{"id":47084,"date":"2023-12-14T12:21:55","date_gmt":"2023-12-14T04:21:55","guid":{"rendered":"https:\/\/swarma.org\/?p=47084"},"modified":"2023-12-14T12:21:55","modified_gmt":"2023-12-14T04:21:55","slug":"%e9%80%9a%e5%be%80%e5%85%b7%e8%ba%ab%e9%80%9a%e7%94%a8%e6%99%ba%e8%83%bd%ef%bc%9a%e5%a6%82%e4%bd%95%e8%ae%a9%e6%9c%ba%e5%99%a8%e4%bb%8e%e8%87%aa%e7%84%b6%e6%a8%a1%e6%80%81%e4%b8%ad%e5%ad%a6%e4%b9%a0","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=47084","title":{"rendered":"\u901a\u5f80\u5177\u8eab\u901a\u7528\u667a\u80fd\uff1a\u5982\u4f55\u8ba9\u673a\u5668\u4ece\u81ea\u7136\u6a21\u6001\u4e2d\u5b66\u4e60\u5230\u4e16\u754c\u6a21\u578b\uff1f"},"content":{"rendered":"
\n
\n
\n
\n

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

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

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

\u5bfc\u8bed<\/strong><\/p>\n<\/section>\n<\/section>\n

\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\u76ee\u524d\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u867d\u7136\u5f3a\u5927\uff0c\u5374\u5e38\u5e38\u7f3a\u4e4f\u4eba\u7c7b\u548c\u52a8\u7269\u7684\u5e38\u8bc6\u548c\u7075\u6d3b\u63a8\u7406\u80fd\u529b\u3002\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u7684\u6700\u7ec8\u76ee\u6807\uff0c\u662f\u8ba9\u673a\u5668\u80fd\u591f\u50cf\u4eba\u7c7b\u4e00\u6837\u601d\u8003\u3001\u5b66\u4e60\u5e76\u89e3\u51b3\u5404\u79cd\u4efb\u52a1\uff0c\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u3002<\/span><\/strong>\u672c\u6587\u4f5c\u8005\u6c88\u9a6c\u6210\u535a\u58eb\u662f\u96c6\u667a\u4ff1\u4e50\u90e8\u300cAGI\u8bfb\u4e66\u4f1a<\/a>\u300d\u53d1\u8d77\u4eba<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong>\u4e4b\u4e00\uff0c\u76ee\u524d\u81f4\u529b\u4e8e\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u7814\u7a76\uff0c\u4ed6\u8ba4\u4e3a\uff0c\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u5173\u952e\u95ee\u9898\u5728\u4e8e\uff0c\u5982\u4f55\u4f7f\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u4ece\u81ea\u7136\u6a21\u6001\u4e2d\u5b66\u4e60\u5230\u5173\u4e8e\u4e16\u754c\u7684\u7ed3\u6784\u5316\u5c42\u7ea7\u5316\u62bd\u8c61\uff0c\u5373\u4e16\u754c\u6a21\u578b\uff08world model\uff09\u3002\u672c\u6587\u4ecb\u7ecd\u4e86\u57fa\u4e8e\u76ee\u524d\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\u7684\u5177\u8eab\u667a\u80fd\u7814\u7a76\u8fd1\u5e74\u6765\u7684\u8fdb\u5c55\u3001\u9762\u4e34\u7684\u6311\u6218\uff0c\u5e76\u63a2\u8ba8\u4e86\u672a\u6765\u53ef\u80fd\u7684\u6539\u8fdb\u65b9\u5411\uff0c\u4f7f\u5f97\u5177\u8eab\u667a\u80fd\u4f53\u80fd\u591f\u5728\u4e0d\u9700\u8981\u4eba\u7c7b\u5e2e\u52a9\u7684\u6761\u4ef6\u4e0b\u81ea\u4e3b\u5b66\u4e60\u4e16\u754c\u6a21\u578b\u5e76\u6d8c\u73b0\u51fa\u901a\u7528\u5b66\u4e60\u80fd\u529b\u3002<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/section>\n

<\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/section>\n
\u7814\u7a76\u9886\u57df\uff1a\u901a\u7528<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong>\u5177\u8eab\u667a\u80fd\uff0c\u5f3a\u5316\u5b66\u4e60\uff0c\u4e16\u754c\u6a21\u578b\uff0c\u5c42\u7ea7\u5316\u62bd\u8c61\uff0c\u6d8c\u73b0<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/section>\n<\/section>\n
\n
<\/strong><\/section>\n<\/section>\n
\n
<\/strong><\/strong><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
<\/section>\n<\/section>\n
\n
<\/strong><\/section>\n
\u6c88\u9a6c\u6210<\/span><\/span> <\/span>| \u4f5c\u8005<\/strong><\/span><\/strong><\/section>\n
<\/strong><\/strong><\/span><\/strong><\/section>\n

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


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
\u76ee\u5f55<\/strong><\/strong><\/section>\n
\n

1. \u4e3b\u6d41\u7684\u5177\u8eab\u667a\u80fd\u8bad\u7ec3\u65b9\u6cd5\u9762\u4e34\u7684\u6311\u6218<\/span><\/strong><\/strong><\/p>\n

2. \u8fd1\u671f\u7a81\u7834\uff1a\u8bed\u8a00\u4efb\u52a1\u9884\u8bad\u7ec3\u5e26\u6765\u7ec4\u5408\/\u7cfb\u7edf\u6027\u6cdb\u5316\u80fd\u529b<\/span><\/strong><\/p>\n

3. \u76ee\u524d\u7684\u5177\u8eab\u667a\u80fd\u7cfb\u7edf\u8ddd\u79bb\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u6839\u672c\u6027\u5dee\u5f02<\/span><\/strong><\/p>\n

4. \u4efb\u52a1\u6307\u5b9a\u95ee\u9898\uff0c\u5e38\u8bc6\uff0c\u81ea\u7136\u62bd\u8c61\u548c\u4e16\u754c\u6a21\u578b<\/span><\/strong><\/p>\n

5. \u5728\u673a\u5668\/\u7edf\u8ba1\u5b66\u4e60\u6846\u67b6\u4e0b\u5982\u4f55\u5b66\u4e60\u4e16\u754c\u6a21\u578b<\/span><\/strong><\/p>\n

6. \u603b\u7ed3<\/span><\/strong><\/p>\n

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


<\/p>\n


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

\u5177\u8eab\u667a\u80fd\uff0c\u901a\u4fd7\u6765\u8bf4\uff0c\u662f\u6307\u7814\u7a76\u5728\u73af\u5883\u4e2d\u5177\u6709\u5b9e\u4f53\u7684\u667a\u80fd\u4f53<\/span>\uff08\u5982\u73b0\u5b9e\u6216\u4eff\u771f\u73af\u5883\u4e0b\u7684\u673a\u5668\u4eba\uff0c\u80fd\u591f\u76f4\u63a5\u4e0e\u73af\u5883\u8fdb\u884c\u7269\u7406\u4ea4\u4e92\uff09<\/span>\u5982\u4f55\u901a\u8fc7\u4e0e\u73af\u5883\u7684\u4ea4\u4e92\u6765\u53d6\u5f97\u8ba4\u77e5\u80fd\u529b\uff0c\u5b66\u4e60\u5e76\u638c\u63e1\u65b0\u6280\u80fd\u65b0\u77e5\u8bc6\u7684\u4e00\u4e2a\u4eba\u5de5\u667a\u80fd\u7684\u5b50\u9886\u57df\u3002<\/span><\/p>\n


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


<\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

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

1. \u4e3b\u6d41\u7684\u5177\u8eab\u667a\u80fd\u8bad\u7ec3\u65b9\u6cd5\u9762\u4e34\u7684\u6311\u6218<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


<\/p>\n

\u5728\u6df1\u5ea6\u5b66\u4e60\u8303\u5f0f\u4e0b\uff0c\u5177\u8eab\u667a\u80fd\u7684\u7814\u7a76\u4e3b\u8981\u96c6\u4e2d\u5728\u6a21\u4eff\u5b66\u4e60<\/span>(Imitation Learning)<\/span>\u548c\u5f3a\u5316\u5b66\u4e60<\/span> (Reinforcement Learning, RL) <\/span>\u4e24\u5927\u5757\u3002\u6a21\u4eff\u5b66\u4e60<\/strong>\u901a\u8fc7\u91c7\u96c6\u7279\u5b9a\u4efb\u52a1\u7684\u8f68\u8ff9\u6570\u636e\u96c6\u5e76\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u6765\u62df\u5408\u72b6\u6001<\/span>(state)<\/span>\u6216\u89c2\u6d4b<\/span>(observation\uff0c\u5982\u7b2c\u4e00\u89c6\u89d2\u7684\u56fe\u50cf) <\/span>\u7684\u65f6\u95f4\u5e8f\u5217\u5230\u52a8\u4f5c<\/span> (action) <\/span>\u7684\u6620\u5c04\u6765\u5b9e\u73b0\u6280\u80fd\u7684\u5b66\u4e60\uff0c\u4e00\u822c\u6765\u8bf4\u6570\u636e\u91c7\u96c6\u6210\u672c\u8f83\u9ad8\u3002\u5f3a\u5316\u5b66\u4e60<\/strong>\u5219\u662f\u901a\u8fc7\u8ba9\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u76f4\u63a5\u4ea4\u4e92\uff0c\u5728\u4ea4\u4e92\u7684\u8fc7\u7a0b\u4e2d\u4f18\u5316\u9884\u5148\u5b9a\u4e49\u597d\u7684\u4e0e\u7279\u5b9a\u4efb\u52a1\u76f8\u5173\u7684\u5956\u52b1\u51fd\u6570<\/span> (reward function) <\/span>\u6765\u5b66\u4e60\u65b0\u6280\u80fd\uff0c\u4e00\u822c\u6765\u8bf4\u8bbe\u8ba1\u5956\u52b1\u51fd\u6570\u9700\u8981\u53cd\u590d\u8fed\u4ee3\uff0c\u4e14\u5f3a\u5316\u5b66\u4e60\u7684\u6837\u672c\u6548\u7387<\/span> (sample-efficiency)<\/span> \u76f8\u6bd4\u4e8e\u6a21\u4eff\u5b66\u4e60\u6765\u8bf4\u4f1a\u4f4e\u5f97\u591a\u3002<\/span><\/p>\n


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

\u65e9\u671f\u7684\u5177\u8eab\u667a\u80fd\u7814\u7a76\u53d1\u73b0\u8981\u4f7f\u5f97\u667a\u80fd\u4f53\u5b66\u5230\u7684\u7b56\u7565 <\/span>(policy)<\/span> \u6cdb\u5316\u5230\u5373\u4fbf\u4e0e\u8bad\u7ec3\u6570\u636e\u975e\u5e38\u76f8\u4f3c\u7684\u4efb\u52a1\u4e5f\u662f\u4e00\u4ef6\u5f02\u5e38\u56f0\u96be\u7684\u4e8b\u60c5\uff0c\u6bd4\u5982\u5728\u201c\u628a\u7269\u4f53\u4ece\u684c\u5b50\u4e2d\u95f4\u63a8\u5230\u53f3\u4e0a\u89d2\u201d\u8fd9\u4e2a\u8bad\u7ec3\u96c6\u4e0a\u5f97\u5230\u7684\u7b56\u7565\u7adf\u7136\u6ca1\u6709\u529e\u6cd5\u76f4\u63a5\u7528\u4e8e\u89e3\u51b3\u201c\u628a\u7269\u4f53\u4ece\u684c\u5b50\u4e2d\u95f4\u63a8\u5230\u53f3\u4e0b\u89d2\u201d\u8fd9\u4e2a\u4efb\u52a1\uff0c\u76f4\u89c9\u4e0a\u6765\u8bf4\u5176\u80cc\u540e\u7684\u539f\u56e0\u662f\u684c\u5b50\u7684\u53f3\u4e0a\u89d2\u548c\u53f3\u4e0b\u89d2\u5728\u56fe\u50cf\u5c42\u9762\u4e0a\u5e76\u4e0d\u76f8\u540c\uff0c\u800c\u795e\u7ecf\u7f51\u7edc\u5e76\u6ca1\u6709\u6570\u636e\u6765\u5b66\u4e60\u5230\u8fd9\u4e24\u8005\u5728\u201c\u66f4\u52a0\u62bd\u8c61\u7684\u5c42\u9762\u4e0a\u662f\u76f8\u7c7b\u4f3c\u7684\u201d\u3002\u56e0\u6b64\u4e00\u4e2a\u5de8\u5927\u7684\u6311\u6218\u662f\u4efb\u52a1\u6570\u91cf\u7684\u7ec4\u5408\u7206\u70b8<\/strong>\uff1a\u53ef\u4ee5\u60f3\u8c61\u5982\u679c\u9700\u8981\u5b66\u4e60\u4ee5100\u79cd\u65b9\u5f0f\u64cd\u4f5c100\u7c7b\u7269\u4f53\u4f7f\u5176\u8fbe\u5230100\u79cd\u7279\u5b9a\u72b6\u6001\uff0c\u90a3\u4e48\u9700\u8981\u5b9a\u4e49100x100x100=\u4e00\u767e\u4e07\u79cd\u4e0d\u540c\u7684\u4efb\u52a1\uff0c\u9700\u8981\u6536\u96c6\u76f8\u5e94\u7684\u6570\u636e\u6765\u8bad\u7ec3\u4e00\u4e2a\u591a\u4efb\u52a1 <\/span>(multi-task) <\/span>\u7b56\u7565\uff0c\u8fd9\u5c31\u975e\u5e38\u4e0d\u73b0\u5b9e\u3002<\/span><\/p>\n


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


<\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

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

2. \u8fd1\u671f\u7a81\u7834\uff1a<\/strong><\/span><\/strong><\/strong><\/p>\n

\u8bed\u8a00\u4efb\u52a1\u9884\u8bad\u7ec3\u5e26\u6765\u7ec4\u5408\/\u7cfb\u7edf\u6027\u6cdb\u5316\u80fd\u529b<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


<\/p>\n

\u8fd9\u4e2a\u7ec4\u5408\u7206\u70b8\u7684\u56f0\u96be\u76f4\u52302021\u5e74\u5de6\u53f3\uff0c\u5b66\u672f\u754c\u4e00\u4e9b\u65e9\u671f\u5de5\u4f5c\u4e2d\u627e\u5230\u4e86\u6f5c\u5728\u7684\u89e3\u51b3\u65b9\u6848\u3002\u76f4\u89c9\u4e0a\u6765\u8bf4\u5176\u539f\u7406\u5c31\u662f\u8ba4\u8bc6\u5230\u8fd9\u4e9b\u4efb\u52a1\u5e76\u975e\u662f\u4e92\u76f8\u72ec\u7acb\u7684\uff0c\u800c\u662f\u4ee5\u4e00\u79cd\u7ec4\u5408<\/span>(compositional)<\/span>\u7684\u5f62\u5f0f\u4e92\u4e3a\u76f8\u5173\uff0c\u5e76\u4e14\u8fd9\u79cd\u7ec4\u5408\u76f8\u5173\u7684\u5f62\u5f0f\u4e0e\u5176\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\u5b58\u5728\u4e00\u5b9a\u7684\u201c\u540c\u6784\u6027\u201d\u3002<\/span><\/p>\n


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

Lu et al 21 [1]\u53d1\u73b0\u5728\u8bed\u8a00\u4efb\u52a1\u4e0a\u9884\u8bad\u7ec3\u8fc7\u7684Transformer\u7adf\u7136\u5728\u4e00\u7cfb\u5217\u770b\u4f3c\u5b8c\u5168\u4e0d\u76f8\u5173\u7684\u4efb\u52a1<\/span>\uff08\u89c6\u89c9\uff0c\u86cb\u767d\u8d28\u6298\u53e0\uff0c\u6570\u503c\u8ba1\u7b97\uff09<\/span>\u4e0a\u5fae\u8c03\u540e\uff0c\u53d6\u5f97\u6bd4\u76f4\u63a5\u8bad\u7ec3\u663e\u8457\u63d0\u9ad8\u7684\u6548\u679c\u3002\u8fd9\u4e00\u7ed3\u679c\u6697\u793a\u4e86\u8bed\u8a00\u4efb\u52a1\u4e2d\u5b58\u5728\u67d0\u79cd\u7ed3\u6784\u5bf9\u4e8e\u8bb8\u591a\u5176\u4ed6\u7c7b\u522b\u7684\u4efb\u52a1\u4e5f\u80fd\u901a\u7528\u3002Eric Jang et al [2]\u5728\u591a\u4efb\u52a1\u5b66\u4e60<\/span> (multi-task learning)<\/span> \u7684\u6846\u67b6\u4e0b\u5229\u7528\u8bed\u8a00\u5d4c\u5165<\/span> (language embedding)<\/span> \u4f5c\u4e3a\u4efb\u52a1\u5d4c\u5165<\/span> (task embedding) <\/span>\u6765\u5b66\u4e60\u673a\u5668\u4eba\u591a\u4efb\u52a1\u7b56\u7565\uff0c\u5c55\u793a\u4e86\u96f6\u6837\u672c\u6cdb\u5316\u7684\u53ef\u80fd\u6027\uff1a\u673a\u5668\u4eba\u8bad\u7ec3\u96c6\u4e2d\u7684\u4efb\u610f\u4e00\u6761\u6570\u636e\u4ece\u672a\u540c\u65f6\u5305\u542b\u201c\u74f6\u5b50\u201d\u548c\u201c\u6258\u76d8\u201d\uff0c\u4e5f\u6ca1\u6709\u88ab\u8bad\u7ec3\u8fc7\u201c\u5c06\u74f6\u5b50\u653e\u5165\u6258\u76d8\u201d\u8fd9\u4e2a\u4efb\u52a1\uff0c\u5374\u53ef\u4ee5\u5728\u6d4b\u8bd5\u96c6\u4e2d\u96f6\u6837\u672c\u6cdb\u5316\u89e3\u51b3\u8fd9\u4e2a\u4efb\u52a1\u3002\u8fd9\u4e2a\u7ed3\u679c\u8868\u660e\u5229\u7528\u8fd9\u79cd\u8bad\u7ec3\u65b9\u5f0f\uff0c\u6a21\u578b\u5b66\u5230\u4e86\u201c\u5c06A\u653e\u5165B\u201d\u8fd9\u6837\u4e00\u4e2a\u62bd\u8c61\u7684\u6982\u5ff5\uff0c\u8fd9\u4e0e\u8bed\u8a00\u7684\u7ec4\u5408\u6027\u5176\u5b9e\u662f\u76f8\u540c\u7684\uff1a\u6211\u4eec\u4ec5\u9700\u8981\u6709\u9650\u7684\u8bcd\u6c47\u548c\u8bed\u6cd5\u7ed3\u6784\u5c31\u53ef\u4ee5\u8fdb\u884c\u65e0\u9650\u79cd\u53ef\u80fd\u7684\u8868\u8fbe\uff0c\u6b63\u662f\u56e0\u4e3a\u6bcf\u79cd\u8bcd\u6c47\u548c\u8bed\u6cd5\u7ed3\u6784\u4e0d\u540c\u65b9\u5f0f\u7684\u7ec4\u5408\u90fd\u53ef\u4ee5\u88ab\u5b9a\u4e49\u6210\u4e00\u4e2a\u65b0\u7684\u6982\u5ff5\uff0c\u800c\u8fd9\u79cd\u7ec4\u5408\u65b9\u5f0f\u662f\u65e0\u7a77\u7684\u3002<\/span><\/p>\n


<\/p>\n

\u8fd1\u671f\u7684\u5177\u8eab\u667a\u80fd\u5927\u6a21\u578b<\/span> (foundation model) <\/span>\u5176\u5b9e\u4e5f\u662f\u6cbf\u7740\u8fd9\u4e2a\u65b9\u5411\u8fdb\u4e00\u6b65\u5730\u89c4\u6a21\u5316<\/span> (scale up)<\/span>\u3002 \u6bd4\u5982Google Deepmind\u7684RT-X[3]\uff0c\u91c7\u7528\u7684\u5c31\u662f\u5728\u5927\u89c4\u6a21\u8bed\u97f3\u6570\u636e\u96c6\u4e0a\u9884\u8bad\u7ec3\u8fc7\u7684Transformer\u4ee5\u53ca\u5728\u89c6\u89c9\u4efb\u52a1\u4e0a\u9884\u8bad\u7ec3\u8fc7\u7684ViT\u4f5c\u4e3abackbone\uff0c\u5728\u591a\u5f62\u6001<\/span>(multiple morphology) <\/span>\u7684\u5177\u8eab\u4efb\u52a1\u6570\u636e\u96c6\u4e0a\u5fae\u8c03\uff0c\u4ece\u800c\u89e3\u9501\u4e86\u96f6\u6837\u672c\uff0c\u7cfb\u7edf\u6027\u6cdb\u5316 [4] \u5230\u65b0\u4efb\u52a1\u7684\u80fd\u529b\uff0c\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u7f13\u89e3\u4e86\u5177\u8eab\u667a\u80fd\u6570\u636e\u91c7\u96c6\u6210\u672c\u9ad8\u6602\u7684\u95ee\u9898\u3002<\/span><\/p>\n


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


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

\n
\n
\n
\n
\n

<\/section>\n

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

3. \u76ee\u524d\u7684\u5177\u8eab\u667a\u80fd\u7cfb\u7edf<\/strong><\/span><\/strong><\/strong><\/p>\n

\u8ddd\u79bb\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u6839\u672c\u6027\u5dee\u5f02<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


<\/p>\n

\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u867d\u7136RT-X\u5c55\u793a\u4e86\u5728\u591a\u79cd\u673a\u5668\u4eba\u5f62\u6001\u4e0a\u5de5\u4f5c\u7684\u7ed3\u679c\uff0c\u4f46\u8fd9\u662f\u901a\u8fc7\u8f83\u4e3a\u5de5\u7a0b\u5316\u7684\u624b\u6bb5\u6765\u5b9e\u73b0\u7684\uff0c\u5177\u4f53\u6765\u8bf4\u6a21\u578b\u4ec5\u8f93\u51fa\u673a\u5668\u4eba\u672b\u7aef\u6267\u884c\u5668<\/span> (end effector)<\/span>\u7684\u8f68\u8ff9\uff0c\u800c\u4e2d\u95f4\u8fde\u63a5\u8f74\u7684\u8fd0\u52a8\u5219\u901a\u8fc7\u9006\u5411\u8fd0\u52a8\u5b66<\/span> (inverse kinematics)<\/span> \u4ee5\u95ed\u5408\u89e3\u7684\u5f62\u5f0f\u5f97\u5230\u3002\u672c\u8d28\u4e0a\u6765\u8bf4\u8fd9\u79cd\u65b9\u6cd5\u610f\u5473\u7740\u673a\u5668\u4eba\u7684\u5177\u4f53\u5f62\u6001\u5df2\u7ecf\u88ab\u4eba\u4e3a\u5730\u62bd\u8c61\u6389\u4e86\uff0c\u6a21\u578b\u5e76\u4e0d\u9700\u8981\u5173\u5fc3\u5177\u4f53\u5f62\u6001\uff0c\u5176\u5bfc\u81f4\u7684\u5c40\u9650\u6027\u662f\u6a21\u578b\u6ca1\u6709\u529e\u6cd5\u50cf\u4eba\u548c\u52a8\u7269\u4e00\u6837\u7075\u6d3b\u5730\u8fd0\u7528\u673a\u5668\u4eba\u7684\u8eab\u4f53\u7ec4\u6210\u90e8\u5206\u6765\u5b8c\u6210\u4efb\u52a1\uff0c\u6bd4\u5982\u8003\u8651\u4e00\u4e2a\u201c\u62ff\u8d77\u74f6\u5b50\u201d\u7684\u4efb\u52a1\uff0c\u4eba\u53ef\u4ee5\u7528\u624b\u6307\u6293\u8d77\uff0c\u6216\u8005\u7528\u624b\u638c\u63e1\u4f4f\uff0c\u751a\u81f3\u7528\u624b\u638c\u548c\u624b\u81c2\u5f2f\u6210\u89d2\u5ea6\u5939\u8d77\u74f6\u5b50\u3002\u8fd9\u79cd\u5bf9\u4e8e\u672c\u4f53\u7684\u8ba4\u77e5\u548c\u7075\u6d3b\u8fd0\u7528\u80fd\u529b\u662f\u76ee\u524d\u7684\u5177\u8eab\u667a\u80fd\u4f53\u505a\u4e0d\u5230\u7684\u3002<\/span><\/p>\n


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

\u53e6\u4e00\u65b9\u9762\uff0c\u76ee\u524d\u5177\u8eab\u667a\u80fd\u4f53\u89e3\u51b3\u7684\u7edd\u5927\u591a\u6570\u4efb\u52a1\u90fd\u662f\u5728\u89c6\u89c9\u80cc\u666f\u4e3a\u684c\u9762\u7684\u73af\u5883\u4e0b<\/span>(tabletop scenario)<\/span>\uff0c\u8fd9\u79cd\u7b80\u5355\u7684\u80cc\u666f\u5b9e\u9645\u4e0a\u5927\u5927\u7b80\u5316\u4e86\u5bf9\u4e8e\u89c6\u89c9\u8ba4\u77e5\u80fd\u529b\u7684\u8981\u6c42\uff0c\u5b9e\u9645\u5b66\u4e60\u5230\u7684\u7b56\u7565\u6ca1\u6709\u529e\u6cd5\u76f4\u63a5\u7528\u5230\u590d\u6742\u7684\u975e\u7ed3\u6784\u5316\u80cc\u666f\u7684\u73af\u5883\u4e0b\uff0c\u539f\u56e0\u662f\u76ee\u524d\u7684\u89c6\u89c9\u81ea\u76d1\u7763\u5b66\u4e60<\/span> (vision self-supervised learning) <\/span>\u8fd8\u6ca1\u80fd\u5b66\u4e60\u5230\u5bf9\u4e8e\u4e16\u754c\u7684\u7ed3\u6784\u5316\u8868\u5f81\uff0c\u4e0d\u5177\u5907\u8db3\u591f\u89c6\u89c9\u8ba4\u77e5\u80fd\u529b\uff0c\u800c\u8fd9\u4e5f\u662f\u5b9e\u73b0\u5177\u8eab\u901a\u7528\u667a\u80fd\u7684\u5173\u952e\u56e0\u7d20\u4e4b\u4e00\u3002\u540c\u65f6\uff0c\u4e0e\u4eba\u548c\u52a8\u7269\u80fd\u591f\u81ea\u4e3b\u5b66\u4e60\u65b0\u7684\u6280\u80fd\u76f8\u6bd4\uff0c\u6ca1\u6709\u4efb\u4f55\u8bc1\u636e\u8868\u660e\u76ee\u524d\u57fa\u4e8e\u6a21\u4eff\u5b66\u4e60\u7684\u65b9\u6cd5\u80fd\u591f\u901a\u8fc7\u5927\u89c4\u6a21\u7684\u9884\u8bad\u7ec3\u6d8c\u73b0\u51fa\u8bad\u7ec3\u96c6\u4e2d\u4ece\u672a\u51fa\u73b0\u8fc7\u7684\u6280\u80fd<\/span>\uff08\u4e0e\u524d\u8ff0\u7684\u96f6\u6837\u672c\u6cdb\u5316\u5230\u65b0\u4efb\u52a1\u4e0d\u540c\uff0c\u6b64\u5904\u5f3a\u8c03\u7684\u662f\u65b0\u6280\u80fd\uff0c\u6bd4\u5982\u201c\u4e3e\u8d77\u74f6\u5b50\u201d\u548c\u201c\u4e3e\u8d77\u676f\u5b50\u201d\u662f\u4e24\u4e2a\u4e0d\u540c\u7684\u4efb\u52a1\uff0c\u4f46\u662f\u7528\u5230\u76f8\u540c\u7684\u6280\u80fd\uff09<\/span>\u3002<\/span><\/p>\n


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

\u76f8\u5bf9\u5730\uff0c\u6a21\u4eff\u5b66\u4e60\u5bf9\u7cbe\u9009\u6570\u636e<\/span> (curated data) <\/span>\u7684\u8981\u6c42\uff0c\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5\u5219\u4e0d\u9700\u8981\u4eba\u4e3a\u6784\u9020\u590d\u6742\u6570\u636e\u96c6\u6765\u5b66\u4e60\u7b56\u7565\uff0c\u800c\u662f\u53ea\u9700\u8981\u4e0e\u73af\u5883\u4ea4\u4e92\uff0c\u5e76\u901a\u8fc7\u4f18\u5316\u5956\u52b1\u51fd\u6570\u5728\u7ebf\u5730\u5b66\u4e60\u7b56\u7565<\/span> (online policy learning)<\/span>\u3002\u5f3a\u5316\u5b66\u4e60<\/span> (\u5c24\u5176\u662f\u65e0\u6a21\u578b\u5f3a\u5316\u5b66\u4e60\uff0cmodel-free RL)<\/span> \u7531\u4e8e\u65e0\u6cd5\u76f4\u63a5\u83b7\u53d6\u73af\u5883\u52a8\u529b\u5b66\u7684\u68af\u5ea6\u4fe1\u606f\uff0c\u5728\u6837\u672c\u6548\u7387\u65b9\u9762\u5f80\u5f80\u6bd4\u6a21\u4eff\u5b66\u4e60\u4f4e\u51e0\u4e2a\u91cf\u7ea7\uff0c\u5bf9\u4e8e\u5b66\u4e60\u5728\u7269\u7406\u73af\u5883\u4e2d\u8fd0\u884c\u7684\u7b56\u7565\uff0c\u8fd9\u6837\u4f4e\u6548\u7684\u5b66\u4e60\u65b9\u5f0f\u662f\u4e0d\u73b0\u5b9e\u7684\u3002\u4e00\u822c\u6765\u8bf4\u6709\u4e24\u79cd\u89e3\u51b3\u65b9\u6848\uff1a<\/span><\/p>\n


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

    \n
  • \n

    \u4e00\u79cd\u662f\u901a\u8fc7\u6784\u5efa\u4e00\u4e2a\u4e0e\u771f\u5b9e\u73af\u5883\u7c7b\u4f3c\u7684\u6a21\u62df\u5668\uff0c\u5728\u6a21\u62df\u5668\u4e2d\u4f7f\u7528\u5927\u91cf\u6570\u636e\u5b66\u4e60\u5230\u4e00\u4e2a\u7b56\u7565\uff0c\u7136\u540e\u5728\u771f\u5b9e\u73af\u5883\u4e2d\u96f6\u6837\u672c\u6cdb\u5316\u6216\u5728\u7ebf\u5fae\u8c03<\/span> (\u5982[5, 6])<\/span>\uff1b<\/span><\/p>\n<\/li>\n

  • \n

    \u53e6\u4e00\u79cd\u65b9\u6848\u662f\u5b66\u4e60\u4e00\u4e2a\u5173\u4e8e\u73af\u5883\u7684\u6a21\u578b\uff0c\u5e76\u5229\u7528\u5b66\u5230\u7684\u6a21\u578b\u6765\u751f\u6210\u5b66\u4e60\u6570\u636e<\/span> (\u5982[7, 8, 9])<\/span>\uff0c\u4ece\u800c\u6781\u5927\u5f97\u51cf\u5c11\u5bf9\u771f\u5b9e\u73af\u5883\u6570\u636e\u7684\u9700\u6c42\uff0c\u57fa\u4e8e\u8fd9\u4e00\u60f3\u6cd5\uff0cPhilipp Wu et.al. [9] \u5c55\u793a\u4e86\u4ec5\u4ec5\u901a\u8fc7\u4e0e\u771f\u5b9e\u7269\u7406\u4e16\u754c\u8fdb\u884c1\u5c0f\u65f6\u7684\u4ea4\u4e92\u5c31\u80fd\u8ba9\u673a\u68b0\u72d7\u5b66\u4f1a\u8d70\u8def\u5e76\u62b5\u6297\u5916\u754c\u7684\u5e72\u6270\u3002<\/span><\/p>\n<\/li>\n<\/ul>\n

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

    \u7136\u800c\uff0c\u76ee\u524d\u6ca1\u6709\u8bc1\u636e\u8868\u660e\u57fa\u4e8e\u5f3a\u5316\u5b66\u4e60\u7684\u65b9\u6cd5\u8bad\u7ec3\u51fa\u7684\u667a\u80fd\u4f53\u80fd\u6d8c\u73b0\u51fa\u5bf9\u5176\u6240\u89e3\u51b3\u7684\u4efb\u52a1\u548c\u73af\u5883\u7684\u8ba4\u77e5\u80fd\u529b\uff0c\u6bd4\u5982\u57fa\u4e8e[9]\u7684\u65b9\u6cd5\u8bad\u7ec3\u9700\u8981\u9891\u7e41\u5730\u4eba\u4e3a\u91cd\u7f6e\u673a\u68b0\u72d7\u7684\u4f4d\u7f6e\uff0c\u56e0\u4e3a\u673a\u68b0\u72d7\u7684\u5956\u52b1\u51fd\u6570\u53ea\u9f13\u52b1\u5b83\u5411\u524d\u8d70\uff0c\u5373\u4f7f\u78b0\u5230\u5899\u4e5f\u4f1a\u53cd\u590d\u5411\u524d\u51b2\u649e\u3002<\/span><\/p>\n


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

    \u8fd9\u4e9b\u7ed3\u679c\u90fd\u6697\u793a\u4e86\u57fa\u4e8e\u76ee\u524d\u7684\u6df1\u5ea6\u5b66\u4e60\u8303\u5f0f\u6765\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u6240\u9762\u4e34\u7684\u4e00\u4e9b\u6839\u672c\u6027\u6311\u6218\uff1a<\/span><\/p>\n


    <\/p>\n

    1\uff09\u76ee\u524d\u7684\u5b66\u4e60\u7cfb\u7edf\u672c\u8d28\u4e0a\u4ecd\u662f\u4e00\u4e2a\u5f00\u73af\u7cfb\u7edf\uff0c\u9700\u8981\u4eba\u7c7b\u667a\u80fd\u7684\u4ecb\u5165<\/strong><\/span>(\u5982\u6839\u636e\u5b66\u4e60\u7ed3\u679c\uff0c\u6709\u9488\u5bf9\u6027\u5730\u91c7\u96c6\u66f4\u591a\u66f4\u597d\u7684\u6570\u636e\uff0c\u8c03\u6574\u6570\u636e\u7684\u6982\u7387\u5206\u5e03\uff0c\u53cd\u590d\u8fed\u4ee3\u4f18\u5316\u5956\u52b1\u51fd\u6570\u7b49)<\/span>\u6765\u5b9e\u73b0\u95ed\u73af<\/strong>\uff0c\u7528Yann Lecun\u7684\u8bdd\u6765\u8bf4\u5c31\u662f\uff0c\u76ee\u524d\u7684\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u662fAssisted Intelligence\uff0c\u800c\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u9700\u8981\u7684\u662fAutonomous Intelligence [10]; <\/span><\/p>\n


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

    2) \u76ee\u524d\u7684\u65b9\u6cd5\u8fd8\u4e0d\u5177\u5907\u4ece\u81ea\u7136\u6a21\u6001\u4e2d\u5b66\u4e60\u5230\u5173\u4e8e\u4e16\u754c\u7684\u7ed3\u6784\u5316\u8868\u5f81\u4e0e\u62bd\u8c61<\/strong><\/span>(\u6216\u8005\u8bf4\u4e16\u754c\u6a21\u578b)<\/span>\u7684\u80fd\u529b<\/span><\/strong>\uff0c\u76f8\u5bf9\u5730\uff0c\u4eba\u7c7b\u548c\u52a8\u7269\u5728\u5a74\u513f\u65f6\u671f\u5c31\u80fd\u4ece\u81ea\u7136\u6a21\u6001<\/span>\uff08\u5982\u89c6\u89c9\uff0c\u542c\u89c9\u7b49\u76f4\u63a5\u6765\u6e90\u4e8e\u5916\u90e8\u4e16\u754c\u7684\u4fe1\u53f7\uff09<\/span>\u4e2d\u5b66\u4e60\u5e76\u57fa\u4e8e\u76f4\u89c9\u7406\u89e3\u7269\u7406\u4e16\u754c\u7684\u7ed3\u6784\u548c\u8fd0\u4f5c\u89c4\u5f8b<\/span>(intuitive physics [11])<\/span>\uff0c\u8fd9\u79cd\u81ea\u7136\u4e60\u5f97\u7684\u8ba4\u77e5\u80fd\u529b\u662f\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u5173\u952e\u3002<\/span><\/p>\n


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


    <\/p>\n

    \n
    \n
    \n
    \n
    \n

    <\/section>\n

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

    4. \u4efb\u52a1\u6307\u5b9a\u95ee\u9898\u3001<\/strong><\/span><\/strong><\/strong><\/p>\n

    \u5e38\u8bc6\u3001\u81ea\u7136\u62bd\u8c61\u548c\u4e16\u754c\u6a21\u578b<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

    \n

    <\/section>\n

    <\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


    <\/p>\n

    \u7531\u4e8e\u7f3a\u4e4f\u4e86\u5b66\u4e60\u5173\u4e8e\u4e16\u754c\u7684\u5c42\u7ea7\u5316\u62bd\u8c61\u7684\u80fd\u529b\uff0c\u4ee5\u76ee\u524d\u7684\u65b9\u6cd5\u6784\u5efa\u5b66\u4e60\u7cfb\u7edf\u9762\u4e34\u4e00\u4e2a\u975e\u5e38\u56f0\u96be\u7684\u5b9e\u9645\u95ee\u9898\uff1a\u4efb\u52a1\u6307\u5b9a\u95ee\u9898<\/strong> [12]<\/span>\uff08\u5176\u672c\u8d28\u4e5f\u662f\u67d0\u79cd\u5bf9\u9f50\u95ee\u9898\uff09<\/span>\uff0c\u901a\u4fd7\u6765\u8bb2\u5c31\u662f\u8bf4\u60f3\u8981\u8bad\u7ec3\u4e00\u4e2a\u6a21\u578b\u6765\u7cbe\u786e\u5730\u5b8c\u6210\u4e00\u9879\u4eba\u7c7b\u5de5\u7a0b\u5e08\u5fc3\u91cc\u5e0c\u671b\u5176\u5b8c\u6210\u7684\u4efb\u52a1\uff0c\u662f\u4e00\u4ef6\u6781\u5176\u56f0\u96be\u7684\u4e8b\u3002<\/span><\/p>\n


    <\/p>\n

    \u6bd4\u5982\u8bf4\uff0c\u8981\u60f3\u8bad\u7ec3\u4e00\u4e2a\u6a21\u578b\u505a\u56fe\u50cf\u8bc6\u522b\uff0c\u6a21\u578b\u88ab\u8bad\u7ec3\u4f18\u5316\u5176\u8f93\u51fa\u4e0e\u56fe\u50cf\u6807\u7b7e\u5728\u4e00\u4e2a\u786e\u5b9a\u6570\u636e\u96c6\u4e0a\u7684\u4ea4\u53c9\u71b5\uff0c\u56e0\u6b64\u8fd9\u4e00\u4efb\u52a1\u5b9e\u9645\u4e0a\u9690\u5f0f\u5730\u662f\u7531\u6570\u636e\u96c6\u6765\u5b9a\u4e49\u7684\uff0c\u8fd9\u5c31\u4ea7\u751f\u4e86\u4e00\u4e2a\u5de8\u5927\u7684\u6311\u6218\uff1a\u4e3a\u4e86\u6307\u5b9a\u4e00\u4e2a\u4efb\u52a1\uff0c\u6211\u4eec\u9700\u8981\u63d0\u4f9b\u6d77\u91cf\u7684\u4fe1\u606f\u3002\u95ee\u9898\u8fd8\u4e0d\u6b62\u4e8e\u6b64\uff0c\u6bd4\u5982\u56fe1\u6240\u793a\uff0c\u5728\u8bad\u7ec3\u96c6\u4e0a\u6a21\u578b\u4f7f\u7528\u4e86\u9519\u8bef\u7684\u65b9\u6cd5<\/span>\uff08\u901a\u8fc7\u68c0\u6d4b\u5230\u7f51\u7403\u548c\u8349\u5730\uff09<\/span>\u5f97\u5230\u4e86\u6b63\u786e\u7684\u5206\u7c7b<\/span>\uff08\u5c0f\u72d7\uff09<\/span>\uff0c\u4e8e\u662f\u5176\u5728\u6d4b\u8bd5\u96c6\u4e0a<\/span>\uff08\u7531\u4e8e\u68c0\u6d4b\u5230\u80cc\u666f\u6c34\u6c60\uff09<\/span>\u9519\u8bef\u5730\u5c06\u5c0f\u72d7\u5206\u7c7b\u4e3a\u6d77\u8c5a\u3002<\/span><\/p>\n


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

    <\/p>\n

    \u56fe1\uff1a\u57fa\u4e8e\u76d1\u7763\u5b66\u4e60\u7684\u56fe\u50cf\u5206\u7c7b\u7b97\u6cd5\u5728\u6d4b\u8bd5\u96c6\uff08\u5982\u5de6\u56fe\u6240\u793a\uff09\u4f18\u5316\u8bad\u7ec3\u635f\u5931\u51fd\u6570\u7684\u7ed3\u679c\u662f\u901a\u8fc7\u7279\u5f81\u4e0e\u56fe\u50cf\u6807\u7b7e\u95f4\u7684\u76f8\u5173\u6027\u6765\u505a\u5206\u7c7b\uff0c\u56e0\u6b64\u5728\u6d4b\u8bd5\u96c6\uff08\u5982\u53f3\u56fe\u6240\u793a\uff09\u4e0a\u7531\u4e8e\u68c0\u6d4b\u5230\u6c34\u6c60\u7684\u80cc\u666f\u9519\u8bef\u5730\u5c06\u5c0f\u72d7\u5206\u7c7b\u4e3a\u4e86\u6d77\u8c5a\uff08\u53d6\u81eaMIT EI Seminar-Pulkit Agrawal-The Task Specification Problem\uff0chttps:\/\/www.youtube.com\/watch?v=2el3GdwS1mg&t=1962s&ab_channel=MITEmbodiedIntelligence)\u3002<\/span><\/section>\n


    <\/p>\n

    \u90a3\u4e48\u8fd9\u91cc\u7684\u6839\u672c\u95ee\u9898\u662f\u4ec0\u4e48\u5462\uff1f\u4e3a\u4ec0\u4e48\u4eba\u80fd\u591f\u5f88\u8f7b\u6613\u5730<\/span>\uff08\u901a\u8fc7\u4ec5\u4ec5\u51e0\u4e2a\u6bd4\u7279\u4fe1\u606f\u91cf\u7684\u81ea\u7136\u8bed\u8a00\uff09<\/span>\u8ba9\u53e6\u5916\u4e00\u4e2a\u4eba\u505a\u4e00\u4ef6\u4e8b\uff0c\u800c\u544a\u8bc9\u673a\u5668\u505a\u4e00\u4ef6\u4e8b\u5374\u5f02\u5e38\u8d39\u529b\uff1f\u6211\u8ba4\u4e3a\u662f\u56e0\u4e3a\u6211\u4eec\u5e0c\u671b\u673a\u5668\u89e3\u51b3\u7684\u4efb\u52a1\u548c\u6211\u4eec\u5bf9\u4efb\u52a1\u7684\u6570\u5b66\u63cf\u8ff0\u4e4b\u95f4\u5b58\u5728\u5dee\u5f02\u6027<\/strong>\uff1a\u62ff\u56fe\u50cf\u5206\u7c7b\u4e3e\u4f8b\uff0c\u6211\u4eec\u5b9e\u9645\u5e0c\u671b\u673a\u5668\u89e3\u51b3\u7684\u4efb\u52a1\u53ef\u80fd\u662f\u201c\u4ee5\u4eba\u8ba4\u4e3a\u6700\u81ea\u7136\u7684\u65b9\u5f0f\u5bf9\u8fd9\u4e9b\u56fe\u50cf\u8fdb\u884c\u5206\u7c7b\u201d\uff0c\u800c\u4e0d\u662f\u201c\u627e\u5230\u80fd\u6700\u597d\u5730\u62df\u5408\u8fd9\u4e9b\u56fe\u50cf\u6807\u7b7e\u7684\u65b9\u6cd5\u201d\u3002\u90a3\u4e48\u4ec0\u4e48\u662f\u201c\u4eba\u8ba4\u4e3a\u6700\u81ea\u7136\u7684\u65b9\u5f0f\u201d\u5462\uff1f\u8fd9\u53ef\u80fd\u5c31\u662f\u6211\u4eec\u6240\u8bf4\u7684\u5e38\u8bc6<\/strong><\/span>(common sense)<\/span>\uff0c\u6bd4\u5982\u6211\u4eec\u5bf9\u56fe\u50cf\u5206\u7c7b\u65f6\u53ea\u4f1a\u8003\u8651\u56fe\u50cf\u7684\u4e3b\u4f53<\/span>\uff08\u770b\u5c0f\u72d7\u800c\u4e0d\u662f\u770b\u5176\u80cc\u540e\u7684\u80cc\u666f\uff09<\/span>\uff0c\u6211\u4eec\u4f1a\u901a\u8fc7\u5bfb\u627e\u5177\u6709\u56e0\u679c\u5173\u7cfb\u7684\u7279\u5f81\u6765\u5206\u8fa8\u4e0d\u540c\u7684\u7269\u4f53<\/span>\uff08\u56e0\u4e3a\u6709\u5c3e\u5df4\u548c\u773c\u775b\u6240\u4ee5\u662f\u5c0f\u72d7\uff09<\/span>\uff0c\u6211\u4eec\u77e5\u9053\u7269\u4f53\u4e0e\u7269\u4f53\u4e4b\u95f4\u7684\u5173\u7cfb<\/span>\uff08\u6bd4\u5982\u72d7\u5c3e\u5df4\u662f\u5c0f\u72d7\u7684\u4e00\u90e8\u5206\uff09<\/span>\u7b49\u7b49\u3002<\/span><\/p>\n


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

    \u8fdb\u4e00\u6b65\u5730\uff0c\u5230\u5e95\u4ec0\u4e48\u662f\u5e38\u8bc6\uff0c\u4eba\u7c7b\u7684\u5e38\u8bc6\u5230\u5e95\u662f\u600e\u4e48\u6765\u7684\uff0c\u4e3a\u4ec0\u4e48\u4e0d\u540c\u7684\u4eba<\/span>\uff08\u751a\u81f3\u5206\u522b\u6765\u81ea\u4ece\u672a\u6709\u8fc7\u63a5\u89e6\u7684\u6587\u660e\uff09<\/span>\u4e5f\u80fd\u4ea7\u751f\u7c7b\u4f3c\u7684\u5e38\u8bc6<\/span>\uff08\u56e0\u800c\u4eba\u548c\u4eba\u4e4b\u95f4\u80fd\u591f\u4f7f\u7528\u8bed\u8a00\u4ea4\u6d41\uff09<\/span>\uff0c\u751a\u81f3\u4eba\u548c\u52a8\u7269\u4e4b\u95f4\u4e5f\u80fd\u7528<\/span>\uff08\u5305\u62ec\u80a2\u4f53\uff09<\/span>\u8bed\u8a00\u8fdb\u884c\u4e00\u5b9a\u7a0b\u5ea6\u7684\u4ea4\u6d41\u3002\u6211\u8ba4\u4e3a\u5176\u539f\u56e0\u5728\u4e8e\u4eba\u7c7b\u548c\u52a8\u7269\u90fd\u6709\u4e00\u5957\u76f8\u7c7b\u4f3c\u7684\u5b66\u4e60\u5173\u4e8e\u4e16\u754c\u7684\u81ea\u7136\u62bd\u8c61<\/span>(natural abstraction [13]\uff0c\u81ea\u7136\u62bd\u8c61\u5047\u8bf4\u8ba4\u4e3a\u6211\u4eec\u7684\u4e16\u754c\u5b58\u5728\u67d0\u4e9b\u7a33\u5b9a\u7684\u4f4e\u7ef4\u7ed3\u6784)<\/span>\u7684\u65b9\u5f0f\uff0c\u4ece\u800c\u4ed6\u4eec\u80fd\u5b66\u5230\u76f8\u4f3c\u7684\u5c42\u7ea7\u5316\u62bd\u8c61\u3002\u8fd9\u4e00\u5957\u5c42\u7ea7\u5316\u62bd\u8c61<\/strong>\u7684\u4f4e\u5c42\u7ea7\u7aef\u901a\u5e38\u662f\u76f8\u5bf9\u9ad8\u7ef4\u5ea6\u7684\uff0c\u96be\u4ee5\u76f4\u63a5\u7528\u8bed\u8a00\u8868\u8ff0\u51fa\u6765\u7684\u5206\u5e03\u5f0f\u8fde\u7eed\u8868\u5f81\uff1b\u800c\u5176\u9ad8\u5c42\u7ea7\u7aef\u901a\u5e38\u662f\u66f4\u52a0\u4f4e\u7ef4\u5ea6\u7684\uff0c\u7ecf\u8fc7\u4e86\u62bd\u8c61\u538b\u7f29\u540e\u5177\u6709\u4e00\u5b9a\u8bed\u4e49\u4fe1\u606f\u7684\u8868\u5f81\uff1b\u5176\u9876\u7aef\u662f\u79bb\u6563\u7684\u5177\u6709\u6e05\u6670\u8bed\u4e49\u7684\u7b26\u53f7\u5316\u8868\u5f81\u3002\u5728\u8ba1\u7b97\u65b9\u9762\uff0c\u4ece\u4f4e\u5c42\u7ea7\u5230\u9ad8\u5c42\u7ea7\u5206\u522b\u5bf9\u5e94\u4e8e\u8ba4\u77e5\u79d1\u5b66\u4e2d\u7cfb\u7edf1\u548c\u7cfb\u7edf2\u4e4b\u95f4\u8fde\u7eed\u6e10\u53d8\u7684\u4fe1\u606f\u5904\u7406\u6a21\u5f0f\uff0c\u800c\u5e38\u8bc6\u5219\u5bf9\u5e94\u4e8e\u8fd9\u5957\u5c42\u7ea7\u5316\u62bd\u8c61\u4f53\u7cfb\u4e2d\u5c42\u7ea7\u4e4b\u95f4\u4e0e\u5c42\u5185\u4e0d\u540c\u90e8\u5206\u4e4b\u95f4\u7684\u8fde\u63a5\u7ed3\u6784<\/strong>\uff0c\u6b63\u662f\u7531\u4e8e\u4e0d\u540c\u7684\u4eba\u548c\u52a8\u7269\u90fd\u80fd\u5b66\u4e60\u5230\u76f8\u7c7b\u4f3c\u7684\u62bd\u8c61\u548c\u8fde\u63a5\u7ed3\u6784\uff0c\u4ec5\u9700\u8981\u901a\u8fc7\u6700\u9ad8\u5c42\u7684\u79bb\u6563\u7b26\u53f7\u8868\u5f81\u4f5c\u4e3a\u4fe1\u606f\u4f20\u9012\u7684\u8f7d\u4f53\u6765\u6307\u5b9a\u4efb\u52a1\uff0c\u6240\u6709\u7684\u4f4e\u5c42\u7ea7\u7684\u4efb\u52a1\u4fe1\u606f\u4e0d\u9700\u8981\u76f4\u63a5\u6307\u5b9a\uff0c\u800c\u662f\u7531\u4fe1\u606f\u7684\u63a5\u53d7\u8005\u6839\u636e\u9ad8\u5c42\u7ea7\u4fe1\u606f\u548c\u81ea\u5df1\u7684\u62bd\u8c61\u4f53\u7cfb\u7528\u7c7b\u4f3c\u4e8e\u8d1d\u53f6\u65af\u63a8\u65ad<\/span>(bayesian inference)<\/span>\u7684\u65b9\u5f0f\u8fdb\u884c\u8865\u5168\u3002<\/span><\/p>\n


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

    \u56e0\u6b64\uff0c<\/span>\u6211\u8ba4\u4e3a\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u4e00\u4e2a\u5173\u952e\u95ee\u9898\uff0c\u662f\u5982\u4f55\u4f7f\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u4ece\u81ea\u7136\u6a21\u6001\u4e2d<\/span><\/strong>\uff08\u5982\u89c6\u89c9\uff0c\u542c\u89c9\uff09<\/span>\u5b66\u4e60\u5230\u5173\u4e8e\u4e16\u754c\u7684\u5c42\u7ea7\u5316\u62bd\u8c61<\/span><\/strong>\uff08\u6216\u8005\u8bf4\u662f\u4e16\u754c\u6a21\u578b\uff0c\u8ba4\u77e5\u5730\u56fe\uff1b\u4ee5\u4e0b\u6211\u4eec\u5c06\u5176\u7b80\u79f0\u4e3a\u4e16\u754c\u6a21\u578b\uff09<\/span>\u3002\u4e3a\u4ec0\u4e48\u8fd9\u4e48\u8bf4\u5462\uff1f\u6211\u8ba4\u4e3a\u81f3\u5c11\u6709\u4ee5\u4e0b\u51e0\u70b9\u539f\u56e0\uff1a<\/span><\/p>\n


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

    \uff081\uff09\u4e16\u754c\u6a21\u578b\u80fd\u5f88\u597d\u5730\u5c06\u611f\u77e5\u548c\u8ba4\u77e5\u7edf\u4e00\u5728\u540c\u4e00\u4e2a\u6846\u67b6\u4e0b\u3002<\/span><\/strong><\/p>\n


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

    \u5728\u76ee\u524d\u7684\u5177\u8eab\u667a\u80fd\u7cfb\u7edf\u4e2d\uff0c\u4ece\u611f\u77e5\u5230\u8ba4\u77e5\u4f3c\u4e4e\u662f\u901a\u8fc7\u76f8\u5bf9\u72ec\u7acb\u7684\u6a21\u5757\u6765\u5b8c\u6210\u7684\uff0c\u6bd4\u5982\u901a\u8fc7\u57fa\u4e8eSLAM\u6216\u8005\u6df1\u5ea6\u5b66\u4e60\u7684\u65b9\u6cd5\u83b7\u5f97\u5173\u4e8e\u5468\u56f4\u73af\u5883\u7684\u4e09\u7ef4\u51e0\u4f55\u4fe1\u606f\u5982\u6df1\u5ea6\u56fe\uff1b\u901a\u8fc7\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u7684\u65b9\u6cd5\u83b7\u5f97\u542b\u6709\u8bed\u4e49\u4fe1\u606f\u7684\u7ed3\u679c\u5982\u76ee\u6807\u68c0\u6d4b\uff0c\u5206\u7c7b\uff0c\u5206\u5272\u7b49\uff1b\u518d\u901a\u8fc7\u624b\u5199\u7684\u89c4\u5219\u6216\u8005\u8f83\u4e3a\u4f20\u7edf\u7684\u4eba\u5de5\u667a\u80fd\u6280\u672f\u5bf9\u7ed3\u679c\u8fdb\u884c\u5904\u7406\u6765\u83b7\u53d6\u4e00\u5b9a\u7684\u8ba4\u77e5\u80fd\u529b\uff0c\u5982\u5bf9\u573a\u666f\u7684\u7406\u89e3\u7b49\u7b49\u3002\u8fd9\u79cd\u6a21\u5757\u5316\u7684\u65b9\u6cd5\u5728\u6a21\u5757\u4e0e\u6a21\u5757\u4e4b\u95f4\u7684\u8854\u63a5\u4e0a\u9762\u4e34\u7740\u5f88\u5927\u7684\u6311\u6218\uff0c\u8fd9\u4e5f\u662f\u9ad8\u7ea7\u522b\u7684\u81ea\u52a8\u9a7e\u9a76\u6280\u672f\u8fdf\u8fdf\u96be\u4ee5\u843d\u5730\u7684\u539f\u56e0\u4e4b\u4e00\u3002<\/span><\/p>\n


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

    \u7a76\u5176\u672c\u8d28\uff0c\u6211\u8ba4\u8bc6\u6839\u672c\u95ee\u9898\u5728\u4e8e\u611f\u77e5\u548c\u8ba4\u77e5\u662f\u540c\u4e00\u4e2a\u8fde\u7eed\u5149\u8c31\u7684\u4e24\u7aef\uff0c\u662f\u4e00\u4e2a\u96be\u4ee5\u5206\u5272\u7684\u6574\u4f53\uff0c\u8981\u5b8c\u7f8e\u89e3\u51b3\u611f\u77e5\u95ee\u9898\u5fc5\u7136\u6d89\u53ca\u5230\u8ba4\u77e5\uff0c\u53cd\u4e4b\u4ea6\u7136\uff0c\u5e76\u4e14\u8ba4\u77e5\u80fd\u529b\u5e76\u4e0d\u662f\u51ed\u7a7a\u4ea7\u751f\u7684\uff0c\u56e0\u6b64\u5408\u7406\u7684\u505a\u6cd5\u5e76\u4e0d\u662f\u901a\u8fc7\u5c06\u8ba4\u77e5\u6a21\u5757\u76f4\u63a5\u8bbe\u8ba1\u8fdb\u667a\u80fd\u7cfb\u7edf\uff0c\u800c\u662f\u5728\u540c\u4e00\u4e2a\u6846\u67b6\u4e0b\u901a\u8fc7\u5bf9\u66f4\u4f4e\u5c42\u7ea7\u7684\u611f\u77e5\u4efb\u52a1\u8fdb\u884c\u4f18\u5316\uff0c\u4f7f\u5f97\u8ba4\u77e5\u80fd\u529b\u81ea\u7136\u5730\u6d8c\u73b0\u51fa\u6765\uff0c\u8fd9\u770b\u8d77\u6765\u4e5f\u662f\u4eba\u7c7b\u548c\u52a8\u7269\u5b66\u4e60\u8ba4\u77e5\u4e16\u754c\u7684\u65b9\u5f0f\u3002<\/span><\/p>\n


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

    \uff082\uff09\u4e16\u754c\u6a21\u578b\u8ba9\u667a\u80fd\u4f53\u4ee5\u8ba1\u7b97\u53ef\u884c\u7684<\/span><\/span><\/strong>\uff08computationally-tractable\uff09<\/span><\/strong><\/span>\u65b9\u5f0f\u5bf9\u672a\u6765\u8fdb\u884c\u957f\u671f\u9884\u6d4b\u548c\u89c4\u5212\u3002<\/span><\/strong><\/p>\n


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

    \u6211\u4eec\u5728\u9884\u6d4b\u548c\u89c4\u5212\u65f6\u7a7a\u4e0a\u9065\u8fdc\u7684\u4e8b\u4ef6\u65f6\u5f80\u5f80\u4f1a\u505a\u5c42\u7ea7\u5316\u7684\u62bd\u8c61\uff0c\u5e76\u4f7f\u7528\u76f8\u5bf9\u7c97\u7c92\u5316\u6982\u5ff5\u5316\u7684\u8868\u5f81\uff0c\u6bd4\u5982\u89c4\u5212\u53bb\u56fd\u5916\u65c5\u884c\uff0c\u6211\u4eec\u4f1a\u5148\u8003\u8651\u9014\u5f84\u54ea\u4e9b\u57ce\u5e02\uff0c\u800c\u4e0d\u662f\u7ec6\u5206\u5230\u4ece\u67d0\u5ea7\u5927\u697c\u600e\u4e48\u5230\u53e6\u4e00\u5ea7\u57ce\u5e02\u7684\u67d0\u5ea7\u5927\u697c\u3002\u8fd9\u79cd\u5728\u5408\u9002\u7c97\u7c92\u5ea6\u4e0a\u8fdb\u884c\u89c4\u5212\u7684\u80fd\u529b\u4f7f\u5f97\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u6709\u9650\u7684\u8ba1\u7b97\u8d44\u6e90\u6709\u6548\u5730\u5bfb\u627e\u5230\u4e00\u4e2a\u9ad8\u5ea6\u4f18\u5316\u7684\u65b9\u6848\uff0c\u8fd9\u5bf9\u65f6\u95f4\u548c\u7a7a\u95f4\u4e0a\u7684\u957f\u8fdc\u671f\u89c4\u5212\u662f\u81f3\u5173\u91cd\u8981\u7684\u3002<\/span><\/p>\n


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

    Tufts\u5927\u5b66\u7684\u53d1\u80b2\u751f\u7269\u5b66\u5bb6<\/span>\uff08developmental biologist\uff09<\/span>Michael Levin\u63d0\u51fa\u4e86\u8ba4\u77e5\u5149\u9525<\/span>\uff08cognitive light-cone\uff09<\/span>\u7684\u6982\u5ff5\u6765\u8861\u91cf\u667a\u80fd\u4f53\u7684\u8ba4\u77e5\u80fd\u529b\uff0c\u8ba4\u4e3a\u667a\u80fd\u4f53\u80fd\u591f\u6d4b\u91cf\uff0c\u6a21\u62df\uff0c\u5e76\u5f71\u54cd\u7684\u4e8b\u4ef6\u5728\u65f6\u95f4\u548c\u7a7a\u95f4\u4e0a\u7684\u8fb9\u754c\u51b3\u5b9a\u4e86\u667a\u80fd\u4f53\u7684\u8ba4\u77e5\u6781\u9650\u3002\u6bd4\u5982\uff0c\u4eba\u7c7b\u4f1a\u5173\u5fc3\u5e76\u7814\u7a76\u65e9\u81f3\u5b87\u5b99\u8d77\u6e90<\/span>\u65f6\u671f\uff0c\u800c\u9065\u8fdc\u81f3\u76ee\u524d\u5b87\u5b99\u89c6\u754c\u8fb9\u754c\u7684\u4e8b\u4ef6\uff0c\u5305\u62ec\u65f6\u95f4\u7a7a\u95f4\u5c3a\u5ea6\u8fdc\u5c0f\u4e8e\/\u5927\u4e8e\u5176\u81ea\u8eab\u751f\u7406\u8fc7\u7a0b\u7684\u4e8b\u4ef6<\/span>\uff08\u5982\u57fa\u672c\u7c92\u5b50\u7684\u76f8\u4e92\u4f5c\u7528\uff0c\u661f\u7cfb\u7684\u6f14\u5316\u7b49\uff09<\/span>\uff0c\u56e0\u800c\u4eba\u7c7b\u5177\u6709\u5f88\u5f3a\u7684\u8ba4\u77e5\u5e76\u6539\u9020\u4e16\u754c\u7684\u80fd\u529b\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u66f4\u52a0\u4f4e\u7ea7\u7684\u667a\u80fd\u4f53\uff0c\u5982\u4e00\u4e2a\u8349\u5c65\u866b\uff0c\u4e00\u4e2a\u7ec6\u80de\uff0c\u53ef\u80fd\u4ec5\u4ec5\u5173\u5fc3\u5176\u8eab\u8fb9\u7684\u6eb6\u6db2\u6d53\u5ea6\u68af\u5ea6\u7b49\u5c0f\u5c3a\u5ea6\u4e8b\u4ef6\uff0c\u5c31\u4e0d\u9700\u8981\u5f88\u9ad8\u7684\u8ba4\u77e5\u80fd\u529b\u3002<\/span><\/p>\n


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

    \uff083\uff09\u4e16\u754c\u6a21\u578b\u5305\u542b\u4e86\u4e8b\u4ef6\u95f4\u7684\u56e0\u679c\u5173\u7cfb<\/span><\/strong>\uff08\u4e00\u79cd\u7279\u6b8a\u7684\u5bf9\u65f6\u5e8f\u7684\u62bd\u8c61\uff09<\/span>\uff0c\u4f7f\u5f97\u667a\u80fd\u4f53\u80fd\u591f\u8fdb\u884c\u53cd\u4e8b\u5b9e\u63a8\u7406<\/span>\uff08counterfactual reasoning\uff09<\/span><\/strong>\uff0c<\/span>\u5e76\u636e\u6b64\u9ad8\u6548\u4e14\u81ea\u52a8\u5730\u5b66\u4e60\u65b0\u77e5\u8bc6\u548c\u65b0\u6280\u80fd\u3002<\/span><\/strong><\/p>\n


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

    \u4eba\u7c7b\u548c\u52a8\u7269\u5b66\u4e60\u6280\u80fd\u76f8\u6bd4\u4e8e\u76ee\u524d\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u5177\u6709\u9ad8\u5f97\u591a\u7684\u6837\u672c\u6548\u7387\uff0c\u5176\u4e2d\u4e00\u4e2a\u5f88\u91cd\u8981\u7684\u539f\u56e0\u5e94\u8be5\u662f\u6211\u4eec\u80fd\u591f\u901a\u8fc7\u4e16\u754c\u6a21\u578b\u6784\u60f3\u51fa\u884c\u4e3a\u548c\u7ed3\u679c\u4e4b\u95f4\u7684\u5bf9\u5e94\u5173\u7cfb<\/span>\uff08\u5373\u56e0\u679c\u5173\u7cfb\uff09<\/span>\u3002\u56e0\u800c\u5f53\u6211\u4eec\u5e0c\u671b\u8fbe\u5230\u67d0\u79cd\u7ed3\u679c<\/span>\uff08\u5b8c\u6210\u67d0\u4e2a\u4efb\u52a1\uff09<\/span>\u7684\u65f6\u5019\uff0c\u5373\u4f7f\u4e8b\u5148\u4ece\u6765\u6ca1\u6709\u505a\u8fc7\uff0c\u6211\u4eec\u4e5f\u53ef\u4ee5\u5f88\u5bb9\u6613\u5730\u89c4\u5212\u51fa\u4e00\u4e2a\u5927\u81f4\u7684\u884c\u4e3a\u8f68\u8ff9\u3002\u5f53\u6211\u4eec\u91cd\u590d\u4e00\u4e2a\u4efb\u52a1\uff0c\u6211\u4eec\u4f1a\u4e0d\u65ad\u505a\u53cd\u4e8b\u5b9e\u63a8\u7406\u6765\u53cd\u601d\u4e2d\u95f4\u7684\u54ea\u4e9b\u6b65\u9aa4\u53ef\u4ee5\u505a\u5f97\u66f4\u597d\uff0c\u8fd9\u4e2a\u5faa\u73af\u5f80\u590d\u7684\u8fc7\u7a0b\u5c31\u662f\u6211\u4eec\u5b66\u4e60\u65b0\u6280\u80fd\u7684\u8fc7\u7a0b\u3002<\/span><\/p>\n


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


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

    \n
    \n
    \n
    \n
    \n

    <\/section>\n

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

    5. \u5728\u673a\u5668\/\u7edf\u8ba1\u5b66\u4e60\u6846\u67b6\u4e0b\u5982\u4f55\u5b66\u4e60\u4e16\u754c\u6a21\u578b<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

    \n

    <\/section>\n

    <\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


    <\/p>\n

    \u90a3\u4e48\u6838\u5fc3\u95ee\u9898\u5c31\u53d8\u6210\u4e86\uff1a\u5982\u4f55\u6784\u5efa\u4e00\u4e2a\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u4f7f\u5176\u80fd\u4ece\u81ea\u7136\u6a21\u6001\u5b66\u4e60\u5230\u4e16\u754c\u6a21\u578b<\/strong>\uff1f\u5e76\u4e14\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u8fd9\u6837\u7684\u5b66\u4e60\u7cfb\u7edf\u5e94\u5f53\u662fautonomous intelligence\uff0c\u800c\u4e0d\u662f\u76ee\u524d\u8fd9\u6837\u9700\u8981\u7531\u4eba\u7c7b\u9891\u7e41\u5e72\u9884\u6765\u5e2e\u52a9\u7684assisted intelligence [10] <\/span>\uff08\u5982\u9700\u8981\u4eba\u7c7b\u91c7\u96c6curated dataset\uff0c\u5e76\u4e14\u53cd\u590d\u8c03\u6574\u6570\u636e\u7684\u5206\u5e03\u6765\u786e\u4fdd\u5176\u591a\u6837\u6027\uff09<\/span>\uff1b\u53e6\u4e00\u65b9\u9762\uff0c\u63d0\u51fa\u7684\u5b66\u4e60\u65b9\u5f0f\u5e94\u5f53\u662f\u5b9e\u9645\u53ef\u884c\u7684\uff0c\u5176\u4e2d\u4e00\u4e2a\u91cd\u8981\u6307\u6807\u5c31\u662f\u8ba1\u7b97\u53ef\u884c\u6027<\/span>(computational tractability)<\/span>\uff0c\u56e0\u4e3a\u5176\u5b9e\u5f88\u5bb9\u6613\u5c31\u80fd\u63d0\u51fa\u4e00\u4e9b\u6ca1\u6709\u4efb\u4f55\u5b9e\u9645\u64cd\u4f5c\u53ef\u884c\u6027\u7684\u65b9\u6cd5\u6765\u5b9e\u73b0\u5177\u8eab\u901a\u7528\u667a\u80fd\uff0c\u6bd4\u5982\u6839\u636e\u7269\u7406\u5b9a\u5f8b\u7cbe\u786e\u6a21\u62df\u5b87\u5b99\u7684\u6f14\u5316\u5e76\u7b49\u5f85\u5176\u4e2d\u51fa\u73b0\u50cf\u4eba\u7c7b\u4e00\u6837\u7684\u667a\u6167\u751f\u7269\uff0c\u53c8\u6bd4\u5982\u8bd5\u56fe\u4ee5\u4e0d\u53ef\u8ba1\u7b97<\/span>\uff08uncomputable\uff09<\/span>\u7684\u91cf\u4f5c\u4e3a\u4f18\u5316\u76ee\u6807<\/span>\uff08\u6bd4\u5982Kolmogorov complexity\uff09<\/span>\u3002<\/span><\/p>\n


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

    \u8981\u56de\u7b54\u8fd9\u4e2a\u95ee\u9898\uff0c\u4ece\u673a\u5668\u5b66\u4e60\/\u7edf\u8ba1\u5b66\u4e60\u7684\u89d2\u5ea6\uff0c\u53ef\u4ee5\u4ece\u56db\u4e2a\u65b9\u9762\u6765\u8003\u8651\uff1a<\/span><\/p>\n


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

    \uff081\uff09\u795e\u7ecf\u7f51\u7edc\u67b6\u6784<\/span><\/strong><\/span><\/h2>\n


    <\/p>\n

    \u5927\u8111\u662f\u4e00\u4e2a\u5177\u6709\u8bb0\u5fc6\u548c\u53cd\u9988\u8fde\u63a5\u7684\u52a8\u529b\u7cfb\u7edf\uff0c\u4e14\u65f6\u523b\u90fd\u5728\u9884\u6d4b\u611f\u5b98\u4fe1\u53f7\u5e76\u7ea0\u6b63\u5176\u9884\u6d4b\uff0c\u8fd9\u79cd\u5c42\u7ea7\u5316\u7684\u9884\u6d4b\u7f16\u7801\u67b6\u6784<\/span>(predictive-coding)<\/span>\u8fde\u63a5\u4e86\u611f\u77e5\u4e0e\u8ba4\u77e5\u3002\u6211\u4eec\u7b2c\u4e00\u89c6\u89d2\u80fd\u611f\u77e5\u5230\u7684\u5173\u4e8e\u4e16\u754c\u7684\u4e00\u5207\u4fe1\u606f\u90fd\u662f\u5b58\u5728\u4e8e\u5927\u8111\u72b6\u6001\u4e2d\u7684\u4fe1\u606f\u6d41\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u5f53\u4eca\u4e3b\u6d41\u7684\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\u662f\u524d\u9988\u7684<\/span>\uff08feedforward\uff09<\/span>\uff0c\u6ca1\u6709\u72b6\u6001\u6216\u4fe1\u606f\u53cd\u9988\u6d41\uff0c\u8fd9\u79cd\u67b6\u6784\u4f3c\u4e4e\u4e0d\u592a\u53ef\u80fd\u7f16\u7801\u8868\u5f81\u4e16\u754c\u7684\u72b6\u6001\u4fe1\u606f\u3002\u8fd1\u671f\u7684\u4e00\u9879\u7814\u7a76[14]\u8868\u660e\uff0ctransformer\u7684\u4e0a\u4e0b\u6587\u5b66\u4e60\u80fd\u529b\u4e0e\u5176\u524d\u5411\u5c55\u5f00<\/span>\uff08forward pass\uff09<\/span>\u8fc7\u7a0b\u4e2d\u7684\u8bb0\u5fc6\u5bb9\u91cf<\/span>\uff08\u6216\u7b49\u6548\u4e8e\u4e2d\u95f4\u72b6\u6001\u7684\u5927\u5c0f\uff09<\/span>\u5b58\u5728\u56e0\u679c\u5173\u7cfb\uff0c\u800c\u4e0d\u662f\u4e0e\u6a21\u578b\u53c2\u6570\u7684\u6570\u91cf\u3002\u8fd9\u4e00\u53d1\u73b0\u8868\u660e\uff0c\u8bb0\u5fc6\u5bb9\u91cf\u5bf9\u4e8e\u901a\u7528\u5b66\u4e60\u80fd\u529b\u7684\u6d8c\u73b0\u8d77\u7740\u81f3\u5173\u91cd\u8981\u7684\u4f5c\u7528\uff0c\u800c\u5982\u4f55\u6709\u6548\u5730\u5e26\u6709\u8bb0\u5fc6\u548c\u53cd\u9988\u94fe\u63a5\u7684\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\u662f\u76ee\u524d\u8fd8\u672a\u89e3\u51b3\u7684\u95ee\u9898\uff0c\u5176\u672c\u8d28\u662f\u4e00\u4e2a\u6781\u5177\u6311\u6218\u6027\u7684temporal credit assignment\u95ee\u9898[15]<\/span> (\u5373\u5982\u4f55\u8861\u91cf\u5f53\u524d\u65f6\u523b\u7684\u67d0\u4e00\u53d8\u5316\u5bf9\u672a\u6765\u67d0\u4e00\u65f6\u523b\u7ed3\u679c\u7684\u5f71\u54cd\uff0c\u5728\u795e\u7ecf\u7f51\u7edc\u8bad\u7ec3\u7684\u8bed\u5883\u4e0b\u5bf9\u5e94\u4e8e\u6839\u636e\u672a\u6765\u67d0\u4e2a\u76ee\u6807\u51fd\u6570\u7684\u503c\u5982\u4f55\u53cd\u63a8\u4ee5\u8c03\u6574\u795e\u7ecf\u7f51\u7edc\u7684\u53c2\u6570\uff1b\u76ee\u524d\u6df1\u5ea6\u5b66\u4e60\u4e2d\u6700\u4e3b\u6d41\u7684\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u5c31\u662f\u4e00\u79cdcredit assignment\u673a\u5236)<\/span>\u3002<\/span><\/p>\n


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

    \u5728\u5b66\u4e60\u65b0\u6280\u80fd\u65f6\uff0c\u5927\u8111\u4f1a\u52a8\u6001\u751f\u6210\u4fc3\u8fdb\u4e13\u95e8<\/span>\uff08task-specific\uff09<\/span>\u795e\u7ecf\u56de\u8def\u5f62\u6210\u548c\u52a0\u5f3a\u7684\u5956\u52b1\u4fe1\u53f7\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u5f53\u524d\u7684\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u4f9d\u8d56\u4e8e\u6570\u636e\u96c6\u6216\u5de5\u7a0b\u5e08\u8bbe\u8ba1\u7684\u5956\u52b1\u51fd\u6570\u4e2d\u5305\u542b\u7684\u4fe1\u606f\uff0c\u8fd8\u65e0\u6cd5\u901a\u8fc7\u81ea\u6211\u8fed\u4ee3\u4ea7\u751f\u8fdc\u8d85\u6570\u636e\u5206\u5e03\u7684\u80fd\u529b\uff0c\u6211\u8ba4\u4e3a\u8fd9\u4ece\u67d0\u79cd\u610f\u4e49\u4e0a\u6765\u8bf4\u53cd\u6620\u4e86\u76ee\u524d\u7684\u7cfb\u7edf\u8fd8\u6ca1\u6709\u80fd\u591f\u5b9e\u73b0\u7c7b\u4f3c\u4e8e\u751f\u547d\u4f53\u901a\u8fc7\u6444\u53d6\u80fd\u91cf\u6765\u964d\u4f4e\u5176\u70ed\u529b\u5b66\u71b5\uff0c\u667a\u80fd\u4f53\u901a\u8fc7\u6574\u5408\u5904\u7406\u4fe1\u606f\u5b9e\u73b0\u4fe1\u606f\u71b5\u964d\u7684\u80fd\u529b\u3002<\/span><\/p>\n


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

    \u4ece\u590d\u6742\u52a8\u529b\u5b66\u7cfb\u7edf\u7684\u89d2\u5ea6\u6765\u770b\uff0c\u5927\u8111\u8868\u73b0\u51fa\u7684\u8fd9\u79cd\u4e0b\u884c\u56e0\u679c\u5173\u7cfb<\/span>\uff08\u5373\u76f8\u5bf9\u9ad8\u5c42\u7ea7\u7684\u8ba4\u77e5\u8fc7\u7a0b\u4f1a\u53cd\u8fc7\u6765\u5f71\u54cd\u5230\u76f8\u5bf9\u4f4e\u5c42\u7ea7\u7684\u795e\u7ecf\u6d3b\u52a8\uff09<\/span>\u662f\u67d0\u4e9b\u590d\u6742\u52a8\u529b\u7cfb\u7edf\u7684\u666e\u904d\u7279\u5f81\uff0c\u8fd9\u4e9b\u7cfb\u7edf\u5728\u591a\u4e2a\u5c3a\u5ea6\u4e0a\u8fd0\u4f5c\uff0c\u4f8b\u5982\u7ec6\u80de\u3001\u4eba\u4f53\u3001\u516c\u53f8\u548c\u4eba\u7c7b\u793e\u4f1a\u7b49\u7b49\u3002\u8fd9\u4e9b\u7cfb\u7edf\u901a\u5e38\u5305\u542b\u591a\u4e2a\u5c42\u7ea7\u7684\u53cd\u9988\u5faa\u73af\uff0c\u5176\u4e2d\u6bcf\u4e2a\u5c3a\u5ea6\u4e0a\u90fd\u6709\u5177\u6709\u4ee3\u7406\u80fd\u529b<\/span>\uff08agency\uff09<\/span>\u7684\u7ec4\u4ef6\u53c2\u4e0e\u89e3\u51b3\u5bf9\u5e94\u7684\u95ee\u9898\uff0c\u540c\u65f6\u53d7\u5230\u66f4\u4f4e\u548c\u66f4\u9ad8\u5c42\u7ea7\u7684\u5f71\u54cd\u548c\u7ea6\u675f\u3002\u6bd4\u5982\u5728\u4e00\u4e2a\u516c\u53f8\u91cc\uff0c\u57fa\u5c42\u7684\u5458\u5de5\u8d1f\u8d23\u5904\u7406\u57fa\u5c42\u7684\u4e8b\u52a1\uff0c\u540c\u65f6\u53d7\u5230\u7ba1\u7406\u5c42\u5458\u5de5\u7684\u7ba1\u7406\uff0c\u7ba1\u7406\u5c42\u5458\u5de5\u540c\u6837\u53d7\u5230\u9ad8\u7ea7\u7ba1\u7406\u5c42\u7684\u7ba1\u7406\uff0c\u6bcf\u4e2a\u5c42\u7ea7\u90fd\u53ea\u9700\u8981\u5173\u5fc3\u81ea\u5df1\u5c42\u7ea7\u7684\u95ee\u9898\uff0c\u800c\u4ed6\u4eec\u7ec4\u5408\u8d77\u6765\u5f62\u6210\u4e86\u4e00\u4e2a\u5177\u6709\u81ea\u4e3b\u51b3\u7b56\u548c\u5e94\u53d8\u80fd\u529b\u7684\u6574\u4f53\u3002Michael Levin\u5c06\u8fd9\u79cd\u67b6\u6784\u79f0\u4e3amulti-scale competency [16]\uff0c\u5e76\u6307\u51fa\u8fd9\u79cd\u67b6\u6784\u6709\u6548\u907f\u514d\u4e86\u4e0a\u5c42\u7ec4\u4ef6\u5bf9\u4e0b\u5c42\u7ec4\u4ef6\u8fdb\u884c\u5fae\u7ba1\u7406<\/span>\uff08micro-management\uff09<\/span>\uff0c\u4f7f\u5f97\u6574\u4e2a\u7cfb\u7edf\u4ece\u8ba1\u7b97\u89d2\u5ea6\u6765\u8bf4\u66f4\u52a0\u9ad8\u6548\uff0c\u4e5f\u66f4\u5177\u6709\u5f39\u6027\u548c\u9c81\u68d2\u6027\u3002\u56e0\u6b64\u6211\u8ba4\u4e3a\u5728\u591a\u4e2a\u5c3a\u5ea6\u4e0a\u7684\u4fe1\u606f\u53cd\u9988\u5faa\u73af<\/strong><\/span>\uff08feedback loop\uff09<\/span>\u53ef\u80fd\u5bf9<\/span>autonomous intelligence\u7684\u6d8c\u73b0\u81f3\u5173\u91cd\u8981<\/span><\/strong>\uff0c\u800c\u8fd9\u5728\u5f53\u524d\u7cfb\u7edf\u4e2d\u662f\u4e0d\u5177\u5907\u7684\u3002<\/span><\/p>\n


    <\/h2>\n

    \uff082\uff09\u5b66\u4e60\u6cd5\u5219<\/span><\/strong><\/span><\/h2>\n


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

    \u4e0a\u9762\u7684\u8ba8\u8bba\u4e2d\u4e5f\u63d0\u5230\uff0c\u76ee\u524d\u7684\u795e\u7ecf\u7f51\u7edc\u666e\u904d\u91c7\u7528\u524d\u9988\u67b6\u6784\uff0c\u5e76\u4e0d\u662f\u56e0\u4e3a\u4e0d\u60f3\u7528\u5e26\u6709\u591a\u5c42\u7ea7\u53cd\u9988\u56de\u8def\u548c\u72b6\u6001\u7684\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c\u800c\u662f\u6211\u4eec\u4e0d\u77e5\u9053\u5982\u4f55\u6709\u6548\u5730\u8bad\u7ec3\u8fd9\u79cd\u67b6\u6784\u3002\u751a\u81f3\u57281986\u5e74Hinton\u7b49\u5b66\u8005\u63d0\u51fa\u53cd\u5411\u4f20\u64ad<\/span>\uff08backpropagation\uff09<\/span>\u4e4b\u524d\uff0c\u6ca1\u6709\u4eba\u77e5\u9053\u5982\u4f55\u6709\u6548\u5730\u8bad\u7ec3\u591a\u5c42\u7684\u524d\u9988\u795e\u7ecf\u7f51\u7edc\uff0c\u5f53\u65f6\u4e3b\u6d41\u7684\u65b9\u6cd5\u8fd8\u662f\u9010\u5c42\u8bad\u7ec3\uff0c\u4f46\u8fd9\u79cd\u5c40\u90e8\u7684\u8bad\u7ec3\u65b9\u6cd5\u7684\u6709\u6548\u6027\u8fdc\u4e0d\u53ca\u53cd\u5411\u4f20\u64ad\u3002<\/span><\/p>\n


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

    \u90a3\u4e48\u5f88\u81ea\u7136\u7684\u4e00\u4e2a\u95ee\u9898\u662f\uff1a\u5927\u8111\u5230\u5e95\u4f7f\u7528\u4e86\u600e\u6837\u7684\u5b66\u4e60\u6cd5\u5219<\/strong>\uff1f\u4ece\u795e\u7ecf\u79d1\u5b66\u548c\u751f\u7269\u5b66\u65b9\u9762\u7684\u8bc1\u636e\u6765\u770b\uff0c\u5927\u8111\u4e0d\u592a\u53ef\u80fd\u5728\u505a\u53cd\u5411\u4f20\u64ad\uff0cZador Anthony et. Al. [15]\u8ba4\u4e3a\u5927\u8111\u5229\u7528\u7075\u6d3b\u7684\u9012\u5f52\u7ed3\u6784\u548c\u673a\u5236\u5728\u7ebf<\/span>\uff08online\uff09<\/span>\u4e14\u9ad8\u6548\u5730\u89e3\u51b3temporal credit assignment\u95ee\u9898\uff0c\u4f46\u5177\u4f53\u7684\u673a\u5236\u548c\u539f\u7406\u8fd8\u6ca1\u6709\u88ab\u5f88\u597d\u5730\u7406\u89e3\u3002\u4f46\u662f\u5f88\u591a\u8bc1\u636e\u8868\u660e\uff0c\u5927\u8111\u7684\u5b66\u4e60\u673a\u5236\u5f88\u6709\u53ef\u80fd\u662f\u5c40\u90e8<\/span>\uff08local\uff09<\/span>\u7684\uff0c\u5c40\u90e8\u5b66\u4e60\u673a\u5236\u7684\u597d\u5904\u5728\u4e8e\u4e0d\u9700\u8981\u540c\u6b65\u5168\u5c40\u7684\u4fe1\u606f\uff0c\u5e76\u4e14\u5bf9\u566a\u58f0<\/span>\uff08\u5927\u8111\u4f5c\u4e3a\u4e00\u4e2a\u7269\u7406\u7cfb\u7edf\uff09<\/span>\u6709\u66f4\u597d\u7684\u9c81\u68d2\u6027\u3002Najarro\uff0cE.\u548cRisi\uff0cS.\u5c55\u793a\u4e86\u53ef\u4ee5\u5b66\u4e60\u4e00\u79cd\u5c40\u90e8\u7684\u8d6b\u5e03\u89c4\u5219<\/span>\uff08hebbian rule\uff09<\/span>\u6765\u4f7f\u5f97\u968f\u673a\u521d\u59cb\u5316\u7684\u7f51\u7edc\u5728\u5176\u751f\u547d\u5468\u671f\u5185\u4e0d\u65ad\u81ea\u7ec4\u7ec7\u5176\u6743\u91cd\uff0c\u4f7f\u5177\u6709\u4e0d\u540c\u5f62\u6001\u7684\u667a\u80fd\u4f53\u80fd\u591f\u5b66\u4e60\u884c\u8d70\uff0c\u540c\u65f6\u5bf9\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u672a\u9047\u5230\u7684\u5f62\u6001\u635f\u4f24<\/span>\uff08\u5982\u5728\u6d4b\u8bd5\u4e2d\u53bb\u6389\u591a\u8db3\u673a\u5668\u4eba\u7684\u4e00\u6761\u817f\uff09<\/span>\u5177\u6709\u9c81\u68d2\u6027\u3002<\/span><\/p>\n


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

    \u8fd9\u4f3c\u4e4e\u8868\u660e\u5c40\u90e8\u5b66\u4e60\u89c4\u5219\u6709\u671b\u89e3\u51b3\u5f53\u524dAI\u7cfb\u7edf\u4e0e\u81ea\u7136\u667a\u80fd\u5728\u5b66\u4e60\u65b9\u9762\u7684\u4e00\u4e9b\u663e\u8457\u5dee\u522b\uff1a<\/span><\/p>\n


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

    1\uff09\u8bad\u7ec3\u548c\u63a8\u7406\u9636\u6bb5\u4e92\u76f8\u5206\u79bb\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6743\u91cd\u901a\u8fc7\u968f\u673a\u68af\u5ea6\u4e0b\u964d<\/span>\uff08SGD\uff09<\/span>\u7684\u53cd\u5411\u4f20\u64ad\u8fdb\u884c\u66f4\u65b0\uff0c\u800c\u5728\u63a8\u7406\u9636\u6bb5\u901a\u5e38\u4fdd\u6301\u4e0d\u53d8\uff1b<\/span><\/p>\n


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

    2\uff09\u57fa\u4e8e\u7edf\u8ba1\u5b66\u4e60\u7684\u65b9\u6cd5\u901a\u5e38\u5047\u8bbe\u73af\u5883\u7684\u5206\u5e03\u662f\u9759\u6001\u7684<\/span>\uff08stationarity\uff09\uff08\u6bd4\u5982\u8bad\u7ec3\u662f\u5728randomly shuffle\u8fc7\u7684\u5c01\u95ed\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u7684\uff09<\/span>\uff0c\u8fd9\u4e00\u5047\u5b9a\u5bf9\u4e8e\u81ea\u7136\u6570\u636e\u6d41\u51e0\u4e4e\u4e0d\u53ef\u80fd\u5f97\u5230\u6ee1\u8db3\uff1b<\/span><\/p>\n


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

    3\uff09\u76ee\u524d\u7684\u5b66\u4e60\u65b9\u6cd5\u5bf9\u4e8e\u975e\u9759\u6001\u7684\u73af\u5883\/\u4efb\u52a1\u5206\u5e03\u65e0\u6cd5\u5b9e\u73b0\u6301\u7eed\u7ec8\u8eab\u5b66\u4e60<\/span>\uff08continual lifelong learning\uff09<\/span>\uff0c\u800c\u4f1a\u78b0\u5230\u707e\u96be\u6027\u9057\u5fd8<\/span>\uff08catastrophic forgetting\uff09<\/span>\u7684\u95ee\u9898\u3002\u800c\u5bf9\u4e8e\u901a\u7528\u5177\u8eab\u667a\u80fd\u6765\u8bf4\uff0c\u5177\u5907\u6301\u7eed\u5b66\u4e60\u80fd\u529b\u81f3\u5173\u91cd\u8981\uff0c\u56e0\u4e3a\u4e16\u754c\u662f\u4e0d\u65ad\u6f14\u5316\u7684\uff0c\u901a\u7528\u667a\u80fd\u4f53\u53ea\u6709\u901a\u8fc7\u4e0d\u65ad\u6539\u53d8\u81ea\u5df1\u5185\u90e8\u7684\u72b6\u6001\u6765\u9002\u5e94\u53d8\u5316\u7684\u73af\u5883\uff0c\u624d\u80fd\u786e\u4fdd\u5176\u80fd\u529b\u7684\u901a\u7528\u6027\u3002\u8bd5\u60f3\u6211\u4eec\u5982\u679c\u60f3\u8ba9\u4e00\u4e2a\u901a\u7528\u667a\u80fd\u89e3\u51b35\u5e74\u4e4b\u540e\u4eba\u7c7b\u5c06\u9047\u5230\u7684\u95ee\u9898\uff0c\u5f88\u663e\u7136\u6211\u4eec\u6ca1\u6709\u529e\u6cd5\u5728\u73b0\u5728\u91c7\u96c6\u5230\u4e0e\u8be5\u4efb\u52a1\u76f8\u5173\u7684\u6570\u636e\uff0c\u53ea\u6709\u8ba9\u5176\u5177\u6709\u5b66\u4e60\u7684\u5143<\/span>\u80fd\u529b\u624d\u80fd\u786e\u4fdd\u5176\u6709\u53ef\u80fd\u89e3\u51b3\u5728\u672a\u6765\u624d\u4f1a\u51fa\u73b0\u7684\u65b0\u4efb\u52a1\uff0c\u56e0\u6b64\u63a2\u7d22\u53d7\u751f\u7269\u542f\u53d1\u7684\u53ef\u5851\u6027<\/span>\uff08plasticity\uff09<\/span>\u548ctemporal credit assignment\u662f\u503c\u5f97\u7814\u7a76\u7684\uff0c\u8fd9\u4e9b\u673a\u5236\u53ef\u80fd\u5728\u5b9e\u73b0\u5177\u8eab\u667a\u80fd\u9002\u5e94\u4e0d\u65ad\u53d8\u5316\u7684\u4e16\u754c\u6240\u9700\u7684\u5728\u7ebf\u6301\u7eed\u5b66\u4e60\u65b9\u9762\u66f4\u52a0\u6709\u6548\u3002<\/span><\/p>\n


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

    \uff083\uff09\u76ee\u6807\u51fd\u6570<\/span><\/strong><\/span><\/h2>\n


    <\/p>\n

    \u5047\u8bbe\u4e16\u754c\u6a21\u578b\u7684\u5b66\u4e60\u53ef\u4ee5\u5f52\u7ed3\u4e3a\u4f18\u5316\u4e00\u4e2a\u76ee\u6807\u51fd\u6570\uff0c\u90a3\u4e48\u8fd9\u4e2a\u76ee\u6807\u51fd\u6570\u53ef\u80fd\u662f\u4ec0\u4e48\u5462\uff1fDeepmind\u7684\u521b\u59cbCTO Shane Legg\u548c\u5176\u5bfc\u5e08Marcus Hutter\u63d0\u51fa\uff0c\u5728\u88ab\u52a8\u73af\u5883\u4e2d<\/span>\uff08\u6307\u667a\u80fd\u4f53\u4e0d\u80fd\u76f4\u63a5\u91c7\u53d6\u884c\u52a8\u6765\u5e72\u9884\u73af\u5883\uff09<\/span>\uff0c\u4e00\u4e2a\u666e\u9002\u7684\u667a\u80fd\u8861\u91cf\u6807\u51c6\u662f\u9884\u6d4b\u672a\u6765\u7684\u80fd\u529b [17]\u3002Yi Ma, Doris Tsao\u548cHeung-Yeung Shum\u63d0\u51fa\uff0c\u5b66\u4e60\u7684\u76ee\u7684\u662f\u627e\u5230\u6570\u636e\u4e2d\u7684\u6a21\u5f0f\u548c\u89c4\u5f8b\uff0c\u8fd9\u53ef\u4ee5\u5f62\u5f0f\u5316\u4e3a\u4f18\u5316\u4e00\u4e2a\u53eb\u505arate distortion\u7684\u4fe1\u606f\u5ea6\u91cf\uff0c\u4f18\u5316\u8fd9\u4e00\u76ee\u6807\u7684\u672c\u8d28\u662f\u5728\u505a\u4fe1\u606f\u538b\u7f29\u548c\u8868\u5f81\u7684\u7a00\u758f\u5316<\/span>\uff08sparsification\uff09<\/span>\uff0c\u5c24\u5176\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\u7684\u662f\uff0c\u4ee5\u68af\u5ea6\u4e0b\u964d\u63a8\u5bfc\u4f18\u5316rate distortion\u7684\u65b9\u6cd5\u53ef\u4ee5\u81ea\u7136\u800c\u7136\u5730\u5f97\u5230\u7c7b\u4f3c\u4e8eResNet\/Transformer\u7684\u67b6\u6784 [18]\u3002\u4ece\u67d0\u79cd\u7a0b\u5ea6\u4e0a\u8bf4\uff0c\u6574\u4e2a\u4eba\u7c7b\u7684\u79d1\u5b66\u53f2\u4e5f\u53ef\u4ee5\u88ab\u8ba4\u4e3a\u662f\u538b\u7f29<\/span>\uff08\u627e\u5230\u89c4\u5f8b\uff09<\/span>\u548c<\/span>\uff08\u57fa\u4e8e\u627e\u5230\u7684\u89c4\u5f8b\u6765\uff09<\/span>\u9884\u6d4b\u3002\u5b9e\u9645\u4e0a\uff0c\u9884\u6d4b\u548c\u538b\u7f29\u4e4b\u95f4\u662f\u5177\u6709\u7b49\u4ef7\u6027\u7684\uff0c\u8be6\u7ec6\u7684\u53ef\u4ee5\u53c2\u8003 [19]\u3002<\/span><\/p>\n


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

    \u7136\u800c\uff0c\u4ec5\u51ed\u9884\u6d4b\/\u538b\u7f29\u5e76\u4e0d\u80fd\u786e\u4fdd\u80fd\u5b66\u5230\u4e16\u754c\u6a21\u578b\uff0c\u56e0\u4e3a\u901a\u8fc7\u4e0e\u73af\u5883\u4ea4\u4e92\u4e3b\u52a8\u83b7\u53d6\u5173\u4e8e\u4e16\u754c\u7684\u4fe1\u606f\u5bf9\u4e8e\u5b66\u4e60\u4e16\u754c\u6a21\u578b\u6765\u8bf4\u662f\u5fc5\u4e0d\u53ef\u5c11\u7684\u8fc7\u7a0b\u3002\u6b64\u5916\uff0c\u4e00\u4e2a\u5177\u6709\u666e\u9002\u6027\u7684\u7c7b\u4eba\u7684\u667a\u80fd\u4f53\u5e94\u8be5\u80fd\u591f\u81ea\u4e3b\u5730\u4e60\u5f97\u65b0\u6280\u80fd\u548c\u77e5\u8bc6\uff0c\u5982\u884c\u8d70\u3001\u8bf4\u8bdd\u3001\u64cd\u4f5c\u7269\u4f53\u3001\u8fdb\u884c\u601d\u7ef4\u5b9e\u9a8c\uff0c\u751a\u81f3\u53d1\u660e\u6570\u5b66\u548c\u7269\u7406\u7406\u8bba\u3002\u662f\u5426\u5b58\u5728\u4e00\u4e2a\u666e\u904d\u7684\u76ee\u6807\uff0c\u4f7f\u5f97\u8fd9\u4e9b\u884c\u4e3a\u5f97\u4ee5\u6d8c\u73b0\u5462\uff1fKarl Friston\u63d0\u51fa\u7684\u81ea\u7531\u80fd\u539f\u7406<\/strong><\/span>\uff08FEP\uff09<\/span>\u4f3c\u4e4e\u662f\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\u7684\u4e00\u4e2a\u5f88\u6709\u5e0c\u671b\u7684\u5019\u9009\u9879 [20]\u3002<\/span>\u81ea\u7531\u80fd\u539f\u7406\u5728\u4e3b\u52a8\u611f\u77e5<\/span><\/span>\uff08active perception\uff09<\/span><\/span>\u7684\u6846\u67b6\u4e0b\u7edf\u4e00\u4e86\u884c\u52a8\u548c\u611f\u77e5\uff0c\u901a\u8fc7\u6700\u5c0f\u5316\u53d8\u5206<\/span><\/span>\uff08variational\uff09<\/span><\/span>\u81ea\u7531\u80fd\uff0c\u9f13\u52b1\u667a\u80fd\u4f53\u4e3b\u52a8\u83b7\u53d6\u4fe1\u606f\u6765\u5b66\u4e60\u4e16\u754c\u6a21\u578b\uff0c\u4ee5\u51cf\u5c11\u5176\u5bf9\u672a\u6765\u89c2\u6d4b\u7684\u610f\u5916\u7a0b\u5ea6<\/span>\uff08surprise\uff09<\/span>\u3002<\/span>\u4e0e\u5f3a\u5316\u5b66\u4e60\u4e0d\u540c\u7684\u662f\uff0c\u5728\u4e3b\u52a8\u611f\u77e5\u7684\u6846\u67b6\u4e0b\u667a\u80fd\u4f53\u5e76\u4e0d\u9700\u8981\u5916\u90e8\u63d0\u4f9b\u7684\u4e0e\u4efb\u52a1\u76f8\u5173\u7684\u5956\u52b1\uff0c\u8fd9\u5c31\u907f\u514d\u4e86\u4eba\u4e3a\u8bbe\u8ba1\u5956\u52b1\u51fd\u6570\u4f1a\u78b0\u5230\u7684\u56f0\u96be\u3002\u7136\u800c\uff0c\u5728\u5b9e\u8df5\u4e2d\u5e94\u7528FEP\u4f1a\u9762\u4e34\u4e0e\u53d8\u5206\u63a8\u65ad<\/span>\uff08variational inference\uff09<\/span>\u7c7b\u4f3c\u7684\u6311\u6218\uff1a\u5f88\u96be\u627e\u5230\u5408\u9002\u7684\u53c2\u6570\u5316\u51fd\u6570\u65cf<\/span>\uff08\u65e2\u80fd\u591f\u8868\u8fbe\u590d\u6742\u7684\u6982\u7387\u5206\u5e03\uff0c\u53c8\u662f\u8ba1\u7b97\u4e0a\u91cf\u4e0a\u53ef\u63a7\u7684\uff09<\/span>\u3002<\/span><\/p>\n


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

    \u5728\u89e3\u51b3\u9ad8\u7ef4\u6570\u636e\u5206\u5e03\u7684\u8ba1\u7b97\u53ef\u884c\u6027\u65b9\u9762\uff0cYann Lecun\uff0cYoshua Bengio\u7b49\u5b66\u8005\u505a\u8fc7\u4e0d\u5c11\u5c1d\u8bd5\u3002\u6bd4\u5982\uff0cYann\u5728 [10]\u4e2d\u6307\u51fa\uff0c\u5bf9\u4e8e\u9ad8\u7ef4\u6570\u636e\u6765\u8bf4\uff0c\u8d85\u8fc7\u4e8c\u9636\u7684\u7edf\u8ba1\u91cf\u51e0\u4e4e\u90fd\u662f\u5728\u8ba1\u7b97\u4e0a\u6ca1\u6709\u5b9e\u9645\u53ef\u64cd\u4f5c\u6027\u7684\uff0c\u6570\u5b66\u4e0a\u6765\u8bf4\u5b66\u4e60\u6982\u7387\u5206\u5e03\u7684\u4e00\u5927\u6311\u6218\u662f\u5982\u4f55\u9ad8\u6548\u5730\u8ba1\u7b97\u51fa\u5f52\u4e00\u5316\u7cfb\u6570<\/span>\uff08partition function\uff09<\/span>\uff0c\u5927\u8111\u4e5f\u4e0d\u592a\u53ef\u80fd\u53bb\u5b66\u4e60\u6574\u4e2a\u9ad8\u7ef4\u6570\u636e\u7684\u5206\u5e03\uff0c\u636e\u6b64\u4ed6\u63d0\u51fa\u57fa\u4e8e\u80fd\u91cf\u7684\u6a21\u578b<\/strong><\/span>\uff08energy-based models\uff09<\/span>\uff0c\u5373\u4e0d\u76f4\u63a5\u53c2\u6570\u5316\u6982\u7387\u5206\u5e03\uff0c\u800c\u662f\u53c2\u6570\u5316\u4e00\u4e2a\u4e0d\u9700\u8981\u5f52\u4e00\u5316\u7684\u51fd\u6570\u6765\u8868\u8fbe\u6570\u636e\u95f4\u7684\u517c\u5bb9\u6027<\/span>\uff08\u76f4\u89c2\u4e0a\u6765\u8bf4\u5c31\u662f\u63cf\u8ff0\u54ea\u4e9b\u60c5\u51b5\u53ef\u80fd\u53d1\u751f\uff0c\u54ea\u4e9b\u60c5\u51b5\u4e0d\u592a\u53ef\u80fd\u53d1\u751f\uff0c\u54ea\u4e9b\u60c5\u51b5\u51e0\u4e4e\u4e0d\u53ef\u80fd\u53d1\u751f\uff09<\/span>\u3002<\/span><\/p>\n


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

    \u76f8\u6bd4\u4e8eFEP\uff0c\u4e00\u4e2a\u66f4\u53ef\u884c\u7684\u65b9\u6cd5\u662f\u901a\u8fc7\u5bf9\u6297\u8bad\u7ec3\u7684\u65b9\u5f0f\u6765\u5b66\u4e60\u4e16\u754c\u6a21\u578b<\/strong>\uff0c\u5982Herrmann V\u3001Kirsch L\u548cSchmidhuber J\u8bbe\u8ba1\u4e86\u4e00\u79cd\u7b97\u6cd5\uff0c\u8ba9\u4e00\u4e2a\u540d\u4e3a\u4e16\u754c\u6a21\u578b\u7684\u795e\u7ecf\u7f51\u7edc\u548c\u4e00\u4e2a\u540d\u4e3a\u63a7\u5236\u5668\u7684\u795e\u7ecf\u7f51\u7edc\u4e4b\u95f4\u8fdb\u884c\u5bf9\u6297\u6027\u535a\u5f08\uff0c\u5176\u4e2d\u63a7\u5236\u5668\u7684\u5185\u5728\u5956\u52b1\u53ef\u4ee5\u662f\u4e16\u754c\u6a21\u578b\u7684\u4fe1\u606f\u589e\u76ca\u3001\u7b97\u6cd5\u538b\u7f29\u8fdb\u5ea6\u6216\u5728\u5177\u6709\u53ef\u8ba1\u7b97\u7b54\u6848\u7684\u66f4\u62bd\u8c61\u95ee\u9898\u4e0a\u7684\u6b63\u786e\u6027\u7b49\u7b49\uff0c\u4ed6\u4eec\u5c55\u793a\u4e86\u901a\u8fc7\u795e\u7ecf\u7f51\u7edc\u6743\u91cd\u7f16\u7801\u7684\u81ea\u6211\u53d1\u660e\u601d\u7ef4\u5b9e\u9a8c\u8fdb\u884c\u5b66\u4e60\u7684\u53ef\u80fd [21]\u3002\u6b64\u5916\uff0cMa\u3001Tsao\u548cShum\u7684\u7814\u7a76\u8868\u660e [18]\uff0c\u4ee5\u4fe1\u606f\u538b\u7f29\u4f5c\u4e3a\u76ee\u6807\u51fd\u6570\uff0c\u5e76\u901a\u8fc7\u4e00\u4e2a\u4f20\u611f\u5668<\/span>\uff08\u7f16\u7801\u5668\uff09<\/span>\u7f51\u7edc\u548c\u4e00\u4e2a\u63a7\u5236\u5668<\/span>\uff08\u89e3\u7801\u5668\uff09<\/span>\u7f51\u7edc\u4e4b\u95f4\u7684\u95ed\u73af\u96f6\u548c\u535a\u5f08\u53ef\u4ee5\u6709\u6548\u5730\u907f\u514d\u795e\u7ecf\u5d29\u6e83 [22]\uff0c\u8fd9\u662f\u4f20\u7edf\u7684<\/span>\uff08\u81ea\uff09<\/span>\u76d1\u7763\u5b66\u4e60\u9762\u4e34\u7684\u4e00\u4e2a\u5f88\u5927\u7684\u6311\u6218\u3002\u8fd9\u4e9b\u53d1\u73b0\u90fd\u8868\u660e\uff0c\u538b\u7f29\u3001\u4fe1\u606f\u589e\u76ca\u548c\u975e\u5408\u4f5c\u535a\u5f08\u53ef\u80fd\u662f\u975e\u5e38\u666e\u9002\u7684\u76ee\u6807\/\u673a\u5236\uff0c\u5e76\u4e14\u6709\u52a9\u4e8e\u901a\u7528\u667a\u80fd\u7684\u6d8c\u73b0\uff0c\u503c\u5f97\u8fdb\u4e00\u6b65\u7814\u7a76\u53ef\u6269\u5c55\u7684\u8ba1\u7b97\u65b9\u6cd5\u6765\u5efa\u6a21\u548c\u4f18\u5316\u8fd9\u4e9b\u76ee\u6807\u51fd\u6570\u3002<\/span><\/p>\n


    <\/h2>\n

    \uff084\uff09\u6570\u636e\/\u73af\u5883<\/span><\/strong><\/span><\/h2>\n


    <\/p>\n

    \u8bad\u7ec3\u667a\u80fd\u4f53\u7684\u6570\u636e\u548c\u73af\u5883\u4e5f\u5bf9\u901a\u7528\u667a\u80fd\u7684\u4ea7\u751f\u8d77\u5230\u5173\u952e\u7684\u4f5c\u7528\u3002\u76ee\u524d\u7684\u4e3b\u6d41\u7684\u505a\u6cd5\u5728\u91c7\u96c6\u6570\u636e\u548c\u8bad\u7ec3\u6a21\u578b\u4e0a\u662f\u5206\u79bb\u7684\uff0c\u800c\u6211\u4eec\u4e5f\u770b\u5230\u8fd9\u5bf9\u4e8e\u590d\u6742\u7684\u5b9e\u9645\u95ee\u9898\u662f\u6709\u5c40\u9650\u6027\u7684\uff0c\u6bd4\u5982\u81ea\u52a8\u9a7e\u9a76\u8f66\u9700\u8981\u5728\u5404\u79cd\u4e0d\u540c\u7684\u8def\u51b5\u548c\u9053\u8def\u5e03\u7f6e\u4e0a\u5b9e\u73b0\u4eba\u7c7b\u6c34\u5e73\u7684\u9a7e\u9a76\u6280\u672f\uff0c\u5c31\u5f88\u5bb9\u6613\u53d1\u73b0\u5728\u56fa\u5b9a\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u662f\u884c\u4e0d\u901a\u7684\uff0c\u800c\u9700\u8981\u4e0d\u65ad\u7528\u5f53\u524d\u5b66\u5230\u7684\u9a7e\u9a76\u7b56\u7565\u53bb\u91c7\u96c6\u65b0\u6570\u636e\uff0c\u5e76\u7528\u65b0\u6570\u636e\u8fdb\u4e00\u6b65\u66f4\u65b0\u9a7e\u9a76\u7b56\u7565\uff0c\u5b9e\u73b0\u53cd\u9988\u95ed\u73af\u3002\u5176\u5b9e\u4eba\u7c7b\u9a7e\u9a76\u4e5f\u662f\u7c7b\u4f3c\u7684\uff0c\u5728\u9a7e\u6821\u4e2d\u5b66\u4e60\u5b8c\u4ee5\u540e\uff0c\u4e0a\u4e86\u5927\u9a6c\u8def\u8fd8\u9700\u8981\u4e00\u6bb5\u65f6\u95f4\u7684\u5b66\u4e60\u9002\u5e94\uff0c\u5728\u4e00\u4e2a\u57ce\u5e02\u4e60\u60ef\u4e86\u4ee5\u540e\uff0c\u5230\u4e86\u53e6\u4e00\u4e2a\u8def\u51b5\u5dee\u522b\u5f88\u5927\u7684\u57ce\u5e02\uff0c\u53c8\u9700\u8981\u4e00\u6bb5\u65f6\u95f4\u7684\u5b66\u4e60\u9002\u5e94\uff0c\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\u5927\u8111\u4f9d\u9760\u8ba4\u77e5\u80fd\u529b\u4e0d\u65ad\u4ea7\u751f\u65b0\u7684\u89e3\u51b3\u4e13\u7528\u4efb\u52a1\u7684\u795e\u7ecf\u56de\u8def\uff0c\u56e0\u6b64\u6211\u4eec\u80fd\u611f\u89c9\u5230\u5728\u8def\u51b5\u53d1\u751f\u53d8\u5316\u65f6\u6211\u4eec\u4f1a\u66f4\u52a0\u5168\u795e\u8d2f\u6ce8<\/span>\uff08\u7cfb\u7edf\u4e8c\u7684\u6ce8\u610f\u529b\u4ecb\u5165\uff09<\/span>\uff0c\u4e00\u6bb5\u65f6\u95f4\u4e4b\u540e\u4fbf\u4e0d\u518d\u9700\u8981\u6ce8\u610f\u529b\u4ecb\u5165\u6162\u6162\u53d8\u8f7b\u677e\u4e86<\/span>\uff08\u4e13\u7528\u795e\u7ecf\u56de\u8def\u5df2\u7ecf\u5f62\u6210\uff09<\/span>\u3002<\/span><\/p>\n


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

    \u8fdb\u4e00\u6b65\u5730\uff0c\u6211\u8ba4\u4e3a\u5728\u56fa\u5b9a\u7684\u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u6a21\u578b\/<\/span>\uff08\u5f3a\u5316\u5b66\u4e60\uff09<\/span>\u5728\u5c01\u95ed\u7684\u73af\u5883\u4e0b\u8bad\u7ec3\u667a\u80fd\u4f53\u5b58\u5728\u7684\u4e00\u4e2a\u6839\u672c\u6027\u95ee\u9898\uff0c\u662f\u667a\u80fd\u4f53\u5f80\u5f80\u4e0d\u9700\u8981\u638c\u63e1\u5b66\u4e60\u7684\u5143<\/span>\u80fd\u529b<\/span>\uff08\u5373\u5bf9\u5e94\u4e8e\u4e0a\u8ff0\u4eba\u7c7b\u9002\u5e94\u65b0\u72b6\u51b5\u7684\u80fd\u529b\uff09<\/span>\u5c31\u80fd\u591f\u89e3\u51b3\u4efb\u52a1\uff0c\u90a3\u4e48\u5f88\u81ea\u7136\u5730\uff0c\u667a\u80fd\u4f53\u5c31\u4e0d\u4f1a\u4e60\u5f97\u8fd9\u79cd\u5143\u80fd\u529b\u3002\u4e0e\u4e4b\u76f8\u5173\u5730\uff0cEvan Hubinger\u7b49\u4eba [23]\u63d0\u51fa\u4e86mesa-optimizer<\/strong>\u7684\u6982\u5ff5\uff0c\u5e76\u8ba4\u4e3a\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u5b58\u5728mesa-optimization\u7684\u73b0\u8c61<\/span>\uff08\u6b64\u5904\u503c\u5f97\u6ce8\u610f\u7684\u662fmesa-optimization\u4e0emeta-optimization\u662f\u4e24\u4e2a\u76f8\u4f3c\u4f46\u4e0d\u540c\u7684\u6982\u5ff5\uff0c\u540e\u8005\u662f\u6307\u5728\u66f4\u9ad8\u7684\u5c42\u7ea7\u4e0a\u5bf9\u4f18\u5316\u7b97\u6cd5\u672c\u8eab\u8fdb\u884c\u4f18\u5316\uff09<\/span>\uff0c\u5373\u4e00\u4e2a\u57fa\u4f18\u5316\u5668<\/span>\uff08base optimizer\uff09<\/span>\uff0c\u8fd9\u91cc\u5f80\u5f80\u6307\u7684\u662f\u57fa\u4e8e\u68af\u5ea6\u4e0b\u964d\u548c\u53cd\u5411\u4f20\u64ad\u7684\u4f18\u5316\u7b97\u6cd5\uff0c\u53ef\u4ee5\u7528\u6765\u4f18\u5316\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\uff0c\u5e76\u4e14\u4f18\u5316\u51fa\u6765\u7684\u795e\u7ecf\u7f51\u7edc\u672c\u8eab\u4e5f\u4f1a\u53d8\u6210\u4e00\u4e2a\u4f18\u5316\u5668<\/strong><\/span>\uff08learned optimizer\/mesa-optimizer\uff09<\/span>\u3002<\/p>\n

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

    \u6bd4\u5982\u5728\u5927\u89c4\u6a21\u6570\u636e\u4e0a\u8bad\u7ec3\u8fc7\u7684transformer\u4f1a\u6d8c\u73b0\u51fain-context learning\u7684\u80fd\u529b [24]\uff0c\u5e76\u4e14\u53d1\u73b0\u8fd9\u79cd\u80fd\u529b\u7684\u6d8c\u73b0\u4e0e\u6570\u636e\u96c6\u4e2d\u4efb\u52a1\u7684\u591a\u6837\u6027\u5448\u73b0\u56e0\u679c\u5173\u7cfb\uff0c\u8fd9\u79cd\u80fd\u529b\u672c\u8eab\u5c31\u7b49\u4ef7\u4e8e\u5728\u505a\u67d0\u79cd\u5b66\u4e60\u4f18\u5316 [25]\u3002\u90a3\u4e48\u5982\u4f55\u7406\u89e3\u8fd9\u79cd\u6d8c\u73b0\u73b0\u8c61\u5462\uff1fEvan Hubinger\u7b49\u4eba [23]\u63d0\u4f9b\u4e86\u4e00\u79cd\u770b\u6cd5\uff1a\u5f53\u4efb\u52a1\u73af\u5883\u7279\u5f81\u5305\u542b\u591a\u6837\u6027\u3001\u591a\u5206\u679d\uff0c\u6d89\u53ca\u65b0\u9896\u60c5\u5883\u6216\u4efb\u52a1\u5b9e\u4f8b\u65f6\uff0c\u66f4\u52a0\u5bb9\u6613\u6d8c\u73b0\u51famesa-optimizer<\/strong>\u3002\u8fd9\u662f\u4ec0\u4e48\u539f\u56e0\u5462\uff1f<\/span>\u53ef\u4ee5\u8fd9\u6837\u7406\u89e3\uff1a\u5f53\u4efb\u52a1\u73af\u5883\u7f3a\u4e4f\u4e0a\u8ff0\u7279\u5f81\u65f6\uff0c\u7edd\u5927\u591a\u6570\u7684\u7b56\u7565\u4f18\u5316<\/span>\uff08policy optimization\uff09<\/span>\u53ef\u4ee5\u5b8c\u5168\u7531\u57fa\u4f18\u5316\u5668\u6765\u5b8c\u6210\u5c31\u80fd\u8fbe\u5230\u5f88\u597d\u7684\u6548\u679c\uff0c\u5b66\u5230\u7684\u7b56\u7565\u672c\u8d28\u4e0a\u662f\u4e00\u79cd\u8c03\u6574\u5730\u7279\u522b\u597d\u7684\u542f\u53d1\u5f0f\u65b9\u6cd5\uff0c\u6709\u70b9\u50cf\u8001\u5e08\u5085\u4e0d\u5047\u601d\u7d22\u7eaf\u51ed\u7ecf\u9a8c\u624b\u611f\u5b8c\u6210\u4e00\u9879\u91cd\u590d\u6027\u5f88\u5f3a\u7684\u6280\u80fd\uff1b\u76f8\u53cd\uff0c\u5982\u679c\u4efb\u52a1\u73af\u5883\u53d8\u5316\u591a\u7aef\u6781\u4e3a\u590d\u6742\u65f6\uff0c\u7eaf\u51ed\u7ecf\u9a8c\u62cd\u8111\u888b\u5c31\u4f1a\u5931\u6548\uff0c\u6bd4\u5982\u8981\u8bad\u7ec3\u4e00\u4e2a\u4f1a\u505a\u6570\u5b66\u5965\u8d5b\u9898\u7684\u667a\u80fd\u4f53\uff0c\u9898\u76ee\u5343\u53d8\u4e07\u5316\uff0c\u4e0d\u53ef\u80fd\u4ec5\u51ed\u8fc7\u53bb\u7684\u7ecf\u9a8c\u5c31\u505a\u5bf9\u6bcf\u4e00\u9053\u65b0\u51fa\u73b0\u7684\u9898\uff0c\u90a3\u4e48\u53ea\u6709\u5f53meta-optimizer\u672c\u8eab\u4e5f\u5177\u6709\u4f18\u5316\u80fd\u529b\u624d\u80fd\u591f\u89e3\u51b3\u8fd9\u79cd\u60c5\u51b5\u3002<\/span><\/p>\n


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

    \u518d\u4e3e\u4e24\u4e2a\u4f8b\u5b50\uff1a1\uff09\u5982\u679c\u6211\u4eec\u60f3\u8bad\u7ec3\u4e00\u4e2a\u4f1a\u8d70\u8ff7\u5bab\u7684\u667a\u80fd\u4f53\uff0c\u8ff7\u5bab\u7684\u5f62\u72b6\u5343\u53d8\u4e07\u5316\uff0c\u8981\u60f3\u901a\u8fc7\u5b66\u4e60\u72b6\u6001\u53d8\u91cf\u5230\u884c\u52a8\u7684\u6620\u5c04<\/span>\uff08\u7c7b\u4f3c\u4e8e\u6839\u636e\u76f4\u89c9\u6765\u8499\uff09<\/span>\u51e0\u4e4e\u4e0d\u53ef\u80fd\u89e3\u51b3\uff0c\u4f46\u5176\u5b9e\u89e3\u51b3\u4efb\u610f\u8ff7\u5bab\u95ee\u9898\u7684\u7b97\u6cd5\u53ea\u9700\u8981\u51e0\u884c\u4ee3\u7801\u7684\u903b\u8f91<\/span>\uff08\u6bd4\u5982\u5e7f\u5ea6\u4f18\u5148\u641c\u7d22\uff09<\/span>\uff0c\u56e0\u6b64\u66f4\u597d\u7684\u89e3\u51b3\u65b9\u6cd5\u662f\u5b66\u5230\u8fd9\u6837\u4e00\u79cd\u5728\u7ebf\u4f18\u5316\u7684\u7b97\u6cd5\u6765\u89e3\u51b3\u8fd9\u4e2a\u8d70\u8ff7\u5bab\u7684\u4efb\u52a1\uff1b<\/span>2\uff09\u6211\u4eec\u6240\u5728\u7684\u5730\u7403\u4e0a\u73af\u5883\u5343\u53d8\u4e07\u5316\uff0c\u81ea\u7136\u6f14\u5316\u4f5c\u4e3a\u4e00\u79cd\u57fa\u4f18\u5316\u5668\u6ca1\u6709\u529e\u6cd5\u76f4\u63a5\u4f18\u5316\u51fa\u4e00\u79cd\u80fd\u89e3\u51b3\u4efb\u4f55\u95ee\u9898\u7684\u751f\u7269\u4f53\uff0c\u56e0\u4e3a\u89e3\u51b3\u4efb\u4f55\u95ee\u9898\u4e5f\u5305\u542b\u4e86\u89e3\u51b3\u672a\u6765\u53ef\u80fd\u9700\u8981\u89e3\u51b3\u7684\u95ee\u9898\uff0c\u8fd9\u4e2a\u95ee\u9898\u7a7a\u95f4\u662f\u65e0\u7a77\u5927\u7684\uff0c\u56e0\u6b64\u81ea\u7136\u6f14\u5316\u6700\u7ec8\u4f18\u5316\u51fa\u4e86\u50cf\u4eba\u7c7b\u548c\u52a8\u7269\u8fd9\u6837\u7684\u901a\u7528\u667a\u80fd\/mesa-optimizer\uff0c\u4ed6\u4eec\u53ef\u4ee5\u901a\u8fc7\u4e0d\u65ad\u5730\u5b66\u4e60\u6765\u9002\u5e94\u53d8\u5316\u7684\u73af\u5883\u3002<\/span>\u7c7b\u4f3c\u7684\u89c2\u70b9\u540c\u6837\u53ef\u4ee5\u7528\u6765\u89e3\u91caGPT\u7684\u901a\u7528\u4efb\u52a1\u89e3\u51b3\u80fd\u529b\uff0cGPT\u6a21\u578b\u5728\u4e92\u8054\u7f51\u7ea7\u522b\u7684\u8bed\u8a00\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u8981\u80fd\u591f\u5728\u5982\u6b64\u591a\u6837\u5316\u7684\u8bad\u7ec3\u96c6\u4e0a\u964d\u4f4e\u635f\u5931\u51fd\u6570\uff0c\u5f88\u53ef\u80fd\u5fc5\u987b\u6d8c\u73b0\u51fa\u7c7b\u4f3c\u4e8ein-context learning\u8fd9\u6837\u5728\u7ebf\u5b66\u4e60\u4f18\u5316\u7684\u80fd\u529b\u3002<\/span><\/p>\n


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

    \u66f4\u8fd1\u4e00\u6b65\u5730\uff0c<\/span>\u6211\u8ba4\u4e3a\u4e00\u4e2a\u9002\u5408\u901a\u7528\u667a\u80fd\u6d8c\u73b0\u7684\u73af\u5883\u5e94\u5f53\u81f3\u5c11\u5177\u6709\u5f00\u653e\u5f0f<\/span>\uff08opened-endness\uff09<\/span>\u975e\u5e73\u7a33\u6027<\/span>\uff08non-stationary\uff09<\/span>\u7684\u7279\u70b9\uff0c\u6bd4\u5982\u8bf4\u6211\u4eec\u6240\u751f\u5b58\u7684\u8fd9\u4e2a\u4e16\u754c\uff0c\u4e0d\u65ad\u5730\u4f1a\u6709\u65b0\u7684\u590d\u6742\u6027\u7684\u51fa\u73b0<\/span><\/strong>\u3002\u76f8\u5bf9\u5730\uff0c\u76ee\u524d\u7684\u7edd\u5927\u591a\u6570\u5f3a\u5316\u5b66\u4e60\u7684\u73af\u5883\u90fd\u662f\u6709\u9650\u7684\uff0c\u7c7b\u4f3c\u4e8e\u5355\u673a\u6e38\u620f\u603b\u6709\u901a\u5173\u7684\u65f6\u5019\uff0c\u5728\u901a\u5173\u4ee5\u540e\u6ca1\u6709\u4efb\u4f55\u8fdb\u4e00\u6b65\u5b66\u4e60\u9002\u5e94\u65b0\u73af\u5883\u7684\u52a8\u673a\u548c\u5fc5\u8981\u6027\u4e86\uff0c\u90a3\u4e48\u81ea\u7136\u4e0d\u9700\u8981\u6d8c\u73b0\u51fa\u9002\u5e94\u6027\u548c\u901a\u7528\u5b66\u4e60\u80fd\u529b\u3002<\/span><\/p>\n


    <\/p>\n


    <\/p>\n

    \n
    \n
    \n
    \n
    \n

    <\/section>\n

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

    6. \u603b\u7ed3<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

    \n

    <\/section>\n

    <\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


    <\/p>\n

    \u672c\u6587\u4ecb\u7ecd\u4e86\u57fa\u4e8e\u76ee\u524d\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\u7684\u5177\u8eab\u667a\u80fd\u7814\u7a76\u6d89\u53ca\u5230\u7684\u65b9\u6cd5<\/span>\uff08\u6a21\u4eff\u5b66\u4e60\u548c\u5f3a\u5316\u5b66\u4e60\uff09<\/span>\u53ca\u5176\u9762\u4e34\u7684\u6311\u6218\uff0c\u5206\u6790\u4e86\u8fd1\u671f\u57fa\u4e8e\u57fa\u7840\u6a21\u578b\u7684\u5de5\u4f5c\u5728\u901a\u5f80\u901a\u7528\u5177\u8eab\u667a\u80fd\u9053\u8def\u4e0a\u53d6\u5f97\u7684\u8fdb\u5c55\u53ca\u5176\u5c40\u9650\u6027\u3002\u5728\u6b64\u57fa\u7840\u4e0a\u8ba8\u8bba\u4e86\u4ee5\u76ee\u524d\u7684\u65b9\u6cd5\u8bad\u7ec3\u5177\u8eab\u667a\u80fd\u7cfb\u7edf\u5c06\u9762\u4e34\u7684\u4e00\u4e2a\u6839\u672c\u6027\u6311\u6218<\/span>\uff08\u4efb\u52a1\u6307\u5b9a\u95ee\u9898\uff09<\/span>\u4ee5\u53ca\u5176\u672c\u8d28\u539f\u56e0<\/span>\uff08\u7f3a\u4e4f\u7ed3\u6784\u5316\u5c42\u7ea7\u5316\u62bd\u8c61\uff0c\u4e16\u754c\u6a21\u578b\uff09<\/span>\uff0c\u5e76\u8bba\u8bc1\u4e86\u4e16\u754c\u6a21\u578b\u5bf9\u4e8e\u5b9e\u73b0\u5177\u8eab\u901a\u7528\u667a\u80fd\u7684\u5173\u952e\u6027\u3002\u6700\u540e\uff0c\u4ece1\uff09\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\uff0c2\uff09\u5b66\u4e60\u6cd5\u5219\uff0c3\uff09\u76ee\u6807\u51fd\u6570\uff0c4\uff09\u8bad\u7ec3\u6570\u636e\/\u73af\u5883\uff0c\u8fd9\u56db\u4e2a\u65b9\u9762\u8ba8\u8bba\u4e86\u76ee\u524d\u7684\u6df1\u5ea6\u5b66\u4e60\u8303\u5f0f\u5728\u672a\u6765\u7684\u53ef\u80fd\u6539\u8fdb\u65b9\u5411\uff0c\u4ee5\u4fbf\u4f7f\u5f97\u5177\u8eab\u667a\u80fd\u4f53\u80fd\u591f\u5b66\u4e60\u5230\u4e16\u754c\u6a21\u578b\u5e76\u6d8c\u73b0\u51fa\u901a\u7528\u7684\u5b66\u4e60\u80fd\u529b\u3002<\/span><\/p>\n


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

    \n
    \n
    \n

    \u81f4\u8c22<\/span><\/h1>\n

    \u4f5c\u8005\u5728\u6b64\u611f\u8c22\u5cb3\u7389\u6d9b\u535a\u58eb\uff0c\u5f90\u535a\u6587\uff0c\u51af\u777f\u6d0b\u5bf9\u672c\u6587\u63d0\u51fa\u7684\u4fee\u6539\u610f\u89c1\u3002<\/span><\/p>\n<\/section>\n<\/section>\n<\/blockquote>\n


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

    \u5b66\u8005\u7b80\u4ecb<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

    <\/section>\n
    \u6c88\u9a6c\u6210<\/span><\/strong>\uff0c\u4ece\u9ebb\u7701\u7406\u5de5\u5b66\u9662\u673a\u68b0\u7cfb\u53d6\u5f97\u535a\u58eb\u5b66\u4f4d\uff0c\u5e08\u4ece\u7f8e\u56fd\u5de5\u7a0b\u9662\u9662\u58ebJonathan How\uff0c\u535a\u58eb\u671f\u95f4\u4e3b\u653b\u7684\u7814\u7a76\u65b9\u5411\u4e3a\u591a\u667a\u80fd\u4f53\u5f3a\u5316\u5b66\u4e60\u548c\u673a\u5668\u4eba\uff0c\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\uff0c\u673a\u5668\u4eba\u611f\u77e5\uff0c\u81ea\u52a8\u9a7e\u9a76\u7b49\u65b9\u9762\u6709\u8f83\u4e3a\u4e30\u5bcc\u7684\u7814\u7a76\u7ecf\u5386\u3002\u5728\u5305\u62ecMeta Reality Lab\uff0cDexterity\u7b49\u591a\u5bb6\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\u548c\u516c\u53f8\u4ece\u4e8b\u8fc7\u76f8\u5173\u7684\u7814\u53d1\u5de5\u4f5c\u3002<\/span>\u76ee\u524d\u7684\u7814\u7a76\u65b9\u5411\u5305\u62ec\u901a\u7528\u5177\u8eab\u667a\u80fd\uff0c<\/span>\u5bf9\u63a2\u7d22\u667a\u80fd\u73b0\u8c61\u80cc\u540e\u7684\u666e\u9002\u6027\u539f\u7406\u4ee5\u53ca\u672a\u6765\u7684\u5f3a\u4eba\u5de5\u667a\u80fd\u53ef\u80fd\u5bfc\u81f4\u7684\u751f\u5b58\u6027\u98ce\u9669\u5f88\u611f\u5174\u8da3\u3002<\/span><\/section>\n


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


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

    \n
    \n
    \n

    \u53c2\u8003\u6587\u732e<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

    \n
    \n
    \n
    \n


    <\/p>\n

    [1] Lu, Kevin, et al. “Pretrained transformers as universal computation engines.” arXiv preprint arXiv:2103.05247<\/em> 1 (2021).<\/span><\/p>\n

    [2] Jang, Eric, et al. “Bc-z: Zero-shot task generalization with robotic imitation learning.” Conference on Robot Learning<\/em>. PMLR, 2022.<\/span><\/p>\n

    [3] Padalkar, Abhishek, et al. “Open x-embodiment: Robotic learning datasets and rt-x models.” arXiv preprint arXiv:2310.08864<\/em> (2023).<\/span><\/p>\n

    [4] Bahdanau, Dzmitry, et al. “Systematic generalization: what is required and can it be learned?.” arXiv preprint arXiv:1811.12889<\/em> (2018).<\/span><\/p>\n

    [5] Akkaya, Ilge, et al. “Solving rubik’s cube with a robot hand.” arXiv preprint arXiv:1910.07113<\/em> (2019).<\/span><\/p>\n

    [6] Zhuang, Ziwen, et al. “Robot parkour learning.” arXiv preprint arXiv:2309.05665<\/em> (2023).<\/span><\/p>\n

    [7] Ha, David, and J\u00fcrgen Schmidhuber. “World models.” arXiv preprint arXiv:1803.10122<\/em> (2018).<\/span><\/p>\n

    [8] Hafner, Danijar, et al. “Dream to control: Learning behaviors by latent imagination.” arXiv preprint arXiv:1912.01603<\/em> (2019).<\/span><\/p>\n

    [9] Wu, Philipp, et al. “Daydreamer: World models for physical robot learning.” Conference on Robot Learning<\/em>. PMLR, 2023.<\/span><\/p>\n

    [10] LeCun, Yann. “A path towards autonomous machine intelligence version 0.9. 2, 2022-06-27.” Open Review<\/em> 62.1 (2022).<\/span><\/p>\n

    [11] Piloto, Luis S., et al. “Intuitive physics learning in a deep-learning model inspired by developmental psychology.” Nature human behaviour<\/em> 6.9 (2022): 1257-1267.<\/span><\/p>\n

    [12] Agrawal, Pulkit. “The task specification problem.” Conference on Robot Learning<\/em>. PMLR, 2022.<\/span><\/p>\n

    [13] https:\/\/www.lesswrong.com\/posts\/gvzW46Z3BsaZsLc25\/natural-abstractions-key-claims-theorems-and-critiques-1<\/span><\/p>\n

    [14] Kirsch, Louis, et al. “General-purpose in-context learning by meta-learning transformers.” arXiv preprint arXiv:2212.04458<\/em> (2022).<\/span><\/p>\n

    [15] Zador, Anthony, et al. “Catalyzing next-generation artificial intelligence through neuroai.” Nature communications<\/em> 14.1 (2023): 1597.<\/span><\/p>\n

    [16] Bongard, Joshua, and Michael Levin. “There\u2019s plenty of room right here: Biological systems as evolved, overloaded, multi-scale machines.” Biomimetics<\/em> 8.1 (2023): 110.<\/span><\/p>\n

    [17] Legg, Shane, and Marcus Hutter. “Universal intelligence: A definition of machine intelligence.” Minds and machines<\/em> 17 (2007): 391-444.<\/span><\/p>\n

    [18] Ma, Yi, Doris Tsao, and Heung-Yeung Shum. “On the principles of parsimony and self-consistency for the emergence of intelligence.” Frontiers of Information Technology & Electronic Engineering<\/em> 23.9 (2022): 1298-1323.<\/span><\/p>\n

    [19] Gr\u00fcnwald, Peter. “Minimum description length tutorial.” Advances in minimum description length: Theory and applications<\/em> 5 (2005): 1-80.<\/span><\/p>\n

    [20] Friston, Karl. “The free-energy principle: a unified brain theory?.” Nature reviews neuroscience<\/em> 11.2 (2010): 127-138.<\/span><\/p>\n

    [21] Herrmann, Vincent, Louis Kirsch, and J\u00fcrgen Schmidhuber. “Learning One Abstract Bit at a Time Through Self-Invented Experiments Encoded as Neural Networks.” International Workshop on Active Inference<\/em>. Cham: Springer Nature Switzerland, 2023.<\/span><\/p>\n

    [22] Kothapalli, Vignesh, Ebrahim Rasromani, and Vasudev Awatramani. “Neural collapse: A review on modelling principles and generalization.” arXiv preprint arXiv:2206.04041<\/em> (2022).<\/span><\/p>\n

    [23] Hubinger, Evan, et al. “Risks from learned optimization in advanced machine learning systems.” arXiv preprint arXiv:1906.01820<\/em> (2019).<\/span><\/p>\n

    [24] Kirsch, Louis, et al. “General-purpose in-context learning by meta-learning transformers.” arXiv preprint arXiv:2212.04458<\/em> (2022).<\/span><\/p>\n

    [25] Aky\u00fcrek, Ekin, et al. “What learning algorithm is in-context learning? investigations with linear models.” arXiv preprint arXiv:2211.15661<\/em> (2022).<\/span><\/p>\n


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

    \n
    \n

    \uff08\u53c2\u8003\u6587\u732e\u53ef\u4e0a\u4e0b\u6ed1\u52a8<\/strong>\u67e5\u770b\uff09<\/p>\n<\/section>\n<\/section>\n


    <\/p>\n


    <\/p>\n


    <\/p>\n

    \u76f8\u5173\u8bfb\u4e66\u4f1a\u63a8\u8350<\/span><\/strong><\/p>\n

    \u5317\u4eac\u5e08\u8303\u5927\u5b66\u7cfb\u7edf\u79d1\u5b66\u5b66\u9662\u535a\u58eb\u751f\u725f\u7267\u4e91\u5173\u4e8e\u5f3a\u5316\u5b66\u4e60\u4e16\u754c\u6a21\u578b\u7684\u4ecb\u7ecd\uff1a<\/span><\/span><\/p>\n

    \u5e26\u9690\u72b6\u6001\u7684\u5f3a\u5316\u5b66\u4e60\u4e16\u754c\u6a21\u578b<\/span><\/a>
    <\/p>\n

    https:\/\/pattern.swarma.org\/study_group_issue\/361<\/span><\/p>\n


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


    <\/p>\n

    \n
    \n
    \n
    \n

    AGI\u8bfb\u4e66\u4f1a<\/strong>
    <\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


    <\/section>\n

    \u4e3a\u4e86\u6df1\u5165\u63a2\u8ba8 AGI \u76f8\u5173\u8bdd\u9898\uff0c\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u96c6\u8403\u6df1\u5ea6\u611f\u77e5\u6280\u672f\u7814\u7a76\u6240\u6240\u957f\u5cb3\u7389\u6d9b\u3001\u9ebb\u7701\u7406\u5de5\u5b66\u9662\u535a\u58eb\u6c88\u9a6c\u6210\u3001\u5929\u666e\u5927\u5b66\u535a\u58eb\u751f\u5f90\u535a\u6587\uff0c\u5171\u540c\u53d1\u8d77 AGI \u8bfb\u4e66\u4f1a<\/a>\uff0c\u6db5\u76d6\u4e3b\u9898\u5305\u62ec\uff1a\u667a\u80fd\u7684\u5b9a\u4e49\u4e0e\u5ea6\u91cf\u3001\u667a\u80fd\u7684\u539f\u7406\u3001\u5927\u8bed\u8a00\u6a21\u578b\u4e0e\u4fe1\u606f\u4e16\u754c\u7684\u667a\u80fd\u3001\u611f\u77e5\u4e0e\u5177\u8eab\u667a\u80fd\u3001\u591a\u89c6\u89d2\u4e0b\u7684\u4eba\u5de5\u667a\u80fd\u3001\u5bf9\u9f50\u6280\u672f\u4e0eAGI\u5b89\u5168\u6027\u3001AGI\u65f6\u4ee3\u7684\u672a\u6765\u793e\u4f1a\u3002\u6b22\u8fce\u611f\u5174\u8da3\u7684\u670b\u53cb\u62a5\u540d\u53c2\u4e0e\uff01<\/span><\/p>\n


    <\/p>\n

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


    <\/p>\n

    \u8be6\u60c5\u8bf7\u89c1\uff1a<\/span><\/p>\n

    AGI \u8bfb\u4e66\u4f1a\u542f\u52a8\uff1a\u8fc8\u5411\u901a\u7528\u4eba\u5de5\u667a\u80fd\u7684\u8de8\u5b66\u79d1\u8def\u5f84<\/a><\/span>
    <\/span><\/p>\n


    <\/p>\n


    <\/p>\n

    \u63a8\u8350\u9605\u8bfb<\/span><\/strong><\/strong>
    <\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
    \n
    1. <\/strong> \u6885\u62c9\u59ae\u00b7\u7c73\u6b47\u5c14Science\u520a\u6587\uff1aAI\u80fd\u5426\u81ea\u4e3b\u5b66\u4e60\u4e16\u754c\u6a21\u578b\uff1f<\/a><\/strong><\/strong><\/span><\/strong><\/section>\n
    2. <\/strong>\u201c\u610f\u8bc6\u673a\u5668\u201d\u521d\u63a2\uff1a\u5982\u4f55\u8ba9\u5927\u8bed\u8a00\u6a21\u578b\u5177\u5907\u81ea\u6211\u610f\u8bc6\uff1f<\/a><\/strong><\/span><\/strong><\/section>\n
    3. <\/strong><\/strong><\/a>\u5e26\u9690\u72b6\u6001\u7684\u5f3a\u5316\u5b66\u4e60\u4e16\u754c\u6a21\u578b\uff5c\u56e0\u679c\u6d8c\u73b0\u8bfb\u4e66\u4f1a<\/a><\/strong><\/strong><\/span><\/strong><\/section>\n
    <\/strong><\/section>\n
    4. <\/strong><\/strong><\/span>\u5f20\u6c5f\uff1a\u7b2c\u4e09\u4ee3\u4eba\u5de5\u667a\u80fd\u6280\u672f\u57fa\u7840\u2014\u2014\u4ece\u53ef\u5fae\u5206\u7f16\u7a0b\u5230\u56e0\u679c\u63a8\u7406 | \u96c6\u667a\u5b66\u56ed\u5168\u65b0\u8bfe\u7a0b<\/a><\/strong><\/strong><\/span><\/section>\n
    5. <\/strong><\/strong>\u52a0\u5165\u96c6\u667a\u5b66\u56edVIP\uff0c\u4e00\u6b21\u6027\u83b7\u53d6\u96c6\u667a\u5e73\u53f0\u6240\u6709\u5185\u5bb9\u8d44\u6e90<\/a><\/strong><\/strong><\/strong><\/strong><\/span><\/a><\/span><\/strong><\/strong><\/strong><\/section>\n
    6. <\/strong>\u52a0\u5165\u96c6\u667a\uff0c\u4e00\u8d77\u590d\u6742\uff01<\/strong><\/span><\/a><\/section>\n


    <\/p>\n


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

    \u70b9\u51fb\u201c\u9605\u8bfb\u539f\u6587\u201d\uff0c\u62a5\u540d\u8bfb\u4e66\u4f1a<\/span><\/strong><\/section>\n

    <\/mp-style-type><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

    \u5bfc\u8bed \u76ee\u524d\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u867d\u7136\u5f3a\u5927\uff0c\u5374\u5e38\u5e38\u7f3a\u4e4f\u4eba\u7c7b\u548c\u52a8\u7269\u7684\u5e38\u8bc6\u548c\u7075\u6d3b\u63a8\u7406\u80fd\u529b\u3002\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u7684\u6700\u7ec8\u76ee\u6807\uff0c\u662f\u8ba9\u673a\u5668\u80fd\u591f\u50cf\u4eba\u7c7b\u4e00\u6837\u601d\u8003\u3001\u5b66\u4e60\u5e76\u89e3\u51b3\u5404\u79cd\u4efb\u52a1\uff0c\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u3002\u672c\u6587\u4f5c\u8005\u6c88\u9a6c\u6210\u535a\u58eb\u662f\u96c6\u667a\u4ff1\u4e50\u90e8\u300cAGI\u8bfb\u4e66\u4f1a\u300d\u53d1\u8d77\u4eba\u4e4b\u4e00\uff0c\u76ee\u524d\u81f4\u529b\u4e8e\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u7814\u7a76\uff0c\u4ed6\u8ba4\u4e3a\uff0c\u5b9e\u73b0\u901a\u7528\u5177\u8eab\u667a\u80fd\u7684\u5173\u952e\u95ee\u9898\u5728\u4e8e\uff0c\u5982\u4f55\u4f7f…<\/p>\n","protected":false},"author":0,"featured_media":47080,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/47084"}],"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"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=47084"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/47084\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/47080"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=47084"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=47084"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=47084"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=47084"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}