

{"id":38484,"date":"2019-12-05T00:00:54","date_gmt":"2019-12-04T16:00:54","guid":{"rendered":"https:\/\/case.ntu.edu.tw\/blog\/?p=38484"},"modified":"2022-01-03T23:15:43","modified_gmt":"2022-01-03T15:15:43","slug":"ai%e4%b9%9f%e7%8e%a9%e6%8d%89%e8%bf%b7%e8%97%8f%ef%bc%81%ef%bc%9f","status":"publish","type":"post","link":"https:\/\/case.ntu.edu.tw\/blog\/?p=38484","title":{"rendered":"AI\u4e5f\u73a9\u6349\u8ff7\u85cf\uff01\uff1f"},"content":{"rendered":"<div class=\"single-post-media clr\">\n<div class=\"post-thumbnail\"><img fetchpriority=\"high\" decoding=\"async\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/12\/AI-hide-and-seek-0.jpg\" alt=\"AI\u4e5f\u73a9\u6349\u8ff7\u85cf\uff01\uff1f\" width=\"500\" height=\"312\" \/><\/div>\n<\/div>\n<div class=\"entry clr\">\n<div class=\"pf-content\">\n<p><strong>\u64b0\u6587\uff0f\u9673\u5101\u7ff0<\/strong><\/p>\n<p><strong>\u6a21\u64ec\u751f\u7269\u7684\u6f14\u5316<\/strong><\/p>\n<p>\u4e00\u500b\u80fd\u8207\u7269\u7406\u4e16\u754c\u4e2d\u7684\u7269\u4ef6\u4e92\u52d5\u3001\u66ff\u4ee3\u4eba\u985e\u5b8c\u6210\u591a\u6a23\u4e14\u8907\u96dc\u4efb\u52d9\u7684\u667a\u6167\u9ad4\uff0c\u4e00\u76f4\u662f\u5f37\u5316\u5b78\u7fd2\u7684\u65b9\u5411\u4e4b\u4e00\u3002\u7136\u800c\u55ae\u667a\u6167\u9ad4\u7684\u5f37\u5316\u5b78\u7fd2\u8a2d\u5b9a\uff0c\u9664\u4e86\u8a2d\u8a08\u51fa\u5408\u9069\u7684\u300c\u734e\u52f5\u51fd\u6578\u300d\uff08reward function\uff09\u9700\u8981\u6295\u5165\u5927\u91cf\u6642\u9593\u8207\u4eba\u529b\u5916\uff0c\u904e\u65bc\u55ae\u4e00\u3001\u7406\u60f3\u5316\u7684\u6a21\u64ec\u74b0\u5883\uff0c\u4e5f\u8207\u73fe\u5be6\u4e16\u754c\u812b\u7bc0\u3002\u4e00\u7a2e\u89e3\u6c7a\u65b9\u6848\uff0c\u662f\u71df\u9020\u591a\u667a\u6167\u9ad4\u7684\u8a13\u7df4\u74b0\u5883\uff0c\u6a21\u64ec\u5730\u7403\u751f\u7269\u5f7c\u6b64\u7af6\u722d\u3001\u5408\u4f5c\u3001\u5055\u540c\u6f14\u5316\u7684\u904e\u7a0b\uff0c\u900f\u904e\u4e3b\u52d5\u5b78\u7fd2\u8207\u63a2\u7d22\uff08Autocurricula\uff09\uff0c\u667a\u6167\u9ad4\u5011\u5f97\u4ee5\u7a4d\u6975\u5229\u7528\u74b0\u5883\u512a\u52e2\u3001\u958b\u767c\u65b0\u6280\u80fd\uff0c\u4e26\u76f8\u4e92\u53cd\u5236\u3002<\/p>\n<div id=\"attachment_83293\" class=\"wp-caption aligncenter\"><img decoding=\"async\" class=\"size-full wp-image-83293\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/12\/AI-hide-and-seek-1.png\" alt=\"\" width=\"1123\" height=\"634\" \/><\/p>\n<p class=\"wp-caption-text\">OpenAI\u642d\u5efa\u7684\u6349\u8ff7\u85cf\u74b0\u5883\uff0c\u7d05\u8272\u7684\u4ee3\u8868\u7375\u4eba\uff0c\u85cd\u8272\u7684\u4ee3\u8868\u7375\u7269\uff0c\u7375\u4eba\u8207\u7375\u7269\u90fd\u80fd\u5920\u8207\u7a7a\u9593\u4e2d\u7684\u7269\u9ad4\u4e92\u52d5\uff0c\u4f8b\u5982\u63a8\u52d5\u7a7a\u9593\u4e2d\u7684\u7bb1\u5b50\u3002\uff08\u5716\u7247\u4f86\u6e90\uff1aKanitscheider et al., 2019.\uff09<\/p>\n<\/div>\n<p>\u4e0a\u5716\uff0c\u662fOpenAI\u5718\u968a\u6240\u642d\u5efa\u7684\u6a21\u64ec\u74b0\u5883\u3002\u900f\u904e\u591a\u500b\u7375\u4eba\uff08\u7d05\u8272\u4eba\u5f62\uff09\u8207\u7375\u7269\uff08\u85cd\u8272\u4eba\u5f62\uff09\u9593\u985e\u4f3c\u6349\u8ff7\u85cf\u7684\u904a\u6232\uff0c\u8a13\u7df4\u667a\u6167\u9ad4\u5011\u5584\u7528\u7a7a\u9593\u4e2d\u7684\u5404\u7a2e\u7269\u4ef6\uff08\u4f8b\u5982\u63a8\u52d5\u7a7a\u9593\u4e2d\u7684\u7bb1\u5b50\uff09\uff0c\u4e26\u767c\u5c55\u51fa\u5408\u4f5c\u6216\u7af6\u722d\u7684\u95dc\u4fc2\u4f86\u9054\u5230\u76ee\u6a19\uff08\u4f8b\u5982\u7375\u4eba\u5408\u4f5c\u6355\u6349\u7375\u7269\uff0c\u6216\u7375\u7269\u9593\u76f8\u4e92\u63a9\u8b77\uff09\u3002<\/p>\n<p>\u6a21\u64ec\u4e00\u958b\u59cb\uff0c\u667a\u6167\u9ad4\u53ea\u6703\u5728\u7a7a\u9593\u4e2d\u96a8\u610f\u8d70\u52d5\uff1b\u5728\u904a\u6232\u9032\u884c\u4e86269\u842c\u56de\u5408\u5f8c\uff0c\u667a\u6167\u9ad4\u5b78\u6703\u4e86\u7c21\u55ae\u7684\u8ffd\u9010\u8207\u9003\u8dd1\uff1b862\u842c\u56de\u5408\u5f8c\uff0c\u7375\u7269\u5b78\u6703\u53ef\u4ee5\u7528\u7bb1\u5b50\u5c07\u5165\u53e3\u5c01\u4f4f\uff0c\u5728\u53f3\u4e0b\u89d2\u7684\u5c0f\u623f\u9593\u4e2d\u8eb2\u904e\u7375\u4eba\u7684\u8ffd\u6355\uff1b\u4f46\u57281450\u842c\u56de\u5408\u5f8c\uff0c\u7375\u4eba\u7adf\u4e5f\u5b78\u6703\u4e86\u53cd\u5236\uff0c\u5c07\u659c\u5761\u62b5\u4f4f\u7246\u58c1\u722c\u9032\u7375\u7269\u6240\u5728\u7684\u7a7a\u9593\u3002\u7136\u800c\uff0c\u7375\u7269\u6700\u5f8c\u53c8\u5360\u4e86\u4e0a\u98a8\uff0c\u5b78\u6703\u5229\u7528\u7375\u4eba\u5728\u6bcf\u56de\u904a\u6232\u958b\u59cb\u6642\u6703\u6709\u6578\u79d2\u7684\u6642\u9593\u7121\u6cd5\u52d5\u5f48\uff0c\u5c07\u659c\u5761\u5148\u62d6\u9032\u5c0f\u623f\u9593\u5167\uff0c\u518d\u7528\u7bb1\u5b50\u5c01\u4f4f\u623f\u9593\u7684\u51fa\u5165\u53e3\uff0c\u4f7f\u5f97\u7375\u4eba\u6beb\u7121\u63a5\u89f8\u5230\u659c\u5761\u7684\u6a5f\u6703\u3002<\/p>\n<div id=\"attachment_83295\" class=\"wp-caption aligncenter\"><img decoding=\"async\" class=\"wp-image-83295 size-full\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/12\/AI-hide-and-seek-5.png\" alt=\"\" width=\"960\" height=\"720\" \/><\/p>\n<p class=\"wp-caption-text\">\uff08\u5716\u7247\u4f86\u6e90\uff1aKanitscheider et al., 2019.\uff09<\/p>\n<\/div>\n<p>\u9019\u6a23\u7684\u904e\u7a0b\u5c55\u793a\u4e86\u667a\u6167\u9ad4\u80fd\u900f\u904e\u4e0d\u65b7\u7684\u5617\u8a66\u8207\u5b78\u7fd2\uff0c\u5f9e\u96f6\u958b\u59cb\u767c\u5c55\u51fa\u8907\u96dc\u7684\u7af6\u722d\u6216\u5408\u4f5c\u7b56\u7565\uff0c\u800c\u4e14\u4e0d\u9700\u8981\u4eba\u70ba\u4ecb\u5165\u3002\u96a8\u8457\u8a13\u7df4\u6b21\u6578\u589e\u52a0\uff0c\u667a\u6167\u9ad4\u5c0d\u74b0\u5883\u7684\u638c\u63e1\u6108\u8da8\u7d14\u719f\uff0c\u6240\u8868\u73fe\u7684\u884c\u70ba\u4e5f\u8d8a\u4f86\u8d8a\u8907\u96dc\u3002<\/p>\n<p><strong>\u6295\u6a5f\u53d6\u5de7<\/strong><\/p>\n<p>\u7d93\u904e\u6f14\u5316\u7b97\u6cd5\u6216\u662f\u5f37\u5316\u5b78\u7fd2\u8a13\u7df4\u51fa\u7684\u667a\u6167\u9ad4\uff0c\u6709\u6642\u5019\u5247\u6703\u8868\u73fe\u51fa\u8d85\u4e4e\u60f3\u50cf\u537b\u53c8\u975e\u5e38\u6709\u6548\u7387\u7684\u6295\u6a5f\u884c\u70ba\u3002\u4ee5\u4e0a\u8ff0\u6349\u8ff7\u85cf\u7684\u74b0\u5883\u4f86\u8aaa\uff0c\u667a\u6167\u9ad4\u5b78\u6703\u5229\u7528\u74b0\u5883\u8a2d\u5b9a\u4e0a\u7684\u6f0f\u6d1e\uff0c\u985e\u4f3c\u73a9\u5bb6\u6feb\u7528\u904a\u6232\u88e1\u7684bug\u4ee5\u9054\u5230\u76ee\u7684\u3002\u4f8b\u5982\uff1a\u7375\u4eba\u9a30\u7a7a\u98db\u8d77\u76f4\u63a5\u98db\u9032\u88ab\u7375\u7269\u5c01\u4f4f\u51fa\u5165\u53e3\u7684\u5c0f\u623f\u9593\u5167\uff1b\u6216\u7375\u7269\u4ee5\u540c\u6a23\u7684\u65b9\u5f0f\u5c07\u659c\u5761\u64e0\u51fa\u5834\u5916\uff0c\u4f7f\u7375\u4eba\u7121\u6cd5\u5229\u7528\u659c\u5761\u9032\u5165\u5c0f\u623f\u9593\u3002\u53e6\u5916\uff0c\u7375\u4eba\u4e5f\u767c\u73fe\u81ea\u5df1\u53ef\u4ee5\u7ad9\u5728\u7bb1\u5b50\u4e0a\u79fb\u52d5\uff0c\u4ee5\u300c\u885d\u6d6a\u300d\u7684\u65b9\u5f0f\u9032\u5165\u7375\u7269\u7684\u85cf\u8eab\u4e4b\u8655\u3002<\/p>\n<p><strong>\u96fb\u7af6\u6210\u5c31<\/strong><\/p>\n<p>OpenAI\u9019\u6b21\u6349\u8ff7\u85cf\u4f7f\u7528\u7684\u6a21\u578b\u70baOpenAI Five\uff0c\u4e8b\u5be6\u4e0a\u65e9\u5df2\u6709\u767b\u5834\u7684\u8a18\u9304\u3002\u9084\u8a18\u5f97\u4eba\u985e\u8207OpenAI\u5728\u96fb\u73a9\u904a\u6232Dota\u4e0a\u7684\u4eba\u6a5f\u5927\u6230\u55ce\uff1f<\/p>\n<p>Dota\u662f\u4e00\u500b5\u5c0d5\u7684\u904a\u6232\uff0c\u6bcf\u4f4d\u73a9\u5bb6\u64cd\u63a7\u4e00\u96bb\u82f1\u96c4\uff0c\u8981\u76e1\u91cf\u5728\u5730\u5716\u4e0a\u64ca\u6bba\u5c0d\u624b\u6216\u53d6\u5f97\u8cc7\u6e90\uff0c\u904e\u7a0b\u4e2d\u6301\u7e8c\u64ca\u7834\u5c0d\u65b9\u7684\u9632\u79a6\u5854\uff0c\u6700\u7d42\u653b\u7834\u5c0d\u65b9\u4e3b\u5821\u7684\u904a\u6232\u3002\u96fb\u8166\u4f7f\u75285\u500b\u63a7\u5236\u5668\u4f86\u63a7\u52365\u500b\u82f1\u96c4\uff0c\u4e5f\u5c31\u662f\u8aaa5\u500b\u667a\u6167\u9ad4\u5fc5\u9808\u6709\u826f\u597d\u7684\u5408\u4f5c\u9ed8\u5951\uff0c\u624d\u80fd\u6253\u6557\u4eba\u985e\u73a9\u5bb6\u6240\u7d44\u6210\u7684\u6230\u968a\u3002<\/p>\n<p>\u7576\u6642OpenAI\u4f7f\u7528\u4e86\u53eb\u505aOpenAI Five\u7684\u6a21\u578b\uff0c\u6a21\u578b\u4e3b\u8981\u6709\u5e7e\u500b\u7279\u8272\uff1a\u7b2c\u4e00\uff0c\u6a21\u578b\u7684\u8f38\u5165\u4f7f\u7528\u4e86\u5927\u91cf\u7684\u7279\u5fb5\u5de5\u7a0b\uff0c\u9078\u53d6\u9069\u7576\u7684\u7279\u5fb5\uff0c\u4f8b\u5982\u82f1\u96c4\u7684\u8840\u91cf\u3001\u6280\u80fd\u72c0\u614b\u3001\u9b54\u529b\u3001\u4f4d\u7f6e\u7b49\u8cc7\u8a0a\uff1b\u7b2c\u4e8c\uff0c\u5f15\u5165\u4e86\u81ea\u7136\u8a9e\u8a00\u88e1\u7684LSTM\u3002LSTM\u7684\u7279\u6027\u662f\u80fd\u5920\u8655\u7406\u5e36\u6709\u6642\u5e8f\u6027\u7684\u8cc7\u6599\uff0c\u800c\u904a\u6232\u7684\u9032\u884c\u672c\u8eab\u5c31\u662f\u6709\u6642\u5e8f\u6027\u7684\uff1b\u7b2c\u4e09\uff0c\u5c0d\u65bc\u6578\u91cf\u4e0d\u4e00\u5b9a\u7684\u5c0f\u5175\u8207\u6575\u4eba\uff0c\u6a21\u578b\u4f7f\u7528\u4e86max pooling\u9032\u884c\u5408\u4f75\u3002<\/p>\n<p>\u5f37\u5316\u5b78\u7fd2\u7684\u734e\u52f5\u4e5f\u4f9d\u64da\u4e0d\u540c\u7684\u8a13\u7df4\u968e\u6bb5\u7d66\u4e88\u8abf\u6574\uff1a\u5728\u8a13\u7df4\u7684\u521d\u671f\u5f37\u8abf\u6bcf\u500b\u63a7\u5236\u5668\u7684\u500b\u5225\u8868\u73fe\uff0c\u4f7f\u96fb\u8166\u80fd\u5feb\u901f\u7684\u4e0a\u624b\u904a\u6232\uff1b\u7b49\u5230\u96fb\u8166\u6709\u4e00\u5b9a\u7684\u63a7\u5236\u80fd\u529b\u4ee5\u5f8c\uff0c\u958b\u59cb\u5f37\u8abf\u591a\u667a\u6167\u9ad4\u9593\u7684\u5408\u4f5c\uff0c\u5171\u540c\u9054\u6210\u7372\u5f97\u6700\u7d42\u52dd\u5229\u7684\u76ee\u6a19\u3002\u85c9\u7531\u6578\u4ee5\u842c\u8a08\u7684CPU\u9032\u884c\u591a\u6b21\u7684\u8a13\u7df4\uff0c\u50c5\u50c5\u7d50\u5408\u5f37\u5316\u5b78\u7fd2\u88e1\u7d93\u5178\u800c\u6709\u6548\u7684\u7b97\u6cd5\uff0c\u4f9d\u7136\u80fd\u7a81\u7834\u591a\u667a\u6167\u9ad4\u8a13\u7df4\u4e0a\u7684\u56f0\u96e3\u3002<\/p>\n<p><strong>\u5ef6\u4f38\u95b1\u8b80\uff1a<\/strong>\u3008<a href=\"http:\/\/highscope.ch.ntu.edu.tw\/wordpress\/?p=79710\">\u904a\u6232AI\u4e0d\u662f\u53ea\u6703\u73a9\u904a\u6232<\/a>\u3009\u3001\u3008<a href=\"http:\/\/highscope.ch.ntu.edu.tw\/wordpress\/?p=79580\">AI\u5316\u8eab\u96fb\u7af6\u9078\u624b\uff0c\u5c55\u73fe\u5718\u968a\u5408\u4f5c\u9ed8\u5951<\/a>\u3009<\/p>\n<p>&nbsp;<\/p>\n<p><strong>\u53c3\u8003\u8cc7\u6599<\/strong><\/p>\n<ol>\n<li>B. Baker, I. Kanitscheider, T. Markov, Y. Wu, G. Powell, B. McGrew, and I. Mordatch, \u201c<a href=\"https:\/\/arxiv.org\/abs\/1909.07528\">Emergent Tool Use From Multi-Agent Autocurricula<\/a>,<em>\u201d\u00a0arXiv.org<\/em>, 2019.<\/li>\n<li>L. Bu\u015foniu, R. Babuska and B. De Schutter (2010).\u00a0<em>\u201cMulti-agent reinforcement learning: An overview,\u201d Chapter 7 in Innovations in Multi-Agent Systems and Applications \u2013 1 (D. Srinivasan and L.C. Jain, eds.), vol. 310 of Studies in Computational Intelligence,\u00a0<\/em>Berlin, Germany: Springer, pp. 183\u2013221.<\/li>\n<\/ol>\n<p>(\u672c\u6587\u7531\u6559\u80b2\u90e8\u88dc\u52a9\u300cAI\u5831\u5831\u2500AI\u79d1\u666e\u63a8\u5ee3\u8a08\u756b\u300d\u57f7\u884c\u5718\u968a\u7de8\u8b6f)<\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>OpenAI\u8fd1\u671f\u767c\u8868\u4e86\u4e00\u7bc7\u6709\u8da3\u7684\u7814\u7a76\uff0c\u5c55\u73fe\u4e86\u4ee5\u5f37\u5316\u5b78\u7fd2\u8a13\u7df4\u51fa\u7684\u591a\u500b\u667a\u6167\u9ad4\u80fd\u5920\u5f7c\u6b64\u5408\u4f5c\u3001\u7af6\u722d\uff0c\u751a\u81f3\u80fd\u627e\u5230\u4e26\u5229\u7528\u6a21\u64ec\u74b0\u5883\u4e2d\u7684\u6f0f\u6d1e\uff0c\u8868\u73fe\u51fa\u8907\u96dc\u884c\u70ba\u3002<\/p>\n","protected":false},"author":21,"featured_media":38485,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3772,3293],"tags":[5766,4989,5767,5513],"aioseo_notices":[],"jetpack_featured_media_url":"https:\/\/case.ntu.edu.tw\/blog\/wp-content\/uploads\/2022\/01\/AI-hide-and-seek-0.jpg","_links":{"self":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38484"}],"collection":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/users\/21"}],"replies":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=38484"}],"version-history":[{"count":1,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38484\/revisions"}],"predecessor-version":[{"id":38486,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38484\/revisions\/38486"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/media\/38485"}],"wp:attachment":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38484"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38484"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38484"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}