{"id":1069944,"date":"2022-11-16T22:58:00","date_gmt":"2022-11-17T06:58:00","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&p=1069944"},"modified":"2024-09-25T03:14:53","modified_gmt":"2024-09-25T10:14:53","slug":"neurips-2022-industry-ai","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/neurips-2022-industry-ai\/","title":{"rendered":"NeurIPS 2022 | \u8d4b\u80fd\u4ea7\u4e1a\u754c\u7684\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u65b0\u8d8b\u52bf"},"content":{"rendered":"\n

\u7f16\u8005\u6309\uff1a\u4f5c\u4e3a\u76ee\u524d\u5168\u7403\u6700\u8d1f\u76db\u540d\u7684\u4eba\u5de5\u667a\u80fd\u76db\u4f1a\u4e4b\u4e00\uff0cNeurIPS (Conference on Neural Information Processing Systems) \u5728\u6bcf\u5e74\u5e74\u672b\u90fd\u662f\u8ba1\u7b97\u673a\u79d1\u5b66\u9886\u57df\u77a9\u76ee\u7684\u7126\u70b9\u3002\u88ab NeurIPS \u63a5\u6536\u7684\u8bba\u6587\uff0c\u4ee3\u8868\u7740\u5f53\u4eca\u795e\u7ecf\u79d1\u5b66\u548c\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u7684\u6700\u9ad8\u6c34\u5e73\u3002\u4eca\u5e74\u7684 NeurIPS \u5927\u4f1a\u5c06\u4e8e11\u670828\u65e5\u81f312\u67089\u65e5\u4e3e\u884c\uff0c\u672c\u5c4a\u5927\u4f1a\u5171\u6536\u523010411\u7bc7\u6709\u6548\u6295\u7a3f\uff0c\u5176\u4e2d2672\u7bc7\u83b7\u63a5\u6536\uff0c\u6700\u7ec8\u63a5\u6536\u7387\u4e3a25.6%\u3002\u76f8\u6bd4\u53bb\u5e74\uff0c\u6295\u7a3f\u6570\u91cf\u7ee7\u7eed\u589e\u52a0\u3002<\/p>\n\n\n\n

\u5728\u672c\u5c4a\u5927\u4f1a\u4e2d\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e5f\u6709\u8bf8\u591a\u8bba\u6587\u5165\u9009\uff0c\u5185\u5bb9\u4e3b\u8981\u6db5\u76d6\u4eba\u5de5\u667a\u80fd\u4e94\u5927\u70ed\u70b9\u8bdd\u9898\uff1a\u4eba\u5de5\u667a\u80fd\u8d70\u5411\u5927\u4e00\u7edf\u3001\u8ba1\u7b97\u673a\u7406\u8bba\u3001\u8d4b\u80fd\u4ea7\u4e1a\u754c\u7684\u4eba\u5de5\u667a\u80fd\u3001\u8d1f\u8d23\u4efb\u7684\u4eba\u5de5\u667a\u80fd\u3001\u4eba\u5de5\u667a\u80fd\u8d4b\u80fd\u5185\u5bb9\u4e0e\u8bbe\u8ba1\u751f\u6210\u3002\u4e0a\u671f\uff0c\u6211\u4eec\u901f\u89c8\u4e86\u4eba\u5de5\u667a\u80fd\u5927\u4e00\u7edf\u4e0e\u7406\u8bba\u7814\u7a76\u7684\u6700\u65b0\u8fdb\u5c55\u3002\u672c\u671f\u4e3a\u5927\u5bb6\u5e26\u6765\u7684\u662f\u4eba\u5de5\u667a\u80fd\u8d4b\u80fd\u4ea7\u4e1a\u754c\u8bdd\u9898\u4e0b\u76845\u7bc7\u8bba\u6587\u3002<\/p>\n\n\n\n

\u81ea\u52a8\u5f3a\u5316\u5b66\u4e60\u8f85\u52a9\u635f\u5931\u51fd\u6570\u8bbe\u8ba1<\/h3>\n\n\n\n
\"text\"<\/figure>\n\n\n\n

\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/reinforcement-learning-with-automated-auxiliary-loss-search\/<\/p>\n\n\n\n

\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/seqml.github.io\/a2ls\/<\/p>\n\n\n\n

\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u901a\u5e38\u9762\u4e34\u7a00\u758f\u5956\u52b1\u3001\u9ad8\u7ef4\u8f93\u5165\u60c5\u51b5\u4e0b\u4f18\u5316\u7a33\u5b9a\u6027\u8f83\u5dee\u7b49\u8bad\u7ec3\u95ee\u9898\u3002\u4e3a\u6b64\uff0c\u7814\u7a76\u8005\u4eec\u6839\u636e\u7ecf\u9a8c\u63d0\u51fa\u4e86\u591a\u79cd\u8f85\u52a9\u635f\u5931\u51fd\u6570\uff0c\u4ee5\u5e2e\u52a9\u5f3a\u5316\u5b66\u4e60\u8fdb\u884c\u8bad\u7ec3\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u4eba\u5de5\u8bbe\u8ba1\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570\u5341\u5206\u4f9d\u8d56\u4eba\u7c7b\u77e5\u8bc6\uff0c\u4e14\u8bbe\u8ba1\u8fc7\u7a0b\u8017\u65f6\u8d39\u529b\uff1b\u540c\u65f6\uff0c\u7531\u4e8e\u672a\u8003\u8651\u5f3a\u5316\u5b66\u4e60\u7684\u539f\u672c\u4f18\u5316\u76ee\u6807\uff0c\u8fd9\u4e9b\u635f\u5931\u51fd\u6570\u5728\u51fd\u6570\u7a7a\u95f4\u4e2d\u90fd\u5c5e\u4e8e\u6b21\u4f18\u89e3\u3002<\/p>\n\n\n\n

\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u81ea\u52a8\u5f3a\u5316\u5b66\u4e60\u7684\u6846\u67b6 A2LS\uff0c\u5728\u5305\u542b\u73b0\u6709\u4eba\u5de5\u8bbe\u8ba1\u7684\u635f\u5931\u51fd\u6570\u7684\u7a7a\u95f4\uff08\u590d\u6742\u5ea6\u7ea6\u4e3a7.5\u00d710^20\uff09\u4e2d\uff0c\u4f7f\u7528\u57fa\u4e8e\u5143\u5b66\u4e60\u7684\u81ea\u52a8\u5316\u7b97\u6cd5\u81ea\u52a8\u641c\u7d22\u66f4\u4f18\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570\u3002\u7b97\u6cd5\u5728\u4e09\u4e2a\u968f\u673a\u9009\u62e9\u7684\u673a\u5668\u4eba\u8bad\u7ec3\u73af\u5883\u4e2d\u7ecf\u8fc74\u81f37\u8f6e\u7684\u641c\u7d22\u540e\uff0c\u627e\u5230\u4e86\u4e00\u4e2a\u8f83\u4f18\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570 A2-winner\u3002\u5e7f\u6cdb\u7684\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u660e\uff0c\u8be5\u8f85\u52a9\u635f\u5931\u51fd\u6570\u4e0d\u4ec5\u80fd\u5927\u5e45\u63d0\u5347\u5f3a\u5316\u5b66\u4e60\u7684\u8bad\u7ec3\u6548\u679c\uff0c\u800c\u4e14\u663e\u8457\u4f18\u4e8e\u4eba\u5de5\u8bbe\u8ba1\u7684\u8f85\u52a9\u51fd\u6570\uff0c\u5176\u5728\u591a\u79cd\u4e0d\u540c\u7684\u573a\u666f\u4e0b\u5177\u6709\u4f18\u79c0\u7684\u8fc1\u79fb\u6cdb\u5316\u6027\uff0c\u5305\u62ec\u4ece\u672a\u5728\u8bad\u7ec3\u73af\u5883\u4e2d\u89c1\u8fc7\u7684\uff1a\uff081\uff09\u57fa\u4e8e\u56fe\u50cf\u7684\u73af\u5883\uff1b\uff082\uff09\u4e0d\u540c\u7684\u573a\u666f\u4f8b\u5982\u6e38\u620f\u573a\u666f\uff1b\uff083\uff09\u4e0d\u540c\u7684\u72b6\u6001\u7c7b\u578b\uff1b\uff084\uff09\u4e0d\u540c\u7684\u7b56\u7565\u7f51\u7edc\u7ed3\u6784\uff1b\uff085\uff09\u90e8\u5206\u72b6\u6001\u53ef\u89c1\uff08partial observable\uff09\u7684\u573a\u666f\u3002\u8bba\u6587\u539f\u6587\u4e2d\u8fd8\u63d0\u4f9b\u4e86\u66f4\u591a\u89c6\u89d2\u7684\u5bf9\u6bd4\u5206\u6790\uff0c\u8868\u660e\u7b97\u6cd5\u7684\u7279\u70b9\u4e0e\u641c\u7d22\u7ed3\u679c\u7684\u4f18\u8d8a\u6027\u3002<\/p>\n\n\n\n

\"\u56fe1\uff1a\u81ea\u52a8\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u6846\u67b6\uff08\u4e0a\uff09\uff1b\u641c\u7d22\u8fc7\u7a0b\u53ef\u89c6\u5316\u56fe\uff08\u4e0b\uff09\"
\u56fe1\uff1a\u81ea\u52a8\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u6846\u67b6\uff08\u4e0a\uff09\uff1b\u641c\u7d22\u8fc7\u7a0b\u53ef\u89c6\u5316\u56fe\uff08\u4e0b\uff09<\/em><\/em><\/figcaption><\/figure>\n\n\n\n

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

\u81ea\u63d0\u5347\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60<\/h3>\n\n\n\n
\"paper<\/figure>\n\n\n\n

\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/bootstrapped-transformer-for-offline-reinforcement-learning\/<\/p>\n\n\n\n

\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/seqml.github.io\/bootorl\/<\/p>\n\n\n\n

\u968f\u7740\u5f3a\u5316\u5b66\u4e60\u5728\u771f\u5b9e\u4e16\u754c\u573a\u666f\u4e2d\u7684\u9700\u6c42\u9010\u6e10\u589e\u5927\uff0c\u4f5c\u4e3a\u4e00\u79cd\u65b0\u7684\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u8303\u5f0f\uff0c\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\uff08offline reinforcement learning\uff09\u53d7\u5230\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5173\u6ce8\u3002\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u5229\u7528\u9884\u5148\u4fdd\u5b58\u7684\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u7684\u4ea4\u4e92\u6570\u636e\u8fdb\u884c\u79bb\u7ebf\u8bad\u7ec3\uff0c\u8fdb\u800c\u83b7\u5f97\u53ef\u7528\u4e8e\u5728\u7ebf\u771f\u5b9e\u73af\u5883\u7684\u667a\u80fd\u4f53\u3002\u76ee\u524d\u8f83\u53d7\u5173\u6ce8\u7684\u4e00\u4e2a\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u6280\u672f\u5206\u652f\u662f\u5f15\u5165 Transformer \u6a21\u578b\u505a\u5e8f\u5217\u5efa\u6a21\uff0c\u5e76\u53d6\u5f97\u4e86\u826f\u597d\u7684\u6210\u6548\u3002<\/p>\n\n\n\n

\u73b0\u6709\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u6570\u636e\u5b58\u5728\u4e24\u4e2a\u95ee\u9898\uff0c\u7b2c\u4e00\u662f\u8986\u76d6\u7387\u7684\u95ee\u9898\uff0c\u4e3b\u8981\u7531\u4e8e\u91c7\u6837\u83b7\u5f97\u8fd9\u4e9b\u79bb\u7ebf\u6570\u636e\u7684\u667a\u80fd\u4f53\u53ef\u80fd\u662f\u4efb\u610f\u7684\u667a\u80fd\u4f53\uff0c\u96be\u4ee5\u4fdd\u8bc1\u79bb\u7ebf\u6570\u636e\u5bf9\u5f3a\u5316\u5b66\u4e60\u771f\u5b9e\u7684\u6570\u636e\u5206\u5e03\u5177\u5907\u826f\u597d\u7684\u8986\u76d6\uff1b\u7b2c\u4e8c\u662f\u8bad\u7ec3\u6570\u636e\u91cf\u7684\u95ee\u9898\uff0c\u76f8\u6bd4\u8f83\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9884\u8bad\u7ec3\u6a21\u578b\u52a8\u8f84\u4e0a\u767e\u4e07\u6587\u672c\u8bed\u53e5\u7684\u8bed\u6599\u5e93\u76f8\u6bd4\uff0c\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u7684\u8bad\u7ec3\u6570\u636e\u91cf\u4e00\u822c\u8f83\u5c0f\u3002<\/p>\n\n\n\n

\u6839\u636e\u76ee\u524d Transformer \u6a21\u578b\u5e8f\u5217\u5efa\u6a21\u4e0e\u751f\u6210\u7684\u7279\u6027\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u81ea\u63d0\u5347\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u6846\u67b6 Bootstrapped Transformer\uff08\u7b80\u79f0BooT\uff09\u4ee5\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898\u3002BooT \u65b9\u6cd5\u5efa\u6a21\u4e86\u79bb\u7ebf\u8bad\u7ec3\u6570\u636e\u7684\u5206\u5e03\u5e76\u540c\u65f6\u751f\u6210\u4e86\u65b0\u7684\u7b26\u5408\u5206\u5e03\u7684\u6570\u636e\u4ee5\u53cd\u54fa\u8bad\u7ec3\u8fc7\u7a0b\u3002\u5728\u901a\u7528\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u57fa\u51c6\u4e2d\uff0cBooT \u663e\u8457\u63d0\u5347\u4e86\u6548\u679c\uff0c\u751a\u81f3\u8d85\u8d8a\u4e86\u4f7f\u752880\u500d\u6570\u636e\u91cf\u7684\u5176\u4ed6\u9884\u8bad\u7ec3\u65b9\u6848\uff1b\u91cf\u5316\u4e0e\u53ef\u89c6\u5206\u6790\u5747\u8868\u660e BooT \u751f\u6210\u7684\u6570\u636e\u66f4\u52a0\u7b26\u5408\u539f\u59cb\u6570\u636e\u5206\u5e03\uff0c\u5e76\u80fd\u5f25\u8865\u539f\u6709\u79bb\u7ebf\u8bad\u7ec3\u6570\u636e\u7684\u4e0d\u8db3\u3002<\/p>\n\n\n\n

\"\u56fe2\uff1aBooT
\u56fe2\uff1aBooT \u5e8f\u5217\u751f\u6210\u8fc7\u7a0b\uff08\u4e0a\uff09\uff1bBooT \u5e8f\u5217\u751f\u6210\u7ed3\u679c\u4e0e\u5206\u5e03\u964d\u7ef4\u53ef\u89c6\u5316\uff08\u4e0b\uff09<\/figcaption><\/figure>\n\n\n\n

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

\u9762\u5411\u5177\u6709\u5206\u6bb5\u7a33\u5b9a\u4e0a\u4e0b\u6587\u7684\u975e\u5e73\u7a33\u73af\u5883\u7684\u81ea\u9002\u5e94\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60<\/h3>\n\n\n\n
\"paper<\/figure>\n\n\n\n

\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/an-adaptive-deep-rl-method-for-non-stationary-environments-with-piecewise-stable-context\/<\/p>\n\n\n\n

\u5728\u73b0\u5b9e\u751f\u6d3b\u4e2d\uff0c\u667a\u80fd\u4f53\u5904\u5728\u4e0d\u65ad\u53d8\u5316\u7684\u73af\u5883\u4e2d\u3002\u8fd9\u662f\u963b\u788d\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u6210\u529f\u5e94\u7528\u7684\u539f\u56e0\u4e4b\u4e00\u3002\u5728\u5f88\u591a\u73b0\u5b9e\u95ee\u9898\u4e2d\uff0c\u73af\u5883\uff08\u4f8b\u5982\u7f51\u7edc\u53ef\u7528\u5e26\u5bbd\u3001\u673a\u5668\u4eba\u6240\u5904\u5730\u5f62\u7b49\uff09\u4f1a\u5728\u4e00\u6bb5\u968f\u673a\u7684\u65f6\u95f4\u5185\u4fdd\u6301\u4e0d\u53d8\uff0c\u968f\u540e\u4ee5\u4e0d\u53ef\u9884\u6d4b\u7684\u65b9\u5f0f\u53d1\u751f\u7a81\u7136\u8df3\u53d8\u3002\u8fd9\u4e9b\u73af\u5883\u53d8\u91cf\u5f80\u5f80\u4e0d\u53ef\u89c2\u6d4b\u3002\u56e0\u6b64\uff0c\u667a\u80fd\u4f53\u9700\u8981\u53ca\u65f6\u68c0\u6d4b\u5230\u8fd9\u79cd\u53d8\u5316\uff0c\u5e76\u5feb\u901f\u8c03\u6574\u81ea\u8eab\u7b56\u7565\u4ee5\u9002\u5e94\u8fd9\u79cd\u53d8\u5316\u3002\u7136\u800c\u73b0\u6709\u5de5\u4f5c\u7684\u5efa\u6a21\u65b9\u5f0f\u90fd\u4e0d\u80fd\u5f88\u597d\u5730\u89e3\u51b3\u5f53\u524d\u95ee\u9898\u3002<\/p>\n\n\n\n

\u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684 Latent Situational MDP (LS-MDP)\uff0c\u5f15\u5165\u4e86\u73af\u5883\u53d8\u91cf C \u4ee5\u53ca\u73af\u5883\u53d8\u91cf\u7ed3\u6784 G \u6765\u7ec6\u81f4\u5730\u523b\u753b\u95ee\u9898\u7684\u7ed3\u6784\uff0c\u5e76\u63d0\u51fa\u4e86 SeCBAD\uff08segmented context belief augmented deep RL\uff09\u7b97\u6cd5\uff0c\u4ee5\u8054\u5408\u63a8\u65ad\u73af\u5883\u53d8\u91cf\u7ed3\u6784\u53ca\u5185\u5bb9\u7684\u65b9\u5f0f\uff0c\u6765\u5b8c\u6210\u5bf9\u73af\u5883\u53d8\u5316\u7684\u5feb\u901f\u68c0\u6d4b\u3002\u540c\u65f6\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u4f7f\u7528\u63a8\u65ad\u7684\u73af\u5883\u53d8\u91cf belief \u5bf9 state \u8fdb\u884c\u589e\u5e7f\uff0c\u4ece\u800c\u4f7f\u5f97\u667a\u80fd\u4f53\u53ef\u4ee5\u5728\u6536\u96c6\u66f4\u591a\u4fe1\u606f\u3001\u4e0e\u5229\u7528\u5df2\u6709\u4fe1\u606f\u6700\u5927\u5316\u56de\u62a5\u8fd9\u4e24\u79cd\u7b56\u7565\u95f4\u8fbe\u5230\u6700\u4f18\u6743\u8861\u3002\u5728\u673a\u5668\u4eba\u63a7\u5236\u3001\u7f51\u7edc\u5e26\u5bbd\u63a7\u5236\u7b49\u5e94\u7528\u4e0a\u5bf9\u8be5\u7b97\u6cd5\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e SeCBAD \u80fd\u591f\u663e\u8457\u63d0\u5347\u6027\u80fd\u3002<\/p>\n\n\n\n

\"\u56fe3\uff1aSeCBAD
\u56fe3\uff1aSeCBAD \u7b97\u6cd5\u6846\u67b6\u56fe<\/em><\/figcaption><\/figure>\n\n\n\n

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

\u57fa\u4e8e\u591a\u6a21\u6001\u65f6\u5e8f\u5bf9\u6bd4\u5b66\u4e60\u7684\u957f\u89c6\u9891-\u8bed\u8a00\u9884\u8bad\u7ec3<\/h3>\n\n\n\n
\"paper<\/figure>\n\n\n\n

\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/long-form-video-language-pre-training-with-multimodal-temporal-contrastive-learning\/<\/p>\n\n\n\n

\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/github.com\/microsoft\/XPretrain<\/p>\n\n\n\n

\u4ee5\u5f80\u5bf9\u89c6\u9891-\u8bed\u8a00\u9884\u8bad\u7ec3\u7684\u7814\u7a76\u4e3b\u8981\u96c6\u4e2d\u5728\u77ed\u89c6\u9891\uff08\u537330\u79d2\u5185\u7684\u89c6\u9891\uff09\u548c\u53e5\u5b50\u4e0a\uff0c\u5ffd\u7565\u4e86\u771f\u5b9e\u573a\u666f\u4e2d\u66f4\u52a0\u5e38\u89c1\u7684\u957f\u89c6\u9891\u7406\u89e3\u3002\u76f4\u63a5\u4ece\u957f\u89c6\u9891\u548c\u8bed\u8a00\u4e2d\u5b66\u4e60\u8868\u5f81\u53ef\u4ee5\u5e2e\u52a9\u8bb8\u591a\u957f\u89c6\u9891-\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\u7684\u53d1\u5c55\uff0c\u4f46\u8fdc\u7a0b\u5173\u7cfb\u5efa\u6a21\u7684\u96be\u5ea6\u548c\u66f4\u591a\u5e27\u5f15\u8d77\u7684\u8ba1\u7b97\u8d1f\u62c5\u6781\u5177\u6311\u6218\u6027\u3002<\/p>\n\n\n\n

\u5728\u672c\u6587\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 Long-Form VIdeo-LAnguage \u9884\u8bad\u7ec3\u6a21\u578b\uff08LF-VILA\uff09\uff0c\u5e76\u5728\u57fa\u4e8e\u89c6\u9891\u6587\u672c\u6570\u636e\u96c6 HD-VILA-100M \u6784\u5efa\u7684\u5927\u89c4\u6a21\u957f\u89c6\u9891\u548c\u6bb5\u843d\u6570\u636e\u96c6 LF-VILA-8M \u4e0a\u8fdb\u884c\u4e86\u8bad\u7ec3\u3002\u4e3a\u4e86\u6709\u6548\u5730\u6355\u6349\u4e30\u5bcc\u7684\u65f6\u5e8f\u52a8\u6001\uff0c\u5e76\u4ee5\u6709\u6548\u7684\u7aef\u5230\u7aef\u7684\u65b9\u5f0f\u66f4\u597d\u5730\u5bf9\u9f50\u89c6\u9891\u548c\u8bed\u8a00\uff0c LF-VILA \u6a21\u578b\u4e2d\u5f15\u5165\u4e86\u4e24\u79cd\u65b0\u8bbe\u8ba1\u3002\u7814\u7a76\u5458\u4eec\u9996\u5148\u63d0\u51fa\u4e86\u4e00\u79cd\u591a\u6a21\u6001\u65f6\u5e8f\u5bf9\u6bd4\u5b66\u4e60\uff08MTC\uff09\u635f\u5931\uff0c\u901a\u8fc7\u9f13\u52b1\u957f\u89c6\u9891\u548c\u6bb5\u843d\u4e4b\u95f4\u7684\u7ec6\u7c92\u5ea6\u5bf9\u9f50\u6765\u5b66\u4e60\u4e0d\u540c\u6a21\u6001\u4e4b\u95f4\u7684\u65f6\u5e8f\u5173\u7cfb\u3002\u5176\u6b21\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u5206\u5c42\u65f6\u95f4\u7a97\u53e3\u6ce8\u610f\u529b\uff08HTWA\uff09\u673a\u5236\uff0c\u4ee5\u6709\u6548\u6355\u83b7\u957f\u671f\u4f9d\u8d56\u5173\u7cfb\uff0c\u540c\u65f6\u964d\u4f4e Transformer \u7684\u8ba1\u7b97\u5f00\u9500\u3002<\/p>\n\n\n\n

\u57287\u4e2a\u4e0b\u6e38\u957f\u89c6\u9891\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\uff08\u5305\u62ec\u6bb5\u843d\u5230\u89c6\u9891\u68c0\u7d22\u548c\u957f\u89c6\u9891\u95ee\u7b54\uff09\u4e0a\u5bf9\u9884\u8bad\u7ec3\u7684 LF-VILA \u6a21\u578b\u8fdb\u7684\u9a8c\u8bc1\u8868\u660e\uff0cLF-VILA \u53d6\u5f97\u4e86\u6700\u597d\u7684\u6027\u80fd\u3002\u5177\u4f53\u800c\u8a00\uff0cLF-VILA \u5728 ActivityNet \u6bb5\u843d\u5230\u89c6\u9891\u68c0\u7d22\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e8616.1%\u7684\u76f8\u5bf9\u6539\u8fdb\uff0c\u5728 How2QA \u4efb\u52a1\u4e0a\u53d6\u5f97\u4e862.4%\u7684\u76f8\u5bf9\u6539\u8fdb\u3002LF-VILA \u7684\u4ee3\u7801\u3001\u6570\u636e\u96c6\u548c\u9884\u8bad\u7ec3\u7684\u6a21\u578b\u5c06\u5f88\u5feb\u5728\u5b98\u65b9\u9879\u76ee\u4e3b\u9875\u4e0a\u53d1\u5e03\uff0c\u6b22\u8fce\u5173\u6ce8\u3002<\/p>\n\n\n\n

\"\u56fe4\uff1aLF-VILA
\u56fe4\uff1aLF-VILA \u9884\u8bad\u7ec3\u6a21\u578b\u6846\u67b6\u56fe(a)\u4e0e MTC \u65b9\u6cd5\u793a\u610f\u56fe(b)<\/em><\/figcaption><\/figure>\n\n\n\n

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

\u9762\u5411\u591a\u4efb\u52a1\u5206\u7ec4\u5b66\u4e60\u7684\u5143\u5b66\u4e60\u6846\u67b6<\/h3>\n\n\n\n
\"paper<\/figure>\n\n\n\n

\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/openreview.net\/forum?id=Rqe-fJQtExY<\/p>\n\n\n\n

\u591a\u4efb\u52a1\u5b66\u4e60\u8303\u5f0f\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u5404\u79cd\u5404\u6837\u7684\u673a\u5668\u5b66\u4e60\u573a\u666f\u4e2d\uff0c\u5305\u62ec\u8bed\u8a00\u7406\u89e3\u3001\u89c6\u89c9\u8bc6\u522b\u3001\u673a\u5668\u4eba\u63a7\u5236\u3001\u836f\u7269\u53d1\u73b0\u3001\u4e34\u5e8a\u6cbb\u7597\u3001\u80fd\u6e90\u7ba1\u7406\u7b49\u7b49\u3002\u7136\u800c\uff0c\u5982\u4f55\u6709\u6548\u5730\u9009\u62e9\u5171\u540c\u8bad\u7ec3\u7684\u4efb\u52a1\u7ec4\u5408\u5374\u662f\u4e00\u4e2a\u6781\u6709\u6311\u6218\u6027\u7684\u95ee\u9898\uff0c\u8fd9\u91cc\u6709\u4e24\u65b9\u9762\u7684\u539f\u56e0\uff1a1\uff09\u4efb\u52a1\u7ec4\u5408\u7684\u6570\u91cf\u968f\u7740\u4efb\u52a1\u6570\u91cf\u7684\u589e\u52a0\u800c\u6210\u6307\u6570\u7ea7\u522b\u5730\u589e\u957f\uff1b2\uff09\u7b80\u5355\u5730\u9009\u62e9\u4e00\u79cd\u5206\u7ec4\u53ef\u80fd\u4f1a\u5bfc\u81f4\u67d0\u4e9b\u4efb\u52a1\u51fa\u73b0\u4e25\u91cd\u7684\u6027\u80fd\u8870\u9000\u3002\u73b0\u6709\u7684\u7814\u7a76\u5de5\u4f5c\u8981\u4e48\u805a\u7126\u4e8e\u7ed9\u5b9a\u4efb\u52a1\u7ec4\u5408\u60c5\u51b5\u4e0b\u7684\u4f18\u5316\u65b9\u6cd5\u548c\u6a21\u578b\u67b6\u6784\uff0c\u8981\u4e48\u91c7\u7528\u4e00\u4e9b\u542f\u53d1\u5f0f\u7684\u65b9\u6cd5\u6765\u5e94\u5bf9\u6307\u6570\u589e\u957f\u7684\u4efb\u52a1\u7ec4\u5408\uff08\u6bd4\u5982\uff1a\u5c06\u591a\u4e2a\u4efb\u52a1\u7ec4\u5408\u5728\u4e00\u8d77\u7684\u6536\u76ca\uff0c\u8fd1\u4f3c\u5747\u644a\u4e3a\u5176\u4e24\u4e24\u7ec4\u5408\u7684\u6536\u76ca\u4e4b\u548c\uff09\u3002\u8bf8\u591a\u65b0\u8fdb\u591a\u4efb\u52a1\u5b66\u4e60\u7814\u7a76\u90fd\u8ba4\u4e3a\u5728\u4efb\u52a1\u6570\u91cf\u8f83\u591a\u65f6\uff0c\u5145\u5206\u63a2\u7d22\u8fd9\u4e2a\u6307\u6570\u7ea7\u522b\u589e\u957f\u7684\u4efb\u52a1\u7ec4\u5408\u7684\u7a7a\u95f4\u662f\u51e0\u4e4e\u4e0d\u53ef\u80fd\u5b8c\u6210\u7684\u4efb\u52a1\u3002<\/p>\n\n\n\n

\u4e3a\u4e86\u5e94\u5bf9\u4efb\u52a1\u7ec4\u5408\u7206\u70b8\u589e\u957f\u7684\u6311\u6218\uff0c\u5728\u672c\u6587\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u4e3a\u591a\u4efb\u52a1\u5206\u7ec4\u5b66\u4e60\u63d0\u51fa\u4e86\u4e00\u5957\u5143\u5b66\u4e60\u6846\u67b6\u3002\u672c\u6587\u7684\u6838\u5fc3\u89c2\u6d4b\u5728\u4e8e\u5c3d\u7ba1\u4efb\u52a1\u7ec4\u5408\u7684\u6570\u91cf\u968f\u7740\u4efb\u52a1\u7684\u6570\u91cf\u6307\u6570\u589e\u957f\uff0c\u4f46\u4efb\u52a1\u7ec4\u5408\u4e0e\u5728\u5176\u4e0a\u505a\u591a\u4efb\u52a1\u5b66\u4e60\u5e26\u6765\u7684\u589e\u76ca\u4e4b\u95f4\u7684\u5173\u7cfb\u5374\u5b58\u5728\u4e8e\u4e00\u4e2a\u4f4e\u7ef4\u7684\u6d41\u5f62\u7a7a\u95f4\u4e2d\u3002\u57fa\u4e8e\u6b64\u89c2\u6d4b\uff0c\u672c\u6587\u5b9a\u4e49\u4e86\u4e00\u4e2a\u4efb\u52a1\u7ec4\u5408\u4e0a\u7684\u5143\u5b66\u4e60\u95ee\u9898\uff1a\u7ed9\u5b9a\u4e00\u4e2a\u4efb\u52a1\u7ec4\u5408\uff0c\u9884\u6d4b\u5176\u591a\u4efb\u52a1\u5b66\u4e60\u76f8\u5bf9\u4e8e\u5355\u4efb\u52a1\u5b66\u4e60\u6240\u5e26\u6765\u7684\u589e\u76ca\u3002\u5728\u6b64\u5143\u5b66\u4e60\u95ee\u9898\u7684\u57fa\u7840\u4e0a\uff0c\u7814\u7a76\u5458\u4eec\u6784\u5efa\u4e86\u4e00\u4e2a\u5143\u6a21\u578b\u5e76\u5f00\u53d1\u4e86\u4e00\u5957\u9010\u6b65\u9009\u62e9\u5143\u5b66\u4e60\u6837\u672c\u7684\u8bad\u7ec3\u65b9\u6cd5\u3002\u57fa\u4e8e\u6b64\u65b9\u6cd5\uff0c\u53ea\u9700\u8981\u8fdb\u884c\u6709\u9650\u6b21\u7684\u591a\u4efb\u52a1\u5b66\u4e60\u4e0e\u8bc4\u4f30\uff0c\u5373\u53ef\u83b7\u5f97\u4e00\u4e2a\u6709\u6548\u7684\u5143\u6a21\u578b\u6765\u51c6\u786e\u5730\u9884\u6d4b\u51fa\u6240\u6709\u672a\u77e5\u4efb\u52a1\u7ec4\u5408\u4e0a\u591a\u4efb\u52a1\u5b66\u4e60\u7684\u589e\u76ca\uff0c\u8fdb\u800c\u5b9e\u73b0\u6709\u6548\u7684\u4efb\u52a1\u5206\u7ec4\u3002<\/p>\n\n\n\n

\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5728\u89c6\u89c9\u3001\u80fd\u6e90\u3001\u533b\u7597\u7b49\u591a\u79cd\u591a\u4efb\u52a1\u5b66\u4e60\u573a\u666f\u4e0b\u9a8c\u8bc1\u4e86\u4ee5\u4e0a\u6846\u67b6\u540e\uff0c\u4e00\u4e2a\u5927\u89c4\u6a21\u7684\u5b9e\u9a8c\u8bc4\u6d4b\uff0827\u4e2a\u4e34\u5e8a\u533b\u7597\u9884\u6d4b\u4efb\u52a1\uff0c\u8d85\u8fc710\u4ebf\u7684\u4efb\u52a1\u7ec4\u5408\uff09\u8868\u660e\uff1a\u5728\u51e0\u4e4e\u540c\u7b49\u7684\u8ba1\u7b97\u5f00\u9500\u4e0b\uff0c\u76f8\u5bf9\u4e8e\u73b0\u6709\u7684\u6700\u4f73\u65b9\u6848\uff0c\u672c\u6587\u6240\u63d0\u51fa\u7684\u5143\u5b66\u4e60\u65b9\u6cd5\u83b7\u5f97\u4e86\u51e0\u4e4e\u7ffb\u500d\u7684\u6027\u80fd\u63d0\u5347\u3002<\/p>\n\n\n\n

\"\u56fe5\uff1a\u5143\u5b66\u4e60\u6846\u67b6\u56fe\"
\u56fe5\uff1a\u5143\u5b66\u4e60\u6846\u67b6\u56fe<\/em><\/figcaption><\/figure>\n\n\n\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

\u7f16\u8005\u6309\uff1a\u4f5c\u4e3a\u76ee\u524d\u5168\u7403\u6700\u8d1f\u76db\u540d\u7684\u4eba\u5de5\u667a\u80fd\u76db\u4f1a\u4e4b\u4e00\uff0cNeurIPS (Conference on Neural Information Processing Systems) \u5728\u6bcf\u5e74\u5e74\u672b\u90fd\u662f\u8ba1\u7b97\u673a\u79d1\u5b66\u9886\u57df\u77a9\u76ee\u7684\u7126\u70b9\u3002\u88ab NeurIPS \u63a5\u6536\u7684\u8bba\u6587\uff0c\u4ee3\u8868\u7740\u5f53\u4eca\u795e\u7ecf\u79d1\u5b66\u548c\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u7684\u6700\u9ad8\u6c34\u5e73\u3002\u4eca\u5e74\u7684 NeurIPS \u5927\u4f1a\u5c06\u4e8e11\u670828\u65e5\u81f312\u67089\u65e5\u4e3e\u884c\uff0c\u672c\u5c4a\u5927\u4f1a\u5171\u6536\u523010411\u7bc7\u6709\u6548\u6295\u7a3f\uff0c\u5176\u4e2d2672\u7bc7\u83b7\u63a5\u6536\uff0c\u6700\u7ec8\u63a5\u6536\u7387\u4e3a25.6%\u3002\u76f8\u6bd4\u53bb\u5e74\uff0c\u6295\u7a3f\u6570\u91cf\u7ee7\u7eed\u589e\u52a0\u3002 \u5728\u672c\u5c4a\u5927\u4f1a\u4e2d\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e5f\u6709\u8bf8\u591a\u8bba\u6587\u5165\u9009\uff0c\u5185\u5bb9\u4e3b\u8981\u6db5\u76d6\u4eba\u5de5\u667a\u80fd\u4e94\u5927\u70ed\u70b9\u8bdd\u9898\uff1a\u4eba\u5de5\u667a\u80fd\u8d70\u5411\u5927\u4e00\u7edf\u3001\u8ba1\u7b97\u673a\u7406\u8bba\u3001\u8d4b\u80fd\u4ea7\u4e1a\u754c\u7684\u4eba\u5de5\u667a\u80fd\u3001\u8d1f\u8d23\u4efb\u7684\u4eba\u5de5\u667a\u80fd\u3001\u4eba\u5de5\u667a\u80fd\u8d4b\u80fd\u5185\u5bb9\u4e0e\u8bbe\u8ba1\u751f\u6210\u3002\u4e0a\u671f\uff0c\u6211\u4eec\u901f\u89c8\u4e86\u4eba\u5de5\u667a\u80fd\u5927\u4e00\u7edf\u4e0e\u7406\u8bba\u7814\u7a76\u7684\u6700\u65b0\u8fdb\u5c55\u3002\u672c\u671f\u4e3a\u5927\u5bb6\u5e26\u6765\u7684\u662f\u4eba\u5de5\u667a\u80fd\u8d4b\u80fd\u4ea7\u4e1a\u754c\u8bdd\u9898\u4e0b\u76845\u7bc7\u8bba\u6587\u3002 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/reinforcement-learning-with-automated-auxiliary-loss-search\/ \u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/seqml.github.io\/a2ls\/ \u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u901a\u5e38\u9762\u4e34\u7a00\u758f\u5956\u52b1\u3001\u9ad8\u7ef4\u8f93\u5165\u60c5\u51b5\u4e0b\u4f18\u5316\u7a33\u5b9a\u6027\u8f83\u5dee\u7b49\u8bad\u7ec3\u95ee\u9898\u3002\u4e3a\u6b64\uff0c\u7814\u7a76\u8005\u4eec\u6839\u636e\u7ecf\u9a8c\u63d0\u51fa\u4e86\u591a\u79cd\u8f85\u52a9\u635f\u5931\u51fd\u6570\uff0c\u4ee5\u5e2e\u52a9\u5f3a\u5316\u5b66\u4e60\u8fdb\u884c\u8bad\u7ec3\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u4eba\u5de5\u8bbe\u8ba1\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570\u5341\u5206\u4f9d\u8d56\u4eba\u7c7b\u77e5\u8bc6\uff0c\u4e14\u8bbe\u8ba1\u8fc7\u7a0b\u8017\u65f6\u8d39\u529b\uff1b\u540c\u65f6\uff0c\u7531\u4e8e\u672a\u8003\u8651\u5f3a\u5316\u5b66\u4e60\u7684\u539f\u672c\u4f18\u5316\u76ee\u6807\uff0c\u8fd9\u4e9b\u635f\u5931\u51fd\u6570\u5728\u51fd\u6570\u7a7a\u95f4\u4e2d\u90fd\u5c5e\u4e8e\u6b21\u4f18\u89e3\u3002 \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u81ea\u52a8\u5f3a\u5316\u5b66\u4e60\u7684\u6846\u67b6 A2LS\uff0c\u5728\u5305\u542b\u73b0\u6709\u4eba\u5de5\u8bbe\u8ba1\u7684\u635f\u5931\u51fd\u6570\u7684\u7a7a\u95f4\uff08\u590d\u6742\u5ea6\u7ea6\u4e3a7.5\u00d710^20\uff09\u4e2d\uff0c\u4f7f\u7528\u57fa\u4e8e\u5143\u5b66\u4e60\u7684\u81ea\u52a8\u5316\u7b97\u6cd5\u81ea\u52a8\u641c\u7d22\u66f4\u4f18\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570\u3002\u7b97\u6cd5\u5728\u4e09\u4e2a\u968f\u673a\u9009\u62e9\u7684\u673a\u5668\u4eba\u8bad\u7ec3\u73af\u5883\u4e2d\u7ecf\u8fc74\u81f37\u8f6e\u7684\u641c\u7d22\u540e\uff0c\u627e\u5230\u4e86\u4e00\u4e2a\u8f83\u4f18\u7684\u8f85\u52a9\u635f\u5931\u51fd\u6570 A2-winner\u3002\u5e7f\u6cdb\u7684\u5b9e\u9a8c\u7ed3\u679c\u8bc1\u660e\uff0c\u8be5\u8f85\u52a9\u635f\u5931\u51fd\u6570\u4e0d\u4ec5\u80fd\u5927\u5e45\u63d0\u5347\u5f3a\u5316\u5b66\u4e60\u7684\u8bad\u7ec3\u6548\u679c\uff0c\u800c\u4e14\u663e\u8457\u4f18\u4e8e\u4eba\u5de5\u8bbe\u8ba1\u7684\u8f85\u52a9\u51fd\u6570\uff0c\u5176\u5728\u591a\u79cd\u4e0d\u540c\u7684\u573a\u666f\u4e0b\u5177\u6709\u4f18\u79c0\u7684\u8fc1\u79fb\u6cdb\u5316\u6027\uff0c\u5305\u62ec\u4ece\u672a\u5728\u8bad\u7ec3\u73af\u5883\u4e2d\u89c1\u8fc7\u7684\uff1a\uff081\uff09\u57fa\u4e8e\u56fe\u50cf\u7684\u73af\u5883\uff1b\uff082\uff09\u4e0d\u540c\u7684\u573a\u666f\u4f8b\u5982\u6e38\u620f\u573a\u666f\uff1b\uff083\uff09\u4e0d\u540c\u7684\u72b6\u6001\u7c7b\u578b\uff1b\uff084\uff09\u4e0d\u540c\u7684\u7b56\u7565\u7f51\u7edc\u7ed3\u6784\uff1b\uff085\uff09\u90e8\u5206\u72b6\u6001\u53ef\u89c1\uff08partial observable\uff09\u7684\u573a\u666f\u3002\u8bba\u6587\u539f\u6587\u4e2d\u8fd8\u63d0\u4f9b\u4e86\u66f4\u591a\u89c6\u89d2\u7684\u5bf9\u6bd4\u5206\u6790\uff0c\u8868\u660e\u7b97\u6cd5\u7684\u7279\u70b9\u4e0e\u641c\u7d22\u7ed3\u679c\u7684\u4f18\u8d8a\u6027\u3002 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/bootstrapped-transformer-for-offline-reinforcement-learning\/ \u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/seqml.github.io\/bootorl\/ \u968f\u7740\u5f3a\u5316\u5b66\u4e60\u5728\u771f\u5b9e\u4e16\u754c\u573a\u666f\u4e2d\u7684\u9700\u6c42\u9010\u6e10\u589e\u5927\uff0c\u4f5c\u4e3a\u4e00\u79cd\u65b0\u7684\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u8303\u5f0f\uff0c\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\uff08offline reinforcement learning\uff09\u53d7\u5230\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5173\u6ce8\u3002\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u5229\u7528\u9884\u5148\u4fdd\u5b58\u7684\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u7684\u4ea4\u4e92\u6570\u636e\u8fdb\u884c\u79bb\u7ebf\u8bad\u7ec3\uff0c\u8fdb\u800c\u83b7\u5f97\u53ef\u7528\u4e8e\u5728\u7ebf\u771f\u5b9e\u73af\u5883\u7684\u667a\u80fd\u4f53\u3002\u76ee\u524d\u8f83\u53d7\u5173\u6ce8\u7684\u4e00\u4e2a\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u6280\u672f\u5206\u652f\u662f\u5f15\u5165 Transformer \u6a21\u578b\u505a\u5e8f\u5217\u5efa\u6a21\uff0c\u5e76\u53d6\u5f97\u4e86\u826f\u597d\u7684\u6210\u6548\u3002 \u73b0\u6709\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u6570\u636e\u5b58\u5728\u4e24\u4e2a\u95ee\u9898\uff0c\u7b2c\u4e00\u662f\u8986\u76d6\u7387\u7684\u95ee\u9898\uff0c\u4e3b\u8981\u7531\u4e8e\u91c7\u6837\u83b7\u5f97\u8fd9\u4e9b\u79bb\u7ebf\u6570\u636e\u7684\u667a\u80fd\u4f53\u53ef\u80fd\u662f\u4efb\u610f\u7684\u667a\u80fd\u4f53\uff0c\u96be\u4ee5\u4fdd\u8bc1\u79bb\u7ebf\u6570\u636e\u5bf9\u5f3a\u5316\u5b66\u4e60\u771f\u5b9e\u7684\u6570\u636e\u5206\u5e03\u5177\u5907\u826f\u597d\u7684\u8986\u76d6\uff1b\u7b2c\u4e8c\u662f\u8bad\u7ec3\u6570\u636e\u91cf\u7684\u95ee\u9898\uff0c\u76f8\u6bd4\u8f83\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9884\u8bad\u7ec3\u6a21\u578b\u52a8\u8f84\u4e0a\u767e\u4e07\u6587\u672c\u8bed\u53e5\u7684\u8bed\u6599\u5e93\u76f8\u6bd4\uff0c\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u7684\u8bad\u7ec3\u6570\u636e\u91cf\u4e00\u822c\u8f83\u5c0f\u3002 \u6839\u636e\u76ee\u524d Transformer \u6a21\u578b\u5e8f\u5217\u5efa\u6a21\u4e0e\u751f\u6210\u7684\u7279\u6027\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u81ea\u63d0\u5347\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u6846\u67b6 Bootstrapped Transformer\uff08\u7b80\u79f0BooT\uff09\u4ee5\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898\u3002BooT \u65b9\u6cd5\u5efa\u6a21\u4e86\u79bb\u7ebf\u8bad\u7ec3\u6570\u636e\u7684\u5206\u5e03\u5e76\u540c\u65f6\u751f\u6210\u4e86\u65b0\u7684\u7b26\u5408\u5206\u5e03\u7684\u6570\u636e\u4ee5\u53cd\u54fa\u8bad\u7ec3\u8fc7\u7a0b\u3002\u5728\u901a\u7528\u7684\u79bb\u7ebf\u5f3a\u5316\u5b66\u4e60\u57fa\u51c6\u4e2d\uff0cBooT \u663e\u8457\u63d0\u5347\u4e86\u6548\u679c\uff0c\u751a\u81f3\u8d85\u8d8a\u4e86\u4f7f\u752880\u500d\u6570\u636e\u91cf\u7684\u5176\u4ed6\u9884\u8bad\u7ec3\u65b9\u6848\uff1b\u91cf\u5316\u4e0e\u53ef\u89c6\u5206\u6790\u5747\u8868\u660e BooT \u751f\u6210\u7684\u6570\u636e\u66f4\u52a0\u7b26\u5408\u539f\u59cb\u6570\u636e\u5206\u5e03\uff0c\u5e76\u80fd\u5f25\u8865\u539f\u6709\u79bb\u7ebf\u8bad\u7ec3\u6570\u636e\u7684\u4e0d\u8db3\u3002 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/an-adaptive-deep-rl-method-for-non-stationary-environments-with-piecewise-stable-context\/ \u5728\u73b0\u5b9e\u751f\u6d3b\u4e2d\uff0c\u667a\u80fd\u4f53\u5904\u5728\u4e0d\u65ad\u53d8\u5316\u7684\u73af\u5883\u4e2d\u3002\u8fd9\u662f\u963b\u788d\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u6210\u529f\u5e94\u7528\u7684\u539f\u56e0\u4e4b\u4e00\u3002\u5728\u5f88\u591a\u73b0\u5b9e\u95ee\u9898\u4e2d\uff0c\u73af\u5883\uff08\u4f8b\u5982\u7f51\u7edc\u53ef\u7528\u5e26\u5bbd\u3001\u673a\u5668\u4eba\u6240\u5904\u5730\u5f62\u7b49\uff09\u4f1a\u5728\u4e00\u6bb5\u968f\u673a\u7684\u65f6\u95f4\u5185\u4fdd\u6301\u4e0d\u53d8\uff0c\u968f\u540e\u4ee5\u4e0d\u53ef\u9884\u6d4b\u7684\u65b9\u5f0f\u53d1\u751f\u7a81\u7136\u8df3\u53d8\u3002\u8fd9\u4e9b\u73af\u5883\u53d8\u91cf\u5f80\u5f80\u4e0d\u53ef\u89c2\u6d4b\u3002\u56e0\u6b64\uff0c\u667a\u80fd\u4f53\u9700\u8981\u53ca\u65f6\u68c0\u6d4b\u5230\u8fd9\u79cd\u53d8\u5316\uff0c\u5e76\u5feb\u901f\u8c03\u6574\u81ea\u8eab\u7b56\u7565\u4ee5\u9002\u5e94\u8fd9\u79cd\u53d8\u5316\u3002\u7136\u800c\u73b0\u6709\u5de5\u4f5c\u7684\u5efa\u6a21\u65b9\u5f0f\u90fd\u4e0d\u80fd\u5f88\u597d\u5730\u89e3\u51b3\u5f53\u524d\u95ee\u9898\u3002 \u4e3a\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684 Latent Situational MDP (LS-MDP)\uff0c\u5f15\u5165\u4e86\u73af\u5883\u53d8\u91cf C \u4ee5\u53ca\u73af\u5883\u53d8\u91cf\u7ed3\u6784 G \u6765\u7ec6\u81f4\u5730\u523b\u753b\u95ee\u9898\u7684\u7ed3\u6784\uff0c\u5e76\u63d0\u51fa\u4e86 SeCBAD\uff08segmented context belief augmented deep RL\uff09\u7b97\u6cd5\uff0c\u4ee5\u8054\u5408\u63a8\u65ad\u73af\u5883\u53d8\u91cf\u7ed3\u6784\u53ca\u5185\u5bb9\u7684\u65b9\u5f0f\uff0c\u6765\u5b8c\u6210\u5bf9\u73af\u5883\u53d8\u5316\u7684\u5feb\u901f\u68c0\u6d4b\u3002\u540c\u65f6\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u4f7f\u7528\u63a8\u65ad\u7684\u73af\u5883\u53d8\u91cf belief […]<\/p>\n","protected":false},"author":34512,"featured_media":1070310,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1069944","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069944","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":4,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069944\/revisions"}],"predecessor-version":[{"id":1086633,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1069944\/revisions\/1086633"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1070310"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1069944"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1069944"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1069944"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1069944"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}