature\uff0c\u5176\u5b9e\u5c31\u53ef\u4ee5\u7406\u89e3\u7c7b\u4f3c\u4e8e\u6210\u6211\u4eec\u4e00\u5f00\u59cb\u8bf4\u7684easiest samples\u7684feature\uff0c\u800c\u540e\uff0c\u751f\u6210\u4e86preferred samples x\uff0cx\u518d\u4f5c\u4e3aC\u7684\u8f93\u5165\u3002\u4f46\u662f\uff0c\u6211\u4eec\u6ce8\u610f\u5230\uff0c\u867d\u7136\u53e3\u53e3\u58f0\u58f0\u8bf4\u8fd9\u4e2a\u5229\u7528G\u751f\u6210prior\uff0c\u4f46\u662f\u7ed9G\u7684h\u5374\u6ca1\u6709\u4efb\u4f55\u7684prior\u3002\u6240\u4ee5\uff0c\u8fd9\u7bc7\u6587\u7ae0\u7684\u6539\u8fdb\u6a21\u578b\u4e00\u5171\u67094\u4e2a\uff0c\u5982\u4e0b\u56fe\u7684a,c,d,e\u6240\u793a\u3002<\/span><\/p>\n

\"\u96c6\u667a\" <\/span><\/p>\n

\u56fea\u65e0\u7591\u662f\u6700\u7b80\u5355\u7684\uff0c\u7b80\u5355\u5230\u4e86\u6839\u672c\u5c31\u6ca1\u6709\u7528\u90a3\u4e2aG\uff0c\u76f4\u63a5\u7528DAE\u6765\u4f5c\u4e3aG\uff0c\u800c\u56fec\u5219\u662f\u5bf9\u56fea\u7684\u6539\u8fdb\uff0c\u7528DAE\u6765\u5904\u7406h\uff0c\u7ed3\u679c\u5582\u7ed9G\uff0cG\u7684\u7ed3\u679c\u5582\u7ed9C\u3002\u6ce8\u610f\uff0c\u8fd9\u91cc\u7684eta\u8868\u793a\u7684\u662fnoise\u3002d\u5219\u662f\u5bf9C\u7684\u6539\u8fdb\uff0c\u662f\u5728x\u548ch\u5904\u5206\u522b\u90fd\u7528\u4e86\u4e00\u4e2aDAE\uff08\u5373E1\u548cE2\uff09\uff0ce\u5c31\u662f\u5e72\u8106\u8ba9h\u548cx\u90fd\u6ca1\u6709noise\uff0c\u53d8\u6210\u4e86\u548cd\u5bf9\u5e94\u7684noiseless\u7248\u672c\u3002<\/span><\/p>\n

 <\/span><\/p>\n

\u5148\u4e0d\u63d0\u566a\u58f0\u7684\u95ee\u9898\uff0c\u5148\u8bf4\u8fd9\u6837\u7684\u7f51\u7edc\u6240\u8fbe\u5230\u7684\u751f\u6210\u7ed3\u679c\u3002\u9996\u5148\uff0c\u8fd9\u6837\u7684\u7f51\u7edc\u8fbe\u5230\u7684\u6548\u679c\u548csynthesizing \u90a3\u7bc7\u6bd4\uff0c\u6709\u4ec0\u4e48\u533a\u522b\u5462\uff0c\u5982\u4e0b\u56fe\u6240\u793a\u3002\u5982\u679c\u4ece\u5de6\u81f3\u53f3\u6807\u53f71\u523012\u7684\u8bdd\uff0c1-3\u548c7-9\u90fd\u662f\u771f\u5b9e\u7684\u56fe\u7247\uff0c4-6\u662f\u7528synthesizing\u6765\u751f\u6210\u7684\u7ed3\u679c\uff0c10-12\u662f\u7528d\u6765\u751f\u6210\u7684\u7ed3\u679c\uff0c\u6240\u4ee5\uff0c\u4ed6\u4eec\u7684\u7ed3\u8bba\u662fsamples\u7684quality\u548cdiversity\u90fd\u6709\u6240\u63d0\u9ad8\u3002\u6ce8\u610f\uff0c\u8fd9\u91cc\u8bf4\u7684\u7ed3\u679c\u4e0d\u662f\u6700\u540e\u7684\u7ed3\u679c\uff0c\u662f\u901a\u8fc7G\u4ee5\u540e\uff0c\u5582\u7ed9C\u7684samples\uff08\u6bd5\u7adf\u53ea\u6539\u8fdb\u4e86\u524d\u9762\u4e00\u90e8\u5206\u561b\uff09\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

<\/span><\/p>\n

\u8fd9\u91cc\u9762\u63d0\u5230\u7684\u5404\u4e2a\u7f51\u7edc\uff08\u6bd4\u5982\uff09\uff0c\u4f5c\u7528\u662f\u7b49\u4ef7\u7684\u5417\uff1f\u5982\u4e0b\u56fe\u6240\u793a\uff08\u8fd9\u91cc\u76f4\u63a5\u628aG+C\u7684\u7ed3\u6784\u79f0\u547c\u4e3aGAN\u4e86\u3002\u3002o(\u256f\u25a1\u2570)o\uff09\u6211\u4eec\u53d1\u73b0d\u7684\u6548\u679c\uff08\u4e5f\u5c31\u662fjoint PPGN-h\uff09\u6548\u679c\u662f\u6700\u597d\u7684\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u800cd\u548ce\u7684\u6bd4\u8f83\u5462\uff0c\u6587\u4e2d\u76f4\u63a5\u8bf4\u4e86\uff0cd\u4e2d\u6709\u4e86noise\uff0c\u56e0\u6b64\u63d0\u9ad8\u4e86samples\u7684diversity\uff0c\u4f46\u662f\u548ce\u8fd9\u4e2anoiseless\u7248\u672c\u76f8\u6bd4\uff0c\u6ca1\u6709noise\u7684quality\u66f4\u9ad8\u3002\u8fd9\u4e2a\u7ed3\u8bba\u672c\u8eab\u662f\u5f88\u597d\u7406\u89e3\u7684\u3002\u60f3\u60f3gradient descent\/ascent\u548cstochastic gradient descent\/ascent\u7684\u533a\u522b\uff0c\u5f53\u6211\u4eec\u4f18\u5316\u7684\u51fd\u6570\u5f88flat\u7684\u65f6\u5019\uff0c\u4f18\u5316\u8fc7\u7a0b\u8fdb\u884c\u4e0d\u4e0b\u53bb\u4e86\uff0c\u6211\u4eec\u5c31\u5f15\u5165noise\u53d8\u6210stochastic\uff0c\u5f53\u51fa\u73b0local minima\u7684\u65f6\u5019\uff0c\u6211\u4eec\u4f9d\u7136\u8fd8\u662f\u5f15\u5165noise\u5c1d\u8bd5\u8df3\u51fa\u90a3\u4e2a\u5c40\u57df\u3002\u5176\u5b9e\u548cd\u4e0ee\u7684\u5bf9\u6bd4\u662f\u4e00\u6837\u7684\u542b\u4e49\u3002<\/span><\/p>\n

 <\/span><\/p>\n

\u518d\u4ecb\u7ecd\u4e00\u7bc72017\u5e74\u521a\u521a\u51fa\u7089\u7684prl (\u8fd9\u4e2a\u7edd\u5bf9\u662f\u7269\u7406\u754c\u7684top journal\u554a\uff01\uff01) \u7684\u6587\u7ae0\uff1a\u201cStochastic thermodynamics of learning\u201d\u3002\u8fd9\u7bc7\u6587\u7ae0\u66f4\u52a0\u5145\u5206\u5730\u4ece\u7edf\u8ba1\u70ed\u529b\u5b66\u7684\u89d2\u5ea6\u89e3\u91ca\u4e86noise\u5bf9\u4e8e\u5b66\u4e60\u8fc7\u7a0b\u7684\u5f71\u54cd\uff0c\u800c\u6587\u7ae0\u4e2d\u7684\u56fe2\u89e3\u91ca\u4e86learning\u7684\u6548\u679c\u968f\u7740noise\u7684\u53d8\u5316\uff0c \u76f4\u63a5\u6210\u4e3a\u4e86ppgn\u8fd9\u7bc7\u6587\u7ae0\u56fe13\u7684\u7269\u7406\u5b66\u4f50\u8bc1\uff01<\/span><\/p>\n


<\/span><\/p>\n

\n
\n
\n
\n
\n

5\u3001<\/span><\/strong><\/inherit><\/p>\n

Visualizing and Understanding Curriculum Learning for Long Short-Term Memory Networks \uff08By Carnegie Mellon University\uff09<\/span><\/strong><\/p>\n<\/section>\n

\n
<\/section>\n
<\/section>\n
<\/section>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n<\/section>\n


<\/p>\n<\/section>\n

\u8fd9\u7bc7\u6587\u7ae0\u5c31\u975e\u5e38\u7b80\u5355\u4e86\uff0c\u5176\u5b9e\u7b97\u662fcurriculum learning\u5bf9rnn\u6216\u8005lstm\u7684\u4e00\u4e2a\u5e94\u7528\u3002\u8fd9\u7bc7\u6587\u7ae0\u6700\u6709\u610f\u4e49\u7684\u7ed3\u679c\u662f\u5bf9\u6bd4\u4e86one-pass learn\u548cbaby learn\u7684\u533a\u522b\u3002<\/span> <\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u6240\u8c13\u7684one-pass\u5462\uff0c\u5c31\u662f\u6807\u51c6\u7684curriculum learning\uff0c\u5c06training set\u5206\u6210\u82e5\u5e72\u4e2a\u5b50\u96c6\uff0c\u9010\u4e00\u8bad\u7ec3\u3002\u53ef\u662f\uff0c\u8fd9\u5176\u5b9e\u662f\u4e0d\u7b26\u5408\u6211\u4eec\u4eba\u5728\u6210\u957f\u8fc7\u7a0b\u4e2d\u5b66\u4e60\u7684\u89c4\u5219\u7684\uff0c\u4f60\u4e0d\u4f1a\u5728\u770b\u5230\u4e00\u7bc7\u6587\u7ae0\u7684\u65f6\u5019\uff0c\u4ece\u5934\u5f00\u59cb\u67e5\u5b57\u5178\uff0c\u4f60\u4f1a\u5e26\u7740\u4e4b\u524d\u7684\u8bb0\u5fc6\uff0c\u5e76\u9010\u6e10\u7684\u4fee\u6b63\u8bb0\u5fc6\u3002\u4e8e\u662f\uff0c\u4ed6\u4eec\u63d0\u51fa\u4e86\u8fd9\u4e2ababy step\uff0c\u5176\u5b9e\u5c31\u662f\u8bad\u7ec3\u5b8cD1\u53bb\u8bad\u7ec3D2\u7684\u65f6\u5019\uff0c\u4e0d\u8bad\u7ec3D2\uff0c\u800c\u662f\u8bad\u7ec3D1\u548cD2\u7684\u5e76\uff0c\u4e0b\u4e00\u6b65\u4e0d\u8bad\u7ec3D3\uff0c\u662f\u8bad\u7ec3D1\uff0cD2\u548cD3\u7684\u5e76\uff0c\u4f9d\u6b21\u7c7b\u63a8\u3002\u6700\u540e\u8bad\u7ec3\u7684\u662f\u6574\u4e2atraining set\u3002<\/span><\/p>\n

\"\u96c6\u667a\" <\/span><\/p>\n

\u7b80\u5355\u8bf4\u660e\uff0c\u8fd9\u6837\u505a\u4e86\u4e4b\u540e\u7684\u6548\u679c\u662f\u4ec0\u4e48\uff0c\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u7ea2\u8272\u7684\u7ebf\u6761\u662fbaby step learn\u3002<\/span> <\/span><\/p>\n

 <\/span>\"\u96c6\u667a\"<\/p>\n

\u8fd9\u53ea\u662f\u5f88\u7b80\u5355\u7684curriculum learning\u554a\uff0c\u53ef\u4ee5\u60f3\u8c61\uff0c\u5982\u679c\u7528\u53e6\u5916\u7684\u7f51\u7edc\u4f5c\u4e3aprior\uff0c\u6548\u679c\u5e94\u8be5\u4f1a\u66f4\u597d\uff01\u5f53\u7136\uff0c\u8fc7\u591a\u7684parameters\u662f\u4e00\u4e2a\u8ba1\u7b97\u96be\u70b9\uff0c\u5e94\u8be5\u5c31\u6709\u7c7b\u4f3c\u4e8elayer normalization\uff0cstandardization layer\u7b49\u5404\u79cdtricks\uff01<\/span><\/p>\n

 <\/span><\/p>\n

\n
\n
\n
\n
\n
\n

\u63a8\u8350\u9605\u8bfb<\/strong><\/span><\/p>\n

\n
\n
<\/section>\n<\/section>\n<\/section>\n
\n


<\/p>\n

\n
\n
\n
\n

<\/section>\n
<\/section>\n
\n
\n
\n
\n
\n

\u7edf\u4e00\u4e4b\u8def\uff1a\u91cf\u5b50\u7ea0\u7f20\u3001\u65f6\u7a7a\u51e0\u4f55\u4e0e\u673a\u5668\u5b66\u4e60<\/span><\/a><\/p>\n

\u6211\u7684\u80a9\u4e0a\u662f\u98ce\uff0c\u98ce\u4e0a\u662f\u95ea\u70c1\u7684\u661f\u7fa4<\/span><\/a><\/p>\n

\u5982\u4f55\u638c\u63e1\u5b87\u5b99\uff1f<\/span><\/a>
What can we know from the given data? | \u539f\u521b\u7cbe\u9009<\/span><\/a><\/p>\n

\u671d\u95fb\u9053\uff0c\u5915\u5982\u4f55\uff1f | \u5c0f\u6728\u7403<\/span><\/a>
<\/p>\n

\u7f51\u7edc\u3001\u51e0\u4f55\u4e0e\u7269\u7406\uff1a\u5c42\u7ea7\u3001\u957f\u7a0b\u5173\u8054\u4e0e\u5c40\u57df\u5316 | \u7ae0\u5f66\u535a<\/span><\/a><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n
\n


<\/h1>\n

<\/span><\/p>\n

\n

\u25c6 <\/span>\u25c6 <\/span>\u25c6<\/span><\/span><\/strong><\/strong><\/span><\/h1>\n

\u96c6\u667a\u4ff1\u4e50\u90e8<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n
\n
\n
\n


<\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n
\n

<\/section>\n
<\/section>\n
<\/section>\n
\n
\n
\n
\n


<\/p>\n

\u82f1\u6587\u540d\uff1a Swarm Agents Club \uff0c
<\/span><\/p>\n

\u7b80\u79f0\u201c SA Club \u201d\uff0c\u6210\u7acb\u4e8e 2008 \u5e74 <\/span><\/p>\n

\u662f\u4e00\u4e2a\u4ece\u4e8b\u5b66\u672f\u7814\u7a76\u3001<\/span><\/p>\n

\u4eab\u53d7\u79d1\u5b66\u4e50\u8da3\u7684\u63a2\u7d22\u8005\u7684\u56e2\u4f53 \u3002<\/span><\/p>\n

\u529b\u56fe\u642d\u5efa\u4e00\u4e2a\u4e2d\u56fd\u7684<\/span><\/p>\n

 \u201c \u6ca1\u6709\u56f4\u5899\u7684\u7814\u7a76\u6240 \u201d\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n

\u96c6\u667aQQ\u7fa4\uff5c292641157<\/span><\/p>\n

\u5546\u52a1\u5408\u4f5c\uff5czhangqian@swarma.org<\/span><\/p>\n

\u6295\u7a3f\u8f6c\u8f7d\uff5cwangting@swarma.org<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\"\u96c6\u667a\"<\/p>\n

\u8ba9\u82f9\u679c\u7838\u5f97\u66f4\u731b\u70c8\u4e9b\u5427\uff01<\/span><\/p>\n

\n

\"\u96c6\u667a\"
<\/p>\n<\/section>\n

\u957f\u6309\u8bc6\u522b\u4e8c\u7ef4\u7801\uff0c\u5173\u6ce8\u96c6\u667aClub\uff0c<\/span><\/p>\n

\u8ba9\u6211\u4eec\u79bb\u79d1\u5b66\u63a2\u7d22\u66f4\u8fd1\u4e00\u6b65\u3002<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/p>\n

\n

\u59cb\u53d1\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7\uff1a \u96c6\u667a\u4ff1\u4e50\u90e8 <\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

 \u8fd9\u662f\u4e00\u4e2a\u7cfb\u5217\u7684\u8bfb\u4e66\u7b14\u8bb0\uff0c\u4e3b\u8981\u56f4\u7ed5\u7740\u5173\u4e8eRNN\uff0cRL\u548cmemory\u4ee5\u53ca\u4ed6\u4eec\u7684\u5404\u79cd\u7ebf\u6027\u7ec4\u5408\u548c\u975e\u7ebf\u6027\u53e0\u52a0\u57282016\u5e74\u7684\u4e00\u4e9b\u6709\u610f\u601d\u7684\u8fdb\u5c55\uff0c\u5f53\u7136\u66f4\u591a\u7684\u56f4\u7ed5\u7740Bengio, Hinton, LeCun, DeepMind, Facebook\u7b49\u65b9\u9762\u7684\u5b9e\u9a8c\u8fdb\u5c55\uff0c\u548cHopfield\u4ee5\u53ca\u8bf8\u591a\u7edf\u8ba1\u7269\u7406supe…<\/p>\n","protected":false},"author":1,"featured_media":7691,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[19],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7690"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7690"}],"version-history":[{"count":1,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7690\/revisions"}],"predecessor-version":[{"id":7710,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7690\/revisions\/7710"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/7691"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7690"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7690"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7690"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=7690"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}