{"id":5015,"date":"2017-06-03T01:00:00","date_gmt":"2017-06-02T17:00:00","guid":{"rendered":"http:\/\/swarma.org\/%swarma-paper%\/"},"modified":"2018-12-05T22:24:09","modified_gmt":"2018-12-05T14:24:09","slug":"%e6%9c%ba%e5%99%a8%e4%b9%9f%e6%9c%89%e5%a5%bd%e5%a5%87%e5%bf%83%e5%90%97%ef%bc%9f-%e6%9d%8e%e5%ab%a3%e7%84%b6","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=5015","title":{"rendered":"\u673a\u5668\u4e5f\u6709\u597d\u5947\u5fc3\u5417\uff1f| \u674e\u5ae3\u7136"},"content":{"rendered":"<div class=\"bpp-post-content\">\n<section class=\"xmteditor\" style=\"display:none;\"  ><\/section>\n<section  style=\"display: none; height: 0px;\">\n<div class=\"post-image\"><img      src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma10-1534232214.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/div>\n<\/section>\n<p style=\"margin-left: 0.5em; margin-right: 0.5em; letter-spacing: 0.5px;\"><img      src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma2-1534232214-1.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<section  class=\"mpa-template\" >\n<section donone=\"shifuMouseDownCard(&#39;shifu_c_029&#39;)\" label=\"Copyright Reserved by PLAYHUDONG.\" style=\"margin: 1em 0px; padding: 0px; max-width: 100%; color: rgb(123, 12, 0); font-size: 16px; white-space: normal; line-height: 25.6px; border-width: 0px; border-style: none; background-color: rgb(255, 255, 255); border-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\">\n<p style=\"margin: 10px 0px 0px 1em; padding: 0px; max-width: 100%; line-height: 1.4; border-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\"><span style=\"margin: 0px; padding: 0px; max-width: 100%; font-size: 14px; border-color: rgb(123, 12, 0); color: rgb(255, 255, 255); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\"><strong style=\"margin: 0px; padding: 0px; max-width: 100%; border-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\"><span style=\"font-size: 14px; border-color: rgb(123, 12, 0); margin: 0px; padding: 0.2em 0.5em; max-width: 100%; font-weight: inherit; text-decoration: inherit; border-radius: 0.3em; text-align: center; background-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\">\u5bfc\u8bed<\/span><\/strong><\/span><\/p>\n<section class=\"\" style=\"margin: -0.7em 0px 0px; padding: 0px; max-width: 100%; border-width: 1px; border-style: solid; border-color: rgb(123, 12, 0); border-radius: 0.4em; color: rgb(123, 12, 0); font-size: 1em; box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\">\n<section style=\"margin: 0px; padding: 1.4em 1em 1em; max-width: 100%; border-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\">\n<section style=\"margin: 0px; padding: 0px; max-width: 100%; font-size: 1em; font-weight: inherit; text-align: inherit; text-decoration: inherit; border-color: rgb(123, 12, 0); box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\">\n<p style=\"max-width: 100%; clear: both; min-height: 1em; border-color: rgb(123, 12, 0); text-align: left; margin-top: 10px; box-sizing: border-box !important; word-wrap: break-word !important; overflow-wrap: break-word !important;\"><span style=\"color: rgb(136, 136, 136); font-size: 13px; font-weight: inherit; text-decoration: inherit; text-align: inherit;\">\u662f\u7684\uff0c\u673a\u5668\u4e5f\u53ef\u4ee5\u6709\u597d\u5947\u5fc3\u3002\u8fd9\u4e0d\uff0c\u79d1\u5b66\u5bb6\u4eec\u6b63\u5728\u52aa\u529b\u8d4b\u4e88\u4eba\u5de5\u667a\u80fd\u7b97\u6cd5\u597d\u5947\u7684\u672c\u80fd\uff0c\u5e2e\u52a9\u5b83\u4eec\u66f4\u597d\u5730\u5b8c\u6210\u5f3a\u5316\u5b66\u4e60\u3002\u5728ICML2017\u7684\u8fd9\u7bc7\u8bba\u6587\u300aCuriosity-driven Exploration by Self-supervised Prediction\u300b\u4e2d\uff0c\u4f5c\u8005\u5c31\u7528\u597d\u5947\u5fc3\u4f5c\u4e3a\u4e00\u79cd\u5956\u52b1\u6fc0\u52b1\uff0c\u6765\u8ba9\u4eba\u5de5\u667a\u80fd\u63a7\u5236\u7684\u9a6c\u5229\u5965\u6551\u51fa\u516c\u4e3b\u3002<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">ICML 2017 accepted paper \u5df2\u7ecf\u516c\u5e03\u3002\u5176\u4e2d\u4e00\u7bc7\u503c\u5f97\u5173\u6ce8\u7684\u5de5\u4f5c\u5c31\u662f\u6765\u81ea UC Berkeley \u7ec4\u7684\u300aCuriosity-driven Exploration by Self-supervised Prediction\u300b\u3002\u8fd9\u7bc7\u5de5\u4f5c\u7684\u4e3b\u5f20\u662f\uff0c\u5373\u4f7f\u6ca1\u6709\u660e\u663e\u7684\uff08\u5916\u754c\uff09reward\uff0c\u7528\u4ed6\u4eec\u7684\u7b97\u6cd5\u4e5f\u53ef\u4ee5\u8ba9\u9a6c\u91cc\u5965\u5927\u53d4\u4e00\u8def\u9876\u8611\u83c7\u5730\u52c7\u5f80\u76f4\u524d\u53bb\u6551\u78a7\u59ec\u516c\u4e3b\u5566\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><img   class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma4-1534232215.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u4e0d\u8fc7\u5176\u5b9e\u5566\uff0c\u5f88\u591a\u5de5\u4f5c\u90fd\u6709\u975e\u5e38\u76f8\u4f3c\u7684\u201c\u524d\u4eba\u201d\u5de5\u4f5c\u3002\u867d\u7136\u4e0d\u4e00\u5b9a\u6ca1\u6709\u65b0\u610f\uff0c\u4f46\u662f\u4e92\u76f8\u5bf9\u6bd4\u6765\u770b\u4f1a\u5bf9\u6574\u4e2a\u5927\u7684\u601d\u8def\u6709\u66f4\u4e0d\u9519\u7684\u7406\u89e3\u3002\u4eca\u5929\u8981\u5bf9\u6bd4\u7684\u5de5\u4f5c\u5206\u522b\u662f\uff1a<br  \/><\/span><\/p>\n<blockquote>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"color: rgb(136, 136, 136);\"><strong><span style=\"font-size: 13px;\">1. \u300aCuriosity-driven Exploration by Self-supervised Prediction\u300b by UCB. ICML 2017.<\/span><\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"color: rgb(136, 136, 136);\"><strong><span style=\"color: rgb(136, 136, 136); font-size: 13px;\">2. \u300aLearning to Act by Predicting the Future\u300b by IntelLab. ICLR 2017 (oral).<\/span><\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"color: rgb(136, 136, 136);\"><strong><span style=\"color: rgb(136, 136, 136); font-size: 13px;\">3. \u300aDeep Successor Reinforcement Learning\u300b by MIT &amp; Harvard. NIPS 2016 workshop.<\/span><\/strong><\/span><\/p>\n<\/blockquote>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u8fd9\u4e09\u7bc7\u5de5\u4f5c\u90fd\u662f\u60f3\u89e3\u51b3\u590d\u6742\u4efb\u52a1\u4e2d\u5956\u52b1\u5f88\u7a00\u758f\u7684\u95ee\u9898\u3002\u6bd4\u5982\u5982\u679c\u6211\u4eec\u53ea\u7528\u662f\u5426\u6700\u7ec8\u5b8c\u6210\u4e86\u4e00\u4e2a\u4efb\u52a1\uff0c\u5b8c\u6210\u5c31\u7ed9\u4e00\u4e2a\u6b63\u7684\u5956\u52b1\uff0c\u6ca1\u5b8c\u6210\u5c31\u4e00\u76f4\u6ca1\u6709\u5956\u52b1\uff0c\u90a3\u4e48\u6574\u4e2a\u9a6c\u5c14\u79d1\u592b\u51b3\u7b56\u8fc7\u7a0b\u53ea\u6709\u6700\u7ec8\u90a3\u4e00\u6b65\u624d\u4f1a\u5f97\u5230\u5956\u52b1\uff0c\u8fd9\u5c31\u662f\u4e00\u4e2a\u5f88\u4e25\u91cd\u7684\u7a00\u758f\u95ee\u9898\u3002\u8fd9\u4e5f\u662f\u5927\u90e8\u5206\u589e\u5f3a\u5b66\u4e60\u7b97\u6cd5\u9762\u4e34\u7684\u6700\u4e3b\u8981\u7684\u95ee\u9898\u4e4b\u4e00\u3002<br  \/><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u4e3a\u4e86\u89e3\u51b3\u8fd9\u6837\u7684\u95ee\u9898\uff0c\u5176\u5b9e\u8fc7\u53bb\u5df2\u7ecf\u6709\u4e0d\u5c11\u5de5\u4f5c\u63d0\u51fa\u4e86\u4e00\u4e9b\u89e3\u51b3\u601d\u8def\u3002\u5982\u679c\u6211\u4eec\u628a\u8fd9\u4e9b\u5b8c\u6210\u4efb\u52a1\u540e\u4f1a\u5f97\u5230\u7684\u6700\u7ec8\u5956\u52b1\u53eb\u505a external reward\uff08\u5916\u90e8\u6fc0\u52b1\uff09\uff0c\u6211\u4eec\u4e5f\u53ef\u4ee5\u901a\u8fc7\u4e00\u4e9b\u522b\u7684\u8bbe\u8ba1\u6765\u4e3a agent \u63d0\u4f9b intrinsic reward\/motivation\uff08\u5185\u90e8\u6fc0\u52b1\uff09\u3002\u8fd9\u7bc7 ICML \u7684\u5de5\u4f5c[1] \u5c31\u662f\u5c06\u8fd9\u79cd\u5185\u90e8\u6fc0\u52b1\u8bbe\u8ba1\u4e3a\u7279\u522b\u7684 curiosity-driven\uff08\u597d\u5947\u5fc3\u9a71\u52a8\uff09\u7684\u5f62\u5f0f\u3002\u5e76\u4e14\u4ed6\u81ea\u5df1\u4e5f\u63d0\u5230\u4e86\uff0c\u8fd9\u79cd\u5f62\u5f0f\u5728\u8fc7\u53bb\u7684\u5de5\u4f5c\u91cc\u4e5f\u5df2\u7ecf\u88ab\u63a2\u7a76\u591a\u6b21\uff08\u6709\u673a\u4f1a\u4e0b\u6b21\u518d\u548c\u5927\u5bb6\u4e00\u8d77\u603b\u7ed3\u8ba8\u8bba\uff09\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u597d\u5947\u5fc3\u9a71\u52a8\u7684\u5de5\u4f5c\u5982\u679c\u653e\u5728\u589e\u5f3a\u5b66\u4e60\u7684\u6846\u67b6\u4e2d\u6765\u770b\uff0c\u4e3b\u8981\u6709\u4e24\u5927\u7c7b\u5177\u4f53\u5f62\u5f0f\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">1\uff09\u6fc0\u52b1 agent \u53bb\u63a2\u7d22\u66f4\u591a\u7684\u65b0\u72b6\u6001\uff08novel state\uff09\uff1b<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">2\uff09\u6fc0\u52b1 agent \u53bb\u6267\u884c\u4e00\u4e9b\u4f1a\u51cf\u5c11\u81ea\u5df1\u5bf9\u4e8e\u672a\u6765\u4e0d\u786e\u5b9a\u6027\u7684\u52a8\u4f5c\uff0c\u6216\u8005\u53ef\u4ee5\u53cd\u8fc7\u6765\u7406\u89e3\u4e3a\u53bb\u6267\u884c\u4e00\u4e9b\u80fd\u589e\u52a0\u81ea\u5df1\u5bf9\u4e8e\u672a\u6765\u7684\u4fe1\u606f\u7684\u52a8\u4f5c\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u8fd9\u4e24\u4e2a\u601d\u60f3\u90fd\u975e\u5e38\u597d\u7406\u89e3\uff0c\u51e0\u4e4e\u53ef\u4ee5\u5b8c\u5168\u5728\u4eba\u7c7b\u7684\u5b66\u4e60\u4e2d\u627e\u5230\u6620\u5c04\u3002\u7136\u800c\u53ea\u8003\u8651\u8fd9\u4e24\u70b9\u5219\u4f1a\u51fa\u73b0\u4e00\u4e9b\u95ee\u9898\uff0c\u5f53\u5f53\u524d\u95ee\u9898\u6216\u8005\u73af\u5883\u5b58\u5728\u5f88\u591a\u566a\u97f3\u7684\u65f6\u5019\uff0c\u8fd9\u7c7b\u9a71\u52a8\u65b9\u6cd5\u5f80\u5f80\u4f1a\u5bfc\u81f4 agent \u88ab\u65e0\u5173\u5e72\u6270\u7269\u5438\u5f15\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0cICML 2017[1] &nbsp;\u7684\u601d\u8def\u5c31\u662f\u53bb\u5c3d\u91cf\u51cf\u5c11\u566a\u97f3\u7684\u5f71\u54cd\u3002\u56e0\u4e3a\u53bb\u6fc0\u52b1 agent \u53bb\u6267\u884c\u90a3\u4e9b\u4f1a\u964d\u4f4e\u81ea\u5df1\u5bf9\u4e8e\u672a\u6765\u4e0d\u786e\u5b9a\u884c\u7684\u52a8\u4f5c\uff0c\u5fc5\u7136\u6d89\u53ca\u5230\u8981\u53bb\u9884\u6d4b\u67d0\u4e2a\u52a8\u4f5c\u4f1a\u4ea7\u751f\u600e\u6837\u7684\u5f71\u54cd\uff0c\u6240\u4ee5\u4ed6\u4eec\u7684\u601d\u8def\u662f\u53ea\u9884\u6d4b\u90a3\u4e9b\u548c\u5f53\u524d\u52a8\u4f5c\u6709\u5173\u7684\u5f71\u54cd\u3002\u4e5f\u5c31\u662f\u8bf4\u5f53 agent \u6267\u884c\u4e00\u4e2a\u52a8\u4f5c\u6216\u8005\u968f\u7740 timestep \u7684\u8fdb\u884c\uff0c\u73af\u5883\u7684\u53d8\u5316\u6709\u4e24\u5927\u7c7b\uff0c\u4e00\u7c7b\u662f\u7531 agent \u9020\u6210\u7684\uff0c\u4e00\u7c7b\u5219\u662f\u566a\u97f3\uff0c\u548c agent \u5982\u4f55\u9009\u62e9\u52a8\u4f5c\u65e0\u5173\u3002\u540e\u8005\u662f ICML 2017[1] &nbsp;\u8fd9\u7bc7\u5de5\u4f5c\u60f3\u8981\u6392\u9664\u6389\u7684\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u6240\u4ee5 ICML 2017 \u7684\u5de5\u4f5c[1] \u5c31\u662f\u53ea\u9884\u6d4b\u90a3\u4e9b\u548c agent \u76f8\u5173\u7684\u73af\u5883\u53d8\u5316\uff0c\u4e5f\u5c31\u662f\u5c06\u9884\u6d4b\u62c6\u89e3\uff08factorize\uff09\u3002\u4ece\u8fd9\u4e2a\u89d2\u5ea6\u53bb\u7406\u89e3\u7684\u8bdd\uff0cICML 2017 \u7684\u5de5\u4f5c[1] \u5219\u548c\u4e00\u7bc7 NIPS 2016 \u7684 workshop[3] \u975e\u5e38\u76f8\u5173\u3002NIPS 2016 workshop \u8fd9\u7bc7\u5de5\u4f5c\u4e2d\u63d0\u51fa\u7528\u4e00\u79cd\u53eb successor representation\uff08SR\uff09\u7684\u65b9\u6cd5\u53bb\u6c42\u89e3 RL \u95ee\u9898\uff0c\u8fd9\u4e2a\u65b9\u6cd5\u7684\u4e3b\u8981\u601d\u8def\u5c31\u662f\u5c06 value function \u62c6\u6210\uff08factorize\uff09\u4e24\u90e8\u5206\uff1a\u4e00\u4e2a\u662f\u901a\u8fc7\u9884\u6d4b\u4efb\u52a1\u5b66\u4e60\u5230\u7684\u8868\u8fbe\uff0c\u548c\u4e00\u4e2a\u4e0e\u8868\u8fbe\u7d27\u5bc6\u76f8\u5173\uff08\u4f4e\u566a\u97f3\u7684\uff09\u6fc0\u52b1\u51fd\u6570\uff0c\u800c\u4e14\u8fd9\u4e2a\u6fc0\u52b1\u51fd\u6570\u8fd8\u662f immediate and intrinsic \u7684\u3002\u4e0b\u9762\u5c31\u4f1a\u4ece\u5bf9\u6bd4\u8fd9\u4e24\u8005\u7684\u89d2\u5ea6\u6765\u4ecb\u7ecd\u8fd9\u7bc7\u5de5\u4f5c\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><strong><span style=\"font-size: 13px;\">ICML 2017 \u4e2d\u63d0\u51fa\u7684\u6a21\u578b\u7ed3\u6784\u53eb Intrinsic Curiosity Module\uff08ICM\uff09\uff1a<\/span><\/strong><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma1-1534232215.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u5176\u4e2d r^e \u4ee3\u8868\u5916\u90e8\u6fc0\u52b1\uff08external reward\uff09\uff0cr^i \u4ee3\u8868\u5185\u90e8\u6fc0\u52b1\uff08intrinsic reward\uff09\u3002\u53ef\u4ee5\u770b\u5230\u9996\u5148\uff0c\u6bcf\u4e2a\u72b6\u6001\u90fd\u4f1a\u88ab ICM \u7f16\u7801\u6210 phi (s_t) \u5f97\u5230\u4e00\u4e2a\u7279\u5f81\u7a7a\u95f4\u4e2d\u7684\u8868\u8fbe\u3002\u8fd9\u4e2a\u7279\u5f81\u7a7a\u95f4\u7684\u8868\u8fbe\uff0c\u6709\u4e24\u4e2a\u4f5c\u7528\uff0c\u4e00\u4e2a\u662f\u7528\u4e8e\u7ecf\u8fc7 inverse model\uff0c\u9884\u6d4b\u5f53\u524d\uff08\u6216\u8005\u8bf4\u4e0a\u4e00\u523b\u7684\uff09\u52a8\u4f5c\uff0c\u53e6\u4e00\u4e2a\u662f\u7528\u4e8e\u7ecf\u8fc7 forward model\uff0c\u9884\u6d4b\u4e0b\u4e00\u523b\u7684\u72b6\u6001\u8868\u8fbe\u3002<strong>\u800c\u8fd9\u79cd\u9884\u6d4b\u7684\u8bef\u5dee\u5c31\u88ab\u8ba4\u4e3a\u662f\u597d\u5947\u5fc3\u6fc0\u52b1\uff08curiosity reward\uff09\u3002<\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma8-1534232215.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma1-1534232215.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma5-1534232216.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">inverse model \u548c forward model \u7ed3\u5408\u5728\u4e00\u8d77\uff0c\u53cc\u5411\u6a21\u578b\u5171\u540c\uff0c\u5c31\u6784\u6210\u4e86\u4e00\u79cd\u81ea\u6211\u76d1\u7763\uff08self-supervised\uff09\u7684\u5b66\u4e60\u65b9\u5f0f\u3002\u4ece inverse model \u7684\u89d2\u5ea6\u770b\uff08\u516c\u5f0f2\uff09\uff0c\u7f51\u7edc g \u53ea\u4f1a\u53bb\u5b66\u4e60\u90a3\u4e9b\u5bf9\u4e8e\u9884\u6d4b\u4e0a\u4e00\u523b\u52a8\u4f5c a_t \u6709\u7528\u7684\u7279\u5f81\u3002\u540c\u7406\uff0c\u4ece forward model \u7684\u89d2\u5ea6\u770b\uff08\u516c\u5f0f4\uff09\uff0c\u7f51\u7edc g \u53ea\u4f1a\u5173\u5fc3\u90a3\u4e9b\u548c\u9884\u6d4b\u6709\u5173\u7684\u7279\u5f81\u3002\u6240\u4ee5\uff0c\u901a\u8fc7\u8fd9\u6837\u7684\u65b9\u5f0f\uff0c\u73af\u5883\u4e2d\u90a3\u4e9b\u4e0d\u80fd\u88ab\u9884\u6d4b\u7684\u53d8\u5316\uff0c\u5373\u566a\u97f3\uff0c\u5c31\u4e0d\u4f1a\u88ab\u201c\u6ce8\u610f\u201d\u5230\uff0c\u81ea\u7136\u4e5f\u5c31\u88ab\u6392\u9664\u6389\u4e86\u3002<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u5728\u56de\u8fc7\u5934\u6765\u770b NIPS 2016 \u7684\u5de5\u4f5c[3]\u3002\u8fd9\u4e2a successor representation\uff08SR\uff09\u5176\u5b9e\u4ece\u540d\u5b57\u4e0a\u5c31\u53cd\u6620\u4e86\u7c7b\u4f3c\u7684\u601d\u60f3\u3002SR \u7684\u516c\u5f0f\u5982\u4e0b\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma4-1534232216.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u901a\u8fc7\u8fd9\u5176\u4e2d\u6db5\u76d6\u7684\u6307\u793a\u51fd\u6570\uff0c\u53ef\u4ee5\u5c06 SR \u7684\u516c\u5f0f\u7406\u89e3\u4e3a\u5df2\u7ecf\u8986\u76d6\/\u8bbf\u95ee\u8fc7\u7684\u72b6\u6001\u7684\u6570\u76ee\uff08state visitation count\uff09\u3002\u628a SR \u7684\u516c\u5f0f\u548c\u66f4\u719f\u6089\u7684 Q-function \u5bf9\u6bd4\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma6-1534232216.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u5c31\u4f1a\u53d1\u73b0\u4e24\u8005\u4e4b\u95f4\u6709\u5f88\u591a\u76f8\u4f3c\u4e4b\u5904\u3002Q-function \u7528\u6765\u8ba1\u7b97\u7684 expected discounted future return, SR \u8ba1\u7b97\u7684\u662f expected discounted future state occupancy. \u6709\u4e86\u8fd9\u6837\u7684 SR \u8868\u8fbe\u5f0f\uff0c<strong>\u5c31\u53ef\u4ee5\u8fdb\u4e00\u6b65\u5c06 Q-function \u62c6\u89e3\uff08factorize\uff09\u6210 SR \u8868\u8fbe\u5f0f\u4e0e\u6fc0\u52b1\u51fd\u6570\u8868\u8fbe\u5f0f\u7684\u5185\u79ef\u4e86\uff1a<\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma8-1534232216.png\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u81f3\u6b64\uff0c\u518d\u56de\u987e\u4e00\u4e0b NIPS 2016 workshop[3] \u7684\u601d\u8def\u3002\u4ed6\u4eec\u901a\u8fc7 SR \u7684\u65b9\u5f0f\uff0c\u5c06\u589e\u5f3a\u5b66\u4e60\u62c6\u89e3\u6210\u7279\u5f81\u7a7a\u95f4\u7684\u8868\u8fbe\u5b66\u4e60\uff08SR\uff09\u548c\u57fa\u4e8e\u8868\u8fbe\u7684\u6fc0\u52b1\u51fd\u6570\u5b66\u4e60\uff08R\uff09\u3002\u5e76\u4e14\u8fd9\u4e2a\u8868\u8fbe\u5b66\u4e60\u7684 SR \u62df\u5408\u7684\u4e5f\u662f\u5bf9\u4e8e\u72b6\u6001\u7a7a\u95f4\u7684\u63a2\u7d22\u7a0b\u5ea6\u2014\u2014\u4e0e curiosity-driven \u6216\u8005\u8bf4\u63a2\u7d22\u66f4\u591a\u672a\u77e5\u9886\u57df\u7684\u601d\u8def\uff0c\u662f\u4e00\u8109\u76f8\u627f\u7684\u3002\u4ece\u6b64\u4e5f\u5c31\u770b\u51fa\u4e86 ICML 2017[1] \u548c NIPS 2016 workshop[3] \u7684\u76f8\u4f3c\u4e4b\u5904\u3002\u800c\u8fd9\u91cc\u7684 R \u53c8\u53ef\u4ee5\u8fdb\u4e00\u6b65\u62c6\u5206\u6210\u5916\u90e8\u6fc0\u52b1\u548c\u5185\u90e8\u6fc0\u52b1\u3002<\/span><\/p>\n<p style=\"margin-left: 0.5em; margin-right: 0.5em; letter-spacing: 0.5px;\"><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><strong><span style=\"font-size: 13px;\">\u518d\u6765\u770b\u53e6\u4e00\u7bc7\u6bd4\u8f83\u76f8\u4f3c\u7684\u5de5\u4f5c\uff0c\u662f\u6765\u81ea\u4e8e ICLR 2017 \u7684 oral \u4e4b\u4e00[2]\uff0c<\/span><\/strong><span style=\"font-size: 13px;\">\u4e3b\u8981 direct prediction of future\u3002\u4ed6\u4eec\u7684\u5de5\u4f5c\u540c\u6837\u60f3\u5bfb\u627e\u9664\u4e86\u6700\u7ec8\u7684\u7a00\u758f\u7684\u5916\u90e8\u6fc0\u52b1\u4ee5\u5916\u7684\u201c\u6fc0\u52b1\u201d\u3002\u5982\u679c\u628a\u5956\u52b1\u51fd\u6570\u770b\u505a\u662f\u5bf9\u5f53\u524d\u7b56\u7565\u6216\u8005\u52a8\u4f5c\u7684\u4e00\u79cd\u8bc4\u4f30\uff0c\u90a3\u4e48 ICLR 2017 \u7684\u5de5\u4f5c\u5219\u53ef\u4ee5\u770b\u505a\u53bb\u5bfb\u627e\u6bcf\u4e2a\u72b6\u6001\u4e0b\u7684\u591a\u79cd\u76f8\u5173\u201c\u8bc4\u4f30\u201d\u6307\u6807\uff0c\u4e5f\u53eb measurement\u3002\u4e5f\u56e0\u6b64\uff0c\u4ed6\u4eec\u5047\u8bbe\uff0c\u4e00\u4e2a\u589e\u5f3a\u5b66\u4e60\u7684\u4efb\u52a1\u662f\u53ef\u4ee5\u8868\u793a\u6210 measurements \u7684\u7ec4\u5408\u7684\u3002\u6709\u4e86\u8fd9\u4e2a\u5047\u8bbe\uff0c\u4ed6\u4eec\u8fdb\u4e00\u6b65\u8ba4\u4e3a\uff0c\u9884\u6d4b\u8fd9\u4e9b measurements \u5c31\u53ef\u4ee5\u5f97\u5230\u5bf9\u4e8e\u81ea\u5df1\u5f53\u524d\u7b56\u7565\u7684\u4e00\u79cd\u8bc4\u4f30\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u81ea\u5df1\u9009\u62e9\u7684\u8fd9\u4e00\u6b65\u52a8\u4f5c\u4f1a\u4ea7\u751f\u4ec0\u4e48\u6837\u7684\u53d8\u5316\uff0c\u53ef\u4ee5\u901a\u8fc7\u9884\u6d4b measurements \u6765\u5b66\u4e60\u5230\u3002\u4e5f\u5c31\u53ef\u4ee5\u5b9e\u73b0\u81ea\u6211\u76d1\u7763\uff08self-supervision\uff09\u3002\u8fd9\u4e2a\u601d\u8def\u662f\u548c ICM[1] \u4e00\u81f4\u7684\u3002\u53ea\u4e0d\u8fc7\u6700\u7ec8\u5bf9\u4e8e reward \u7684\u523b\u753b\uff0c\u4e24\u8fb9\u91c7\u7528\u4e86\u4e0d\u540c\u7684\u65b9\u5f0f\u3002<\/span><\/p>\n<p style=\"margin-left: 0.5em; margin-right: 0.5em; letter-spacing: 0.5px;\"><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u81f3\u6b64\uff0c\u4e09\u7bc7\u5de5\u4f5c\u7684\u5927\u81f4\u601d\u8def\u90fd\u4ecb\u7ecd\u5b8c\u6bd5\u4e86\u3002\u53ef\u4ee5\u770b\u5230\u4e09\u8005\u975e\u5e38\u76f8\u4f3c\uff0c\u90fd\u662f\u5c3d\u53ef\u80fd\u5730\u63d0\u4f9b\u66f4\u591a\u7684 immediate reward\u3002\u8fd9\u79cd\u601d\u8def\u4e5f\u88ab\u7edf\u79f0\u4e3a reward shaping\u3002\u540c\u65f6\uff0c\u6709\u4e00\u4e9b\u5de5\u4f5c\u8fd8\u4f1a\u5c06\u8fd9\u4e9b\u4e2d\u95f4\u72b6\u6001\u7684\u6fc0\u52b1\uff0c\u53eb\u505a subgoal\uff0cskill \u6216\u8005 option\u3002\u4e09\u7bc7\u5de5\u4f5c\u751a\u81f3\u8fd8\u505a\u4e86\u76f8\u540c\u7684\u5b9e\u9a8c\u3002ICM[1] \u9664\u4e86\u6253\u9a6c\u91cc\u5965\u4ee5\u5916\uff0c\u8fd8\u6253\u4e86 Doom\u3002\u800c[2][3] \u4e5f\u6253\u4e86 Doom\u3002\u6765\u770b\u770b\u4e09\u8005\u6253\u7684 Doom \u7ed3\u679c\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><img   class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma5-1534232216.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><strong style=\"color: rgb(136, 136, 136);\"><span style=\"font-size: 11px;\">ICM [1] \u7684\u8868\u73b0<\/span><\/strong><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><img   class=\"\"   style=\"white-space: normal;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma1-1534232218.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><span style=\"text-decoration: none; color: rgb(136, 136, 136);\"><strong><span style=\"color: rgb(136, 136, 136); text-decoration: none; font-size: 11px;\">ICLR[2] \u7684\u8868\u73b0<\/span><\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><img   class=\"\"   style=\"color: rgb(136, 136, 136); font-size: 14px; text-align: center; white-space: normal;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma5-1534232221.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><span style=\"text-decoration: none; color: rgb(136, 136, 136);\"><strong><span style=\"text-decoration: none; font-size: 11px;\">NIPS 2016 worksho, SR[3] \u7684\u8868\u73b0<\/span><\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u4e09\u8005\u90fd\u5c55\u73b0\u4e86\u81ea\u5df1\u5bf9\u4e8e\u672a\u77e5\u5730\u56fe\u6216\u8005\u8bf4\u672a\u77e5\u73af\u5883\u7684\u6cdb\u5316\u6027\uff0c\u540c\u65f6\u4e5f\u5c55\u73b0\u4e86\u8fd9\u79cd reward shaping \u8fd9\u79cd\u65b9\u5f0f\u5bf9\u4e8e\u589e\u5f3a\u5b66\u4e60\u8bad\u7ec3\u7684\u597d\u5904\uff1a\u7a33\u5b9a\u6027\u63d0\u9ad8\u3001\u6536\u655b\u53d8\u5feb\u7b49\u7b49\u3002ICM[1] \u7684\u5de5\u4f5c\u6bd4\u8f83\u60ca\u8273\u7684\u5730\u65b9\u5728\u4e8e\uff0c\u4ed6\u4eec\u5f7b\u5e95\u5730\u5c06\u5916\u90e8\u6fc0\u52b1\u632a\u53bb\u540e\uff0c\u4f9d\u7136\u53ef\u4ee5\u53ea\u901a\u8fc7\u5185\u90e8\u6fc0\u52b1\u4fc3\u4f7f agent \u5b66\u51fa\u5f88\u597d\u7684\u7b56\u7565\uff0c\u5b8c\u6210\u6e38\u620f\u3002\u800c ICLR[2] \u7684\u65b9\u6cd5\u5219\u76f4\u63a5\u593a\u5f97\u4e86\u53bb\u5e74 ViZDoom 2016 \u7684\u51a0\u519b\uff0c\u4ece\u89c6\u9891\u4e2d\u53ef\u4ee5\u770b\u51fa\u5b83\u7684\u7b56\u7565\u4f1a\u66f4\u52a0\u7ec6\u81f4\uff0c\u4e0d\u4f1a\u201c\u6a2a\u51b2\u76f4\u649e\u201d\uff0c\u4f1a\u6355\u6349\u5f88\u7ec6\u5fae\u5730\u79fb\u52a8\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u6536\u96c6\u7269\u54c1\u548c\u5c04\u51fb\u654c\u4eba\u7684\u51c6\u786e\u5ea6\uff08\u5b8c\u6574\u89c6\u9891\u8bf7\u770b\u8bba\u6587\u63d0\u4f9b\u7684\u5730\u5740\uff0c\u8fd9\u91cc\u53ea\u622a\u53d6\u4e86\u5341\u79d2\uff09\u3002\u800c SR[3] \u7684\u5de5\u4f5c\u8fd8\u6709\u4e00\u4e9b\u5f88\u6709\u8da3\u7684\u53ef\u89c6\u5316\u3002SR \u53ef\u4ee5\u7528\u4e8e\u63d0\u53d6\u5177\u4f53\u7684 subgoal\uff0c\u4ece\u800c\u5c55\u73b0\u51fa\u5b66\u4e60\u6548\u679c\uff1a<\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma10-1534232221.jpg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px;\"><span style=\"font-size: 13px;\">\u5176\u5b9e\u5982\u679c\u628a\u4e09\u7bc7\u5de5\u4f5c\u90fd\u770b\u505a subgoal \u6216\u8005 reward shaping \u7684\u5de5\u4f5c\uff0c\u5219 hierarchical RL \u7684\u76f8\u5173\u5de5\u4f5c\u4e5f\u90fd\u662f\u8fd9\u4e9b\u5de5\u4f5c\u7684\u201c\u8fdc\u4eb2\u201d\uff08\u4e0a\u6b21\u4e5f\u6709\u63d0\u5230\u4e00\u4e9b\u603b\u7ed3\uff0c\u70b9\u51fb\u9605\u8bfb\u539f\u6587\uff09\u3002<\/span><\/p>\n<section  class=\"mpa-template\" >\n<p style=\"max-width: 100%; min-height: 1em; color: rgb(62, 62, 62); font-size: 16px; white-space: normal; text-align: center; line-height: 3em; background-color: rgb(255, 255, 255); box-sizing: border-box !important; word-wrap: break-word !important;\"><img class=\"__bg_gif\"     style=\"margin: 0px; padding: 0px; max-width: 100%; height: auto !important; box-sizing: border-box !important; word-wrap: break-word !important; visibility: visible !important; width: 64px !important;\" width=\"100%\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma6-1534232221.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><br  \/><\/p>\n<\/section>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: center;\"><span style=\"color: rgb(123, 12, 0);\"><strong>\u63a8\u8350\u8bfe\u7a0b<\/strong><\/span><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\"><\/span><\/strong><\/p>\n<hr  \/>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\">\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u7b80\u4ecb\u2014\u2014\u53f2\u96ea\u677e<\/span><\/strong><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\">http:\/\/campus.swarma.org\/gcou=9?src=2<span style=\"color: rgb(123, 12, 0);\"><\/span><\/span><\/strong><\/p>\n<div class=\"post-image\"><img      style=\"text-align: center; white-space: normal; width: 85%; height: auto !important;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma6-1534232222.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/div>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\"><\/span><\/strong><\/p>\n<hr  \/>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\">AI\u6253\u6e38\u620f\u80cc\u540e\u7684\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u6280\u672f\u2014\u2014\u5f20\u6c5f<\/span><\/strong><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\">http:\/\/campus.swarma.org\/gcou=10282?src=2<\/span><\/strong><\/p>\n<div class=\"post-image\"><img    style=\"width: 90%; height: auto !important;\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma5-1534232222.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/div>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><span style=\"font-size: 13px;\"><\/span><\/p>\n<hr  \/>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><span style=\"font-size: 13px;\"><strong>\u5f3a\u5316\u5b66\u4e60\u5165\u95e8\u7cfb\u5217\u2014\u2014\u83ab\u70e6<\/strong><\/span><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><span style=\"font-size: 13px;\"><strong>http:\/\/campus.swarma.org\/gpac=16?src=2<\/strong><\/span><\/p>\n<div class=\"post-image\"><img      style=\"width: 90%; height: auto !important;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma5-1534232222-1.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/div>\n<p><br  \/><\/p>\n<section  class=\"mpa-template\" >\n<p style=\"max-width: 100%; min-height: 1em; color: rgb(62, 62, 62); font-size: 16px; white-space: normal; text-align: center; line-height: 3em; background-color: rgb(255, 255, 255); box-sizing: border-box !important; word-wrap: break-word !important;\"><img class=\"__bg_gif\"     style=\"margin: 0px; padding: 0px; max-width: 100%; height: auto !important; box-sizing: border-box !important; word-wrap: break-word !important; visibility: visible !important; width: 64px !important;\" width=\"100%\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma3-1534232222.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><br  \/><\/p>\n<\/section>\n<p style=\"margin: 10px 0.5em; letter-spacing: 0.5px; text-align: center; line-height: 2em;\"><span style=\"color: rgb(123, 12, 0);\"><strong>\u63a8\u8350\u9605\u8bfb<\/strong><\/span><\/p>\n<p style=\"margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485769&amp;idx=1&amp;sn=17928ec512da74b250e356ccba25a750&amp;chksm=e89453c4dfe3dad2a9bc7d73f9957379b15a61ef77c6a4d8b5d680b88cef22f51329ffb13a31&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485769&amp;idx=1&amp;sn=17928ec512da74b250e356ccba25a750&amp;chksm=e89453c4dfe3dad2a9bc7d73f9957379b15a61ef77c6a4d8b5d680b88cef22f51329ffb13a31&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">\u67ef\u6d01\u8f93\u4e86\uff0c\u6240\u4ee5\u6765\u5b66\u5b66AlphaGo\u80cc\u540e\u7684\u4eba\u5de5\u667a\u80fd\u5427&nbsp;<\/a><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485769&amp;idx=4&amp;sn=8eb327f42836ab3ebbf14e5235eac449&amp;chksm=e89453c4dfe3dad2e3ad7c7662ee658337f8638dea2d1487a98e69fb85b3254ddb84128f1213&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485769&amp;idx=4&amp;sn=8eb327f42836ab3ebbf14e5235eac449&amp;chksm=e89453c4dfe3dad2e3ad7c7662ee658337f8638dea2d1487a98e69fb85b3254ddb84128f1213&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">\u5de6\u53f3\u4e92\u640f\uff0c\u9752\u51fa\u4e8e\u84dd\u800c\u80dc\u4e8e\u84dd\uff1f<\/a><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485740&amp;idx=2&amp;sn=8210fa9a0275b40ebf490dd43e3f3ef9&amp;chksm=e89453a1dfe3dab718c59f14bf96816db86fd4366e59b640c846bb3efabe5748502e1fb3689f&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485740&amp;idx=2&amp;sn=8210fa9a0275b40ebf490dd43e3f3ef9&amp;chksm=e89453a1dfe3dab718c59f14bf96816db86fd4366e59b640c846bb3efabe5748502e1fb3689f&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">2017\u5317\u4eac\u5e08\u8303\u5927\u5b66\u590d\u6742\u7cfb\u7edf\u6691\u671f\u5b66\u6821\u62db\u751f\u542f\u4e8b<\/a><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485677&amp;idx=1&amp;sn=c87541996d9f734390b9c6b2ab9abde5&amp;chksm=e8945260dfe3db76d5839523786f649d4fe7222f7f2a3a35be5775e3dc8d032a4bffb1544eeb&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485677&amp;idx=1&amp;sn=c87541996d9f734390b9c6b2ab9abde5&amp;chksm=e8945260dfe3db76d5839523786f649d4fe7222f7f2a3a35be5775e3dc8d032a4bffb1544eeb&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">\u63ea\u7740\u81ea\u5df1\u7684\u5934\u53d1\u79bb\u5f00\u5730\u7403<\/a><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485723&amp;idx=1&amp;sn=ef60e683e1d9377f0ebc583d761a0066&amp;chksm=e8945396dfe3da808b84c9dcafe65c02920b77847ed3831066a376c01112de6777a476a3de2e&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485723&amp;idx=1&amp;sn=ef60e683e1d9377f0ebc583d761a0066&amp;chksm=e8945396dfe3da808b84c9dcafe65c02920b77847ed3831066a376c01112de6777a476a3de2e&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">\u900f\u89c6\u6df1\u5ea6\u5b66\u4e60\uff0c\u7545\u60f3\u672a\u6765\u5e94\u7528<\/a><br  \/><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><a href=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485679&amp;idx=1&amp;sn=364fec8b0a805ed03cc27851de03c3b4&amp;chksm=e8945262dfe3db749eb76c65b5de80b161e0dff1d0eb115ca32c357ef0fe1f6d7316d15b3047&amp;scene=21#wechat_redirect\" target=\"_blank\" data_ue_src=\"http:\/\/mp.weixin.qq.com\/s?__biz=MzIzMjQyNzQ5MA==&amp;mid=2247485679&amp;idx=1&amp;sn=364fec8b0a805ed03cc27851de03c3b4&amp;chksm=e8945262dfe3db749eb76c65b5de80b161e0dff1d0eb115ca32c357ef0fe1f6d7316d15b3047&amp;scene=21#wechat_redirect\" style=\"text-decoration: underline; font-size: 13px;\">\u96c6\u667a\u5fd7\u613f\u8005\u62db\u52df\u5566\uff01<\/a><\/p>\n<p style=\"margin-top: 10px; margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><strong><span style=\"font-size: 13px;\">\u8f6c\u8f7d\u58f0\u660e\uff1a\u672c\u6587\u8f6c\u8f7d\u81ea\u300c\u7a0b\u5e8f\u5a9b\u7684\u65e5\u5e38\u300d\uff0c\u641c\u7d22\u300cgirlswhocode\u300d\u5373\u53ef\u5173\u6ce8\u3002<\/span><\/strong><br  \/><\/p>\n<p style=\"margin: 10px 0.5em; line-height: 2em; letter-spacing: 0.5px; text-align: left;\"><img    class=\"\"   src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma0-1534232222.jpg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><br  \/><\/p>\n<hr style=\"margin-right: 0.5em; margin-left: 0.5em; white-space: normal; letter-spacing: 0.5px;\"  \/>\n<p style=\"margin-right: 0.5em; margin-left: 0.5em; white-space: normal; text-align: center; letter-spacing: 0.5px;\"><img width=\"100%\" class=\"__bg_gif\"     style=\"color: rgb(62, 62, 62); font-size: 16px; background-color: rgb(255, 255, 255); visibility: visible !important; width: 64px !important;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma8-1534232222.gif\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<section   style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 16px; white-space: normal; color: rgb(62, 62, 62); line-height: 25.6px; letter-spacing: 0.5px; background-color: rgb(255, 255, 255);\">\n<section   style=\"margin-right: 0.5em; margin-left: 0.5em; line-height: 25.6px; text-align: center; outline: none;\">\n<section   style=\"margin-right: 0.5em; margin-left: 0.5em; line-height: 25.6px; outline: none;\">\n<section   style=\"outline: none;\">\n<section  style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 15px; outline: none;\">\n<section  style=\"line-height: 25.6px; border-color: rgb(123, 12, 0);\">\n<p style=\"margin-top: 10px; margin-bottom: 10px; padding-right: 3px; padding-left: 3px; transform: translate3d(0px, 0px, 0px); border-color: rgb(123, 12, 0); line-height: 1.5em;\"><strong><span style=\"font-size: 12px; color: rgb(136, 136, 136);\">\u96c6\u667aQQ\u7fa4\uff5c292641157<br  \/>\u5546\u52a1\u5408\u4f5c\uff5czhangqian@swarma.org<br  \/>\u6295\u7a3f\u8f6c\u8f7d\uff5cwangting@swarma.org<\/span><\/strong><\/p>\n<section   style=\"margin-right: 0.5em; margin-left: 0.5em; outline: none;\">\n<h1 style=\"margin-top: 10px; margin-bottom: 10px; line-height: 1.75em;\"><strong style=\"font-size: 14px; white-space: pre-wrap; color: rgb(0, 112, 192); line-height: 25.6px;\"><strong style=\"line-height: 28px; white-space: normal; color: rgb(61, 170, 214); font-size: 20px;\"><span style=\"font-size: 14px; color: rgb(136, 136, 136);\"><span style=\"color: rgb(255, 76, 0);\">\u25c6&nbsp;<\/span><span style=\"color: rgb(0, 128, 255);\">\u25c6&nbsp;<\/span><span style=\"color: rgb(61, 170, 214);\">\u25c6<\/span><\/span><\/strong><\/strong><\/h1>\n<\/section>\n<p style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 19px; font-family: \u5fae\u8f6f\u96c5\u9ed1; color: rgb(71, 193, 168); line-height: 23.2727px; overflow-wrap: break-word !important;\"><span style=\"color: rgb(123, 12, 0);\"><strong><span style=\"font-size: 14px;\">\u641c\u7d22\u516c\u4f17\u53f7\uff1a\u96c6\u667a\u4ff1\u4e50\u90e8<\/span><\/strong><\/span><\/p>\n<p style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 19px; font-family: \u5fae\u8f6f\u96c5\u9ed1; color: rgb(71, 193, 168); line-height: 23.2727px; overflow-wrap: break-word !important;\"><span style=\"color: rgb(0, 0, 0);\"><strong><span style=\"font-size: 14px;\"><br  \/><\/span><\/strong><\/span><\/p>\n<p style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 19px; font-family: \u5fae\u8f6f\u96c5\u9ed1; color: rgb(71, 193, 168); line-height: 23.2727px; overflow-wrap: break-word !important;\"><span style=\"color: rgb(0, 0, 0);\"><strong><span style=\"font-size: 14px;\">\u52a0\u5165\u201c\u6ca1\u6709\u56f4\u5899\u7684\u7814\u7a76\u6240\u201d<\/span><\/strong><\/span><\/p>\n<section style=\"margin-right: 0.5em; margin-left: 0.5em; font-size: 14px; color: rgb(71, 193, 168); line-height: 20px; overflow-wrap: break-word !important;\">\n<p style=\"margin: 5px auto; padding: 10px; width: 180px; border-width: 2px; border-style: dashed; border-color: rgb(132, 132, 132); line-height: normal; overflow-wrap: break-word !important;\"><img width=\"auto\"      style=\"overflow-wrap: break-word !important; visibility: visible !important; width: auto !important;\" src=\"http:\/\/swarma.org\/wp-content\/uploads\/2018\/08\/swarma7-1534232222.jpeg\" alt=\"\u96c6\u667a\" title=\"\u96c6\u667a\"  \/><\/p>\n<\/section>\n<p><span style=\"font-size: 14px;\">\u8ba9\u82f9\u679c\u7838\u5f97\u66f4\u731b\u70c8\u4e9b\u5427\uff01<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<blockquote class='keep-source'>\n<p>\u59cb\u53d1\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7\uff1a                         \u96c6\u667a\u4ff1\u4e50\u90e8                      <\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u5bfc\u8bed \u662f\u7684\uff0c\u673a\u5668\u4e5f\u53ef\u4ee5\u6709\u597d\u5947\u5fc3\u3002\u8fd9\u4e0d\uff0c\u79d1\u5b66\u5bb6\u4eec\u6b63\u5728\u52aa\u529b\u8d4b\u4e88\u4eba\u5de5\u667a\u80fd\u7b97\u6cd5\u597d\u5947\u7684\u672c\u80fd\uff0c\u5e2e\u52a9\u5b83\u4eec\u66f4\u597d\u5730\u5b8c\u6210\u5f3a\u5316\u5b66\u4e60\u3002\u5728ICML2017\u7684\u8fd9\u7bc7\u8bba\u6587\u300aCuriosity-driven Exploration by Self-supervised Prediction\u300b\u4e2d\uff0c\u4f5c\u8005\u5c31\u7528\u597d\u5947\u5fc3\u4f5c\u4e3a\u4e00\u79cd\u5956\u52b1\u6fc0\u52b1\uff0c\u6765\u8ba9\u4eba&#8230;<\/p>\n","protected":false},"author":1,"featured_media":5016,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[19],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/5015"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5015"}],"version-history":[{"count":1,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/5015\/revisions"}],"predecessor-version":[{"id":5043,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/5015\/revisions\/5043"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/5016"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5015"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5015"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5015"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=5015"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}