{"id":35347,"date":"2022-06-22T20:17:08","date_gmt":"2022-06-22T12:17:08","guid":{"rendered":"https:\/\/swarma.org\/?p=35347"},"modified":"2022-06-22T20:17:08","modified_gmt":"2022-06-22T12:17:08","slug":"%e5%b4%94%e9%b9%8f%e5%9b%a2%e9%98%9f%ef%bc%9a%e4%b8%87%e5%ad%97%e9%95%bf%e6%96%87%e6%a2%b3%e7%90%86%e3%80%8c%e7%a8%b3%e5%ae%9a%e5%ad%a6%e4%b9%a0%e3%80%8d%e5%85%a8%e6%99%af%e5%9b%be","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=35347","title":{"rendered":"\u5d14\u9e4f\u56e2\u961f\uff1a\u4e07\u5b57\u957f\u6587\u68b3\u7406\u300c\u7a33\u5b9a\u5b66\u4e60\u300d\u5168\u666f\u56fe"},"content":{"rendered":"
\n
\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
\n

\u5c3d\u7ba1\u57fa\u4e8e\u76f8\u5173\u6027\u7684\u673a\u5668\u5b66\u4e60\u5728\u5f88\u591a\u9886\u57df\u90fd\u53d6\u5f97\u4e86\u6210\u529f\uff0c\u4f46\u662f\u865a\u5047\u76f8\u5173\u5e26\u6765\u7684\u6f5c\u5728\u98ce\u9669\u8fd8\u662f\u9650\u5236\u4e86\u8fd9\u4e9b\u6a21\u578b\u5728\u4e0d\u5c11\u98ce\u9669\u654f\u611f\u9886\u57df\u7684\u5e94\u7528\u3002\u7a33\u5b9a\u5b66\u4e60\u88ab\u63d0\u51fa\u6765\u5e94\u5bf9\u8fd9\u4e2a\u6311\u6218\uff0c\u5b83\u8bd5\u56fe\u5728\u4e0d\u5f71\u54cd\u6a21\u578b\u6027\u80fd\u7684\u60c5\u51b5\u4e0b\u6784\u5efa\u66f4\u52a0\u53ef\u4fe1\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u3002\u5728\u4eca\u5e74\u76842\u670823\u65e5\uff0c\u6e05\u534e\u5927\u5b66\u8ba1\u7b97\u673a\u7cfb\u5d14\u9e4f\u526f\u6559\u6388\u4e0e\u65af\u5766\u798f\u5927\u5b66Susan Athey\uff08\u7f8e\u56fd\u79d1\u5b66\u9662\u9662\u58eb\uff0c\u56e0\u679c\u9886\u57df\u56fd\u9645\u6743\u5a01\uff09\u5408\u4f5c\uff0c\u5728\u5168\u7403\u9876\u7ea7\u671f\u520aNature Machine Intelligence<\/em>\uff08\u5f71\u54cd\u56e0\u5b50IF=15.51\uff0c2020\uff09\u4e0a\u53d1\u8868\u4e86\u9898\u4e3a\u201cStable Learning Establishes Some Common Ground Between Causal Inference and Machine Learning\u201d\uff08\u7a33\u5b9a\u5b66\u4e60\uff1a\u5efa\u7acb\u56e0\u679c\u63a8\u65ad\u548c\u673a\u5668\u5b66\u4e60\u7684\u5171\u8bc6\uff09\u7684\u89c2\u70b9\u8bba\u6587\uff0c\u6df1\u5165\u63a2\u8ba8\u548c\u603b\u7ed3\u4e86\u56e0\u679c\u63a8\u7406\u5728\u673a\u5668\u5b66\u4e60\u548c\u4eba\u5de5\u667a\u80fd\u9886\u57df\u53d6\u5f97\u7684\u5173\u6ce8\uff0c\u63d0\u51fa\u673a\u5668\u5b66\u4e60\u548c\u56e0\u679c\u63a8\u7406\u4e4b\u95f4\u5e94\u8be5\u5f62\u6210\u5171\u8bc6\uff0c\u800c\u7a33\u5b9a\u5b66\u4e60\u6b63\u5728\u5411\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\u7684\u65b9\u5411\u8fc8\u8fdb\u3002\u8fd9\u7bc7\u63a8\u6587\u57fa\u4e8e\u4e0a\u8ff0\u5b66\u672f\u89c2\u70b9\uff0c\u7efc\u8ff0\u4e86\u7a33\u5b9a\u5b66\u4e60\u7684\u4e00\u7cfb\u5217\u8fdb\u5c55\uff0c\u8bd5\u56fe\u4e3a\u8bfb\u8005\u63d0\u4f9b\u4e00\u5e45\u5173\u4e8e\u8fd9\u4e00\u7814\u7a76\u65b9\u5411\u7684\u5168\u666f\u56fe\u3002<\/strong><\/p>\n


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

\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u96f6\u7280\u79d1\u6280\u63a8\u51fa\u56e0\u679c\u6d3e\u54c1\u724c\u5e76\u7b56\u5212\u7cfb\u5217\u6d3b\u52a8\u3002\u57287\u67082\u65e514:00-17:00\u56e0\u679c\u6d3e\u8bba\u575b\u7b2c\u4e00\u671f\u300c\u56e0\u679c\u9769\u547d\u2014\u2014\u4e0b\u4e00\u4ee3\u53ef\u4fe1AI<\/a>\u300d\u4e0a\uff0c\u6211\u4eec\u9080\u8bf7\u5230\u5d14\u9e4f\u8001\u5e08\u4ee5\u300c\u63a8\u8350\u7cfb\u7edf\u7684\u5206\u5e03\u5916\u6cdb\u5316\u300d\u505a\u62a5\u544a\uff0c\u8be6\u60c5\u89c1\u6587\u672b\uff0c\u6b22\u8fce\u5927\u5bb6\u62a5\u540d\u53c2\u4e0e~<\/strong><\/p>\n


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

\u7814\u7a76\u9886\u57df\uff1a\u673a\u5668\u5b66\u4e60\uff0c\u7a33\u5b9a\u5b66\u4e60<\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/p>\n<\/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

\u4e01\u5584\u4e00\u3001\u9648\u5929\u8c6a\u3001\u845b\u6625\u6c5f<\/span> | \u4f5c\u8005<\/strong><\/span><\/p>\n

\u9093\u4e00\u96ea<\/span> | \u7f16\u8f91<\/strong><\/span>
<\/p>\n

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


<\/p>\n

<\/span><\/strong><\/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><\/section>\n
\n
<\/span><\/strong><\/section>\n
\u5bfc\u8a00 <\/span><\/strong><\/section>\n
\u3007\u3001\u673a\u5668\u5b66\u4e60\u9762\u4e34\u7684\u6311\u6218<\/span><\/strong><\/section>\n
\u4e00\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u63d0\u51fa<\/span><\/strong><\/section>\n
\u4e8c\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u4e3b\u8981\u65b9\u6cd5<\/span><\/strong><\/section>\n
\u4e09\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u5e94\u7528<\/span><\/strong><\/section>\n
\u56db\u3001\u5c0f\u7ed3 <\/span><\/strong><\/section>\n
<\/span><\/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


<\/p>\n

\n
\n
\n
\n

<\/section>\n

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

\u3007\u3001\u673a\u5668\u5b66\u4e60\u9762\u4e34\u7684\u6311\u6218<\/span><\/strong><\/span><\/p>\n<\/section>\n

\n

<\/section>\n

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


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

\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5df2\u7ecf\u5728\u8bb8\u591a\u9762\u5411\u4e92\u8054\u7f51\u7684\u573a\u666f\u53d6\u5f97\u6210\u529f\u3002\u5728\u8bf8\u5982\u9884\u6d4b\u70b9\u51fb\u91cf\u6216\u5bf9\u56fe\u50cf\u8fdb\u884c\u5206\u7c7b\u7b49\u5e94\u7528\u573a\u666f\u4e2d\uff0c\u6a21\u578b\u505a\u51fa\u9519\u8bef\u51b3\u7b56\u7684\u4ee3\u4ef7\u4f3c\u4e4e\u5e76\u4e0d\u9ad8\uff0c\u56e0\u6b64\u4ece\u4e1a\u8005\u91c7\u7528\u201c\u6027\u80fd\u9a71\u52a8\u201d\u7684\u6a21\u5f0f\u4f18\u5316\u4eba\u5de5\u667a\u80fd\u6280\u672f\uff0c\u5373\u53ea\u5173\u6ce8\u8be5\u6a21\u578b\u5728\u5b8c\u6210\u76ee\u6807\u4efb\u52a1\u65f6\u4f53\u73b0\u51fa\u7684\u6027\u80fd\u800c\u4e0d\u592a\u5173\u6ce8\u6280\u672f\u53d1\u751f\u9519\u8bef\u65f6\u7684\u98ce\u9669\u3002\u5f53\u4efb\u52a1\u73af\u5883\u53d1\u751f\u53d8\u5316\uff0c\u9884\u6d4b\u51fa\u73b0\u9519\u8bef\u7684\u65f6\u5019\uff0c\u4eba\u4eec\u901a\u8fc7\u9891\u7e41\u5730\u66f4\u65b0\u9ed1\u76d2\u6a21\u578b\u4ee5\u4fdd\u8bc1\u9884\u6d4b\u7684\u6027\u80fd\u3002\u7136\u800c\uff0c\u5728\u8bf8\u5982\u533b\u7597\u4fdd\u5065\u3001\u5de5\u4e1a\u5236\u9020\u3001\u91d1\u878d\u548c\u53f8\u6cd5\u7b49\u4e0e\u793e\u4f1a\u751f\u6d3b\u606f\u606f\u76f8\u5173\u7684\u9886\u57df\uff0c\u673a\u5668\u5b66\u4e60\u6a21\u578b\u505a\u51fa\u7684\u9519\u8bef\u9884\u6d4b\u7684\u540e\u679c\u5f80\u5f80\u662f\u96be\u4ee5\u63a5\u53d7\u7684\uff0c\u8fd9\u4e9b\u573a\u666f\u4e5f\u56e0\u6b64\u88ab\u79f0\u4e3a\u98ce\u9669\u654f\u611f\u7684\u573a\u666f\u3002\u7531\u4e8e\u6570\u636e\u83b7\u53d6\u56f0\u96be\u4ee5\u53ca\u4f26\u7406\u95ee\u9898\uff0c\u5728\u98ce\u9669\u654f\u611f\u7684\u573a\u666f\u4e2d\u56e0\u4e3a\u73af\u5883\u53d8\u5316\u800c\u91cd\u65b0\u8bad\u7ec3\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u4ee3\u4ef7\u4f1a\u6bd4\u8f83\u6602\u8d35\uff0c\u56e0\u6b64\u6a21\u578b\u7684\u77ed\u671f\u9884\u6d4b\u6027\u80fd\u4e4b\u5916\u7684\u7279\u6027\u4e5f\u5341\u5206\u91cd\u8981\u3002\u4e3a\u4e86\u4fc3\u8fdb\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5728\u66f4\u591a\u98ce\u9669\u654f\u611f\u573a\u666f\u7684\u5e94\u7528\uff0c\u6211\u4eec\u9700\u8981\u4ed4\u7ec6\u5206\u6790\u673a\u5668\u5b66\u4e60\u6a21\u578b\u9762\u4e34\u7684\u6280\u672f\u6027\u98ce\u9669\uff0c\u5e76\u91c7\u53d6\u529e\u6cd5\u514b\u670d\u8fd9\u4e9b\u98ce\u9669\u3002<\/span>
<\/p>\n


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

\u6e05\u534e\u5927\u5b66\u5d14\u9e4f\u526f\u6559\u6388\u548c\u65af\u5766\u798f\u5927\u5b66\u7684Susan Athey\u5728\u4e00\u7bc7\u4eca\u5e74\u53d1\u8868\u4e8eNature Machine Intelligence<\/em>\u7684\u89c2\u70b9\u8bba\u6587\u4e2d\u6df1\u5165\u5206\u6790\u4e86\u673a\u5668\u5b66\u4e60\u53d1\u98ce\u9669\u6765\u6e90\uff0c\u4ed6\u4eec\u63d0\u51fa\u7f3a\u4e4f\u7a33\u5b9a\u6027\u3001\u53ef\u89e3\u91ca\u6027\u548c\u516c\u5e73\u6027\u662f\u5f53\u4eca\u673a\u5668\u5b66\u4e60\u4e2d\u5fc5\u987b\u89e3\u51b3\u7684\u6700\u5173\u952e\u548c\u6700\u7d27\u8feb\u7684\u4e09\u4e2a\u98ce\u9669\u56e0\u7d20\uff0c\u5e76\u6307\u51fa\u53ef\u4ee5\u5229\u7528\u56e0\u679c\u5173\u7cfb\u7684\u89c2\u70b9\u6765\u6539\u5584\u673a\u5668\u5b66\u4e60\u548c\u9884\u6d4b\u5efa\u6a21\u3002<\/span><\/p>\n


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

\u7f3a\u4e4f\u7a33\u5b9a\u6027\u6307\u7684\u662f\u6a21\u578b\u5728\u672a\u77e5\u73af\u5883\u4e0b\u9884\u6d4b\u6027\u80fd\u4e0d\u7a33\u5b9a\u3002\u7531\u4e8e\u76ee\u524d\u5927\u591a\u6570\u673a\u5668\u5b66\u4e60\u6a21\u578b\u4f9d\u8d56\u4e8eI.I.D.\u5047\u8bbe<\/span>\uff08Independent and identically distributed\uff09<\/span>\uff0c\u5373\u8bad\u7ec3\u6570\u636e\u96c6\u548c\u6d4b\u8bd5\u6570\u636e\u96c6\u7684\u6570\u636e\u5206\u5e03\u662f\u76f8\u4f3c\u7684\u3002\u5728\u5b9e\u8df5\u4e2d\uff0c\u6a21\u578b\u7684\u5e94\u7528\u73af\u5883\u5e38\u5e38\u662f\u672a\u77e5\u7684\uff0c\u5373\u53ef\u80fd\u51fa\u73b0\u7684<\/span>\uff08\u6d4b\u8bd5\u6570\u636e\u96c6\uff09<\/span>\u5206\u5e03\u662f\u65e0\u6cd5\u5b8c\u5168\u9884\u89c1\u7684\uff0cI.I.D.\u5047\u8bbe\u5728\u8fd9\u65f6\u4f1a\u88ab\u7834\u574f\uff0c\u6a21\u578b\u7684\u6027\u80fd\u65e0\u6cd5\u4fdd\u8bc1\u3002\u7f3a\u4e4f\u53ef\u89e3\u91ca\u6027\u662f\u7531\u4e8e\u76ee\u524d\u5927\u591a\u6570\u73b0\u6210\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u90fd\u662f\u9ed1\u76d2\u6a21\u578b\uff1a\u7b97\u6cd5\u8fc7\u7a0b\u548c\u9884\u6d4b\u7ed3\u679c\u90fd\u4e0d\u5bb9\u6613\u5411\u4eba\u7c7b\u89e3\u91ca\u3002\u5728\u98ce\u9669\u654f\u611f\u9886\u57df\uff0c\u7531\u4e8e\u9519\u8bef\u51b3\u7b56\u4f1a\u5bfc\u81f4\u4e25\u91cd\u7684\u540e\u679c\uff0c\u5f80\u5f80\u9700\u8981\u8ba9\u4eba\u7c7b\u7559\u5728\u51b3\u7b56\u5faa\u73af\u4e2d\u6210\u4e3a\u51b3\u7b56\u7684\u6700\u7ec8\u7684\u770b\u95e8\u4eba\u3002\u6b64\u65f6\uff0c\u901a\u7528\u7684\u8bed\u8a00\u6765\u8ba9\u7b97\u6cd5\u548c\u4eba\u7c7b\u7406\u89e3\u548c\u5408\u4f5c\u662f\u5fc5\u8981\u7684\u3002\u6700\u540e\uff0c\u4e3b\u6d41\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u53ef\u80fd\u4f1a\u653e\u5927\u6570\u636e\u4e2d\u5b58\u5728\u7684\u504f\u5dee\uff0c\u4ece\u800c\u5bfc\u81f4\u4e0d\u516c\u5e73\u7684\u7ed3\u679c\uff0c\u56e0\u6b64\u662f\u7f3a\u4e4f\u516c\u5e73\u6027\u7684\u3002<\/span><\/p>\n


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

\u56e0\u4e3a\u5e38\u89c1\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u662f\u57fa\u4e8e\u8f93\u5165\u53d8\u91cf\u548c\u8f93\u51fa\u53d8\u91cf\u7684\u76f8\u5173\u6027\u505a\u9884\u6d4b\uff0c\u5f53\u5b83\u9047\u5230\u53d8\u91cf\u4e4b\u95f4\u7684\u865a\u5047\u76f8\u5173\u65f6\uff0c\u4fbf\u65e0\u6cd5\u907f\u514d\u4e0a\u8ff0\u4e09\u4e2a\u98ce\u9669\u56e0\u7d20\u3002\u6570\u636e\u4e2d\u7684\u76f8\u5173\u6027\u80fd\u5f52\u7c7b\u6210\u4e09<\/span>\u79cd\u6a21\u5f0f\uff1a\u56e0\u679c\u5bfc\u81f4\u7684\u76f8\u5173\uff0c\u6df7\u6dc6\u5bfc\u81f4\u7684\u76f8\u5173\u548c\u9009\u62e9\u6027\u504f\u5dee\u5bfc\u81f4\u7684\u76f8\u5173\uff0c\u5982\u56fe1\u6240\u793a\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe1\uff1a\u76f8\u5173\u6027\u7684\u4e09\u79cd\u6a21\u5f0f\uff0c\uff08a\uff09\u56e0\u679c\u76f8\u5173\uff1b\uff08b\uff09\u6df7\u6dc6\u5bfc\u81f4\u7684\u76f8\u5173\uff1b\uff08c\uff09\u6570\u636e\u9009\u62e9\u6027\u504f\u5dee\u5bfc\u81f4\u7684\u76f8\u5173\u3002<\/span><\/p>\n


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

\u5728\u8fd9\u4e09\u79cd\u4ea7\u751f\u76f8\u5173\u6027\u7684\u6a21\u5f0f\u4e2d\uff0c\u53ea\u6709\u56e0\u679c\u5173\u7cfb\u4ea7\u751f\u7684\u76f8\u5173\u6027\u53cd\u6620\u4e86\u53d8\u91cf\u4e4b\u95f4\u7684\u5185\u5728\u4f9d\u8d56\u5173\u7cfb\uff1b\u53e6\u5916\u4e24\u79cd\u7c7b\u578b\u662f\u5bf9\u7279\u5f81\u7684\u8054\u5408\u5206\u5e03\u548c\u6570\u636e\u6536\u96c6\u8fc7\u7a0b\u654f\u611f\u7684\u865a\u5047\u76f8\u5173\u3002\u7531\u4e8e\u9ed1\u76d2\u6a21\u578b\u751a\u81f3\u6ca1\u6709\u5c1d\u8bd5\u533a\u5206\u4ea7\u751f\u8fd9\u4e9b\u76f8\u5173\u6027\u7684\u4e09\u79cd\u4e0d\u540c\u6a21\u5f0f\uff0c\u5b83\u4eec\u7684\u9884\u6d4b\u6027\u80fd\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u53d6\u51b3\u4e8e\u6d4b\u8bd5\u5206\u5e03\u4e0e\u8bad\u7ec3\u5206\u5e03\u7684\u504f\u79bb\u7a0b\u5ea6\uff0c\u4ece\u800c\u5bfc\u81f4\u5728\u4e0d\u540c\u7684\u6d4b\u8bd5\u5206\u5e03\u4e0b\u6027\u80fd\u4e0d\u7a33\u5b9a\u3002\u540c\u65f6\uff0c\u57fa\u4e8e\u865a\u5047\u76f8\u5173\u6027\u7684\u9884\u6d4b\u6a21\u578b\u4e5f\u53ef\u80fd\u662f\u4e0d\u516c\u5e73\u7684\uff0c\u56e0\u4e3a\u5b83\u8fd8\u653e\u5927\u4e86\u9009\u62e9\u6027\u504f\u5dee\u3002<\/span><\/p>\n


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

\u5d14\u9e4f\u8001\u5e08\u5728\u8fd9\u7bc7\u53d1\u8868\u5728Nature Machine Intelligence<\/em>\u7684\u89c2\u70b9\u8bba\u6587\u4e2d\u63d0\u51fa\uff0c\u4e3a\u4e86\u4ece\u6839\u672c\u4e0a\u89e3\u51b3\u7a33\u5b9a\u6027\u3001\u53ef\u89e3\u91ca\u6027\u548c\u516c\u5e73\u6027\u7684\u98ce\u9669\uff0c\u4eba\u4eec\u9700\u8981\u5728\u673a\u5668\u5b66\u4e60\u6846\u67b6\u4e2d\u63a5\u53d7\u5e76\u5f3a\u8c03\u56e0\u679c\u5173\u7cfb\u3002\u56e0\u679c\u5173\u7cfb\u5b9e\u9645\u4e0a\u523b\u753b\u4e86\u6570\u636e\u751f\u6210\u673a\u5236\uff0c\u5982\u679c\u80fd\u591f\u8bc6\u522b\u548c\u4f30\u8ba1\u6570\u636e\u7684\u56e0\u679c\u7ed3\u6784\uff0c\u9884\u6d4b\u5efa\u6a21\u81ea\u7136\u53ef\u4ee5\u4f5c\u4e3a\u526f\u4ea7\u54c1\u6765\u89e3\u51b3\u3002\u9075\u5faa\u7740\u5f3a\u8c03\u56e0\u679c\u5173\u7cfb\u7684\u6280\u672f\u8def\u5f84\uff0c\u7814\u7a76\u8005\u5fc5\u987b\u89e3\u51b3\u56e0\u679c\u63a8\u65ad\u7684\u6240\u6709\u6311\u6218\u3002\u81f3\u6b64\uff0c\u673a\u5668\u5b66\u4e60\u548c\u56e0\u679c\u63a8\u65ad\u4e4b\u95f4\u5f62\u6210\u4e86\u5171\u8bc6\u3002\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u548c\u56fd\u5185\u5916\u5176\u4ed6\u56e2\u961f\u5728\u8fd9\u4e2a\u95ee\u9898\u4e0a\u8fdb\u884c\u4e86\u8bf8\u591a\u63a2\u7d22\uff0c\u5e76\u5f62\u6210\u4e86\u201c\u7a33\u5b9a\u5b66\u4e60\u201d\u7684\u65b9\u5411\uff0c\u671d\u7740\u6c9f\u901a\u673a\u5668\u5b66\u4e60\u548c\u56e0\u679c\u63a8\u65ad\u7684\u76ee\u6807\u524d\u8fdb\u3002<\/span><\/p>\n


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

\u5728\u672c\u7bc7\u63a8\u9001\u4e2d\uff0c\u6211\u4eec\u9996\u5148\u7ed9\u51fa\u4e86\u7a33\u5b9a\u5b66\u4e60\u4efb\u52a1\u7684\u52a8\u673a\u548c\u5b9a\u4e49\uff0c\u5e76\u603b\u7ed3\u4e86\u7a33\u5b9a\u5b66\u4e60\u7684\u4e3b\u8981\u65b9\u6cd5\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u5305\u62ec\u57fa\u4e8e\u6837\u672c\u52a0\u6743\u7684\u53d8\u91cf\u53bb\u76f8\u5173\u3001\u5bf9\u6297\u7a33\u5b9a\u5b66\u4e60\u3001\u5f02\u8d28\u6027\u98ce\u9669\u6700\u5c0f\u5316\u548c\u7a33\u5b9a\u5b66\u4e60\u7684\u7406\u8bba\u89e3\u91ca\u7b49\u3002\u968f\u540e\uff0c\u6211\u4eec\u5c55\u793a\u4e86\u7a33\u5b9a\u5b66\u4e60\u5728\u56fe\u5b66\u4e60\uff0c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u5e94\u7528\uff0c\u5e76\u8ba8\u8bba\u4e86\u5b83\u662f\u5982\u4f55\u4fc3\u8fdb\u7b97\u6cd5\u7684\u516c\u5e73\u6027\uff0c\u548c\u5904\u7406\u9886\u57df\u81ea\u9002\u5e94\u95ee\u9898\u7684\u3002<\/span><\/p>\n


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

\n
\n


<\/p>\n

\n
\n

<\/section>\n

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

\u4e00\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u63d0\u51fa<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


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

\u7a33\u5b9a\u5b66\u4e60\u7684\u6982\u5ff5\u662f\u57282018\u5e74\u7684KDD\u4f1a\u8bae\u4e0a\u7531\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u7684\u3002\u5b83\u65e8\u5728\u89e3\u51b3\u8fd9\u6837\u4e00\u4e2a\u95ee\u9898\uff0c\u5f53\u6211\u4eec\u7528\u673a\u5668\u5b66\u4e60\u8fdb\u884c\u9884\u6d4b\u5efa\u6a21\u7684\u65f6\u5019\uff0c\u5982\u679c\u5bf9\u6d4b\u8bd5\u6570\u636e\u96c6\u6ca1\u6709\u4efb\u4f55\u5148\u9a8c\u77e5\u8bc6\uff0c\u5982\u4f55\u4fdd\u8bc1\u6a21\u578b\u5728\u672a\u77e5\u5206\u5e03\u4e0a\u505a\u51fa\u7a33\u5b9a\u9884\u6d4b\uff0c\u5373\u5728\u4fdd\u8bc1\u9884\u6d4b\u6027\u80fd\u7684\u540c\u65f6\uff0c\u5c3d\u91cf\u51cf\u5c0f\u5b83\u5728\u672a\u77e5\u6d4b\u8bd5\u96c6\u7684\u65b9\u5dee\u3002\u56fe2\u5c55\u793a\u4e86\u7a33\u5b9a\u5b66\u4e60\u548c\u4f20\u7edf\u673a\u5668\u5b66\u4e60<\/span>\uff08\u72ec\u7acb\u540c\u5206\u5e03\u5b66\u4e60, I.I.D.\uff09<\/span>\u4e0e\u8fc1\u79fb\u5b66\u4e60\u7684\u5173\u7cfb\uff0c\u540e\u4e24\u8005\u7684\u4f18\u5316\u76ee\u6807\u662f\u5728\u5177\u6709\u6d4b\u8bd5\u96c6\u5206\u5e03\u5148\u9a8c\u77e5\u8bc6\u7684\u60c5\u51b5\u4e0b\uff0c\u6700\u5927\u5316\u6a21\u578b\u7684\u9884\u6d4b\u6027\u80fd<\/span>\uff08\u51c6\u786e\u7387\uff09<\/span>\u3002\u540c\u65f6\uff0c\u7a33\u5b9a\u5b66\u4e60\u5047\u8bbe\u5728\u6d4b\u8bd5\u6570\u636e\u4e2d\u5b58\u5728\u591a\u4e2a\u73af\u5883\uff0c\u56e0\u6b64\u5bf9\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u63d0\u51fa\u4e86\u66f4\u9ad8\u7684\u8981\u6c42\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe2\uff1a\u7a33\u5b9a\u5b66\u4e60\uff0c\u4f20\u7edf\u673a\u5668\u5b66\u4e60\uff08I.I.D.\u5b66\u4e60\uff09\u548c\u8fc1\u79fb\u5b66\u4e60\u7684\u6bd4\u8f83<\/span><\/p>\n


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

\u57282018\u5e74\u53d1\u8868\u5728KDD\u4f1a\u8bae\u7684\u8bba\u6587\u4e2d\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u53d7\u5230\u56e0\u679c\u63a8\u65ad\u4e2d\u5229\u7528\u534f\u53d8\u91cf\u5e73\u8861\u4f30\u8ba1\u56e0\u679c\u6548\u5e94\u7684\u601d\u60f3\u7684\u542f\u53d1\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u6df1\u5ea6\u5168\u5c40\u5e73\u8861\u56de\u5f52<\/span>\uff08Deep Global Balancing Regression, DGBR\uff09<\/span>\u7b97\u6cd5\uff0c\u8054\u5408\u4f18\u5316\u4e00\u4e2a\u7528\u4e8e\u7279\u5f81\u9009\u62e9<\/span>\uff08\u964d\u7ef4\uff09<\/span>\u7684\u6df1\u5ea6\u81ea\u52a8\u7f16\u7801\u5668\u6a21\u578b\uff0c\u548c\u4e00\u4e2a\u7528\u4e8e\u672a\u77e5\u73af\u5883\u4e0b\u7a33\u5b9a\u9884\u6d4b\u7684\u5168\u5c40\u5e73\u8861\u6a21\u578b\u3002\u8fd9\u4e2a\u5168\u5c40\u5e73\u8861\u6a21\u578b\u8bc4\u4f30\u4e86\u4efb\u610f\u4e00\u4e2a\u7279\u5f81\u5728\u88ab\u9694\u79bb<\/span>\uff08isolated\uff09<\/span>\u7684\u60c5\u51b5\u4e0b\uff0c\u8be5\u4e2a\u7279\u5f81\u5bf9\u54cd\u5e94\u7684\u56e0\u679c\u6548\u5e94\uff0c\u5e76\u4ee5\u6b64\u6784\u5efa\u4e86\u4e00\u7ec4\u4f7f\u5f97\u53d8\u91cf\u88ab\u5e73\u8861<\/span>\uff08variable balancing\uff09<\/span>\u7684\u6837\u672c\u6743\u91cd\uff0c\u8ba9\u52a0\u6743\u540e\u7684\u6837\u672c\u7684\u6240\u6709\u7279\u5f81\u90fd\u5c3d\u53ef\u80fd\u76f8\u4e92\u72ec\u7acb\u3002\u8fd9\u4e2a\u7b97\u6cd5\u5728\u9ad8\u7ef4\u5ea6\u7684\u6570\u636e\u4e2d\u4e5f\u9002\u7528\u3002<\/span><\/p>\n


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

\n
\n


<\/p>\n

\n
\n

<\/section>\n

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

\u4e8c\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u4e3b\u8981\u65b9\u6cd5<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


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

DGBR\u7b97\u6cd5\u9996\u6b21\u89e3\u51b3\u4e86\u4e8c\u5143\u9884\u6d4b\u53d8\u91cf<\/span>\uff08\u7279\u5f81\uff09<\/span>\u548c\u4e8c\u5143\u79bb\u6563\u54cd\u5e94\u53d8\u91cf\u8bbe\u5b9a\u4e0b\u7684\u7a33\u5b9a\u9884\u6d4b\u95ee\u9898\u3002<\/span>\u6b64\u540e\uff0c\u4e00\u7cfb\u5217\u7a33\u5b9a\u5b66\u4e60\u65b9\u6cd5\u6765\u88ab\u63d0\u51fa\u6765\uff0c\u7528\u4e8e\u89e3\u51b3\u66f4\u591a\u4e0d\u540c\u8bbe\u5b9a\u4e0b\u7684\u7a33\u5b9a\u9884\u6d4b\u95ee\u9898\u3002\u800c\u540e\u7eed\u8bbe\u8ba1\u7684\u7a33\u5b9a\u5b66\u4e60\u65b9\u6cd5\u4e0d\u4ec5\u4ec5\u518d\u5c40\u9650\u4e8e\u56e0\u679c\u63a8\u65ad\u7684\u89d2\u5ea6\uff0c\u8fd8\u5305\u62ec\u4e86\u7edf\u8ba1\u5b66\u4e60\u3001\u4f18\u5316\u8fc7\u7a0b\u7b49\u4e0d\u540c\u89c6\u89d2\uff0c\u5b83\u4eec\u5c06\u5728\u672c\u8282\u88ab\u9010\u4e00\u4ecb\u7ecd<\/span>\u3002<\/span><\/p>\n


<\/p>\n

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

\n
\n
\n

2.1 \u57fa\u4e8e\u6837\u672c\u52a0\u6743\u7684\u53d8\u91cf\u53bb\u76f8\u5173<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u9488\u5bf9\u5177\u6709\u6a21\u578b\u9519\u4f30<\/span>\uff08model misspecification\uff0c\u5373\u6a21\u578b\u4e0e\u6570\u636e\u751f\u6210\u673a\u5236\u4e0d\u4e00\u81f4\uff09<\/span>\u7a33\u5b9a\u9884\u6d4b\u95ee\u9898\u8fdb\u884c\u4e86\u8fdb\u4e00\u6b65\u63a2\u7d22\u3002Zheyan Shen\u7b49\u4eba\u7814\u7a76\u4e86\u5728\u7ebf\u6027\u6a21\u578b\u4e2d\uff0c\u53d8\u91cf\u95f4\u7684\u5171\u7ebf\u6027\u662f\u5982\u4f55\u5f71\u54cd\u9884\u6d4b\u7a33\u5b9a\u6027\u7684\uff0c\u5e76\u63d0\u51fa\u4e86\u4e00\u79cd\u4e00\u822c\u7684\u6570\u636e\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u901a\u8fc7\u8bad\u7ec3\u96c6\u6837\u672c\u91cd\u52a0\u6743\u6765\u53bb\u9664\u9884\u6d4b\u53d8\u91cf<\/span>\uff08\u7279\u5f81\uff09<\/span>\u95f4\u7684\u76f8\u5173\u6027\uff0c\u4ee5\u51cf\u5c0f\u5171\u7ebf\u6027\u6548\u5e94\u3002\u51b5\u7428\u7b49\u4eba\u7684\u5de5\u4f5c\u5219\u8fdb\u4e00\u6b65\u6539\u8fdb\u4e86DGBR\u7b97\u6cd5\uff0c\u63d0\u51fa\u4e86\u53bb\u76f8\u5173\u52a0\u6743\u56de\u5f52\u7b97\u6cd5<\/span>\uff08Decorrelated Weighting Regression, DWR\uff09<\/span>\uff0c\u7ed3\u5408\u4e86\u53d8\u91cf\u53bb\u76f8\u5173\u6b63\u5219\u5316\u4e0e\u52a0\u6743\u56de\u5f52\u6a21\u578b\uff0c\u89e3\u51b3\u4e86\u5728\u8fde\u7eed\u7684\u9884\u6d4b\u53d8\u91cf<\/span>\uff08\u7279\u5f81\uff09<\/span>\u7684\u8bbe\u5b9a\u4e0b\uff0c\u6a21\u578b\u7684\u7a33\u5b9a\u9884\u6d4b\u95ee\u9898\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe3\uff1aDWR\u6a21\u578b\u548c\u57fa\u7ebf\u6a21\u578b\u5728\u4e0d\u540c\u6d4b\u8bd5\u73af\u5883\u4e0b\u6027\u80fd\u7684\u6bd4\u8f83\uff1b\u5de6\u56fe\uff1a\u6a2a\u8f74\u4ee3\u8868\u4e0d\u540c\u7684\u73af\u5883\uff0c\u7eb5\u8f74\u4ee3\u8868\u9884\u6d4b\u8bef\u5dee\uff0c\u4e0d\u540c\u989c\u8272\u4ee3\u8868\u4e0d\u540c\u7684\u6a21\u578b\uff1b\u53f3\u56fe\uff1a\u6a2a\u8f74\u4ee3\u8868\u4e0d\u540c\u7684\u6a21\u578b\uff0c\u7eb5\u8f74\u4ee3\u8868\u5728\u4e0d\u540c\u73af\u5883\u4e0b\u7684\u5e73\u5747\u9519\u8bef\uff0c\u6027\u80fd\u7684\u65b9\u5dee\u5728\u6761\u67f1\u9876\u7aef\u7528\u9ed1\u5b9e\u7ebf\u6807\u51fa\u3002<\/span><\/p>\n


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

\u53bb\u9664\u6240\u6709\u53d8\u91cf\u4e4b\u95f4\u7684\u76f8\u5173\u6027\u56fa\u7136\u662f\u627e\u5230\u56e0\u679c\u76f8\u5173\uff0c\u5e73\u8861\u534f\u53d8\u91cf\u5e76\u5b9e\u73b0\u7a33\u5b9a\u9884\u6d4b\u7684\u597d\u4e3b\u610f\uff0c\u5b83\u7684\u4ee3\u4ef7\u662f\u5927\u5927\u964d\u4f4e\u4e86\u6709\u6548\u6837\u672c\u91cf<\/span>\uff08efficient sample size\uff09<\/span>\uff0c\u800c\u8fd9\u5728\u673a\u5668\u5b66\u4e60\u8bad\u7ec3\u4e2d\u662f\u707e\u96be\u6027\u7684\u3002Zheyan Shen\u7b49\u4eba\u901a\u8fc7\u4f7f\u7528\u6765\u81ea\u4e0d\u540c\u73af\u5883\u7684\u672a\u6807\u6ce8\u6570\u636e\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e\u53d8\u91cf\u805a\u7c7b\u7684\u53d8\u91cf\u5206\u89e3\u7684\u7b97\u6cd5\uff0c\u79f0\u4e3a\u533a\u5206\u6027\u53d8\u91cf\u53bb\u76f8\u5173<\/span>\uff08Differentiated Variable Decorrelation, DVD\uff09<\/span>\u3002\u8fd9\u4e2a\u65b9\u6cd5\u662f\u6ce8\u610f\u5230\u4fdd\u7559\u56e0\u679c\u6027\u53d8\u91cf\u4e4b\u95f4\u7684\u76f8\u5173\u6027\u672a\u5fc5\u4f1a\u5bfc\u81f4\u6a21\u578b\u5728\u672a\u77e5\u73af\u5883\u4e2d\u7684\u6027\u80fd\u4e0d\u7a33\u5b9a\u3002\u4ee5\u7279\u5f81\u95f4\u76f8\u5173\u6027\u7684\u5728\u8bad\u7ec3\u96c6\u6570\u636e\u548c\u672a\u6807\u6ce8\u4e4b\u95f4\u7684\u6570\u636e\u7a33\u5b9a\u6027\u4f5c\u4e3a\u805a\u7c7b\u7684\u6307\u6807\uff0c\u53ef\u4ee5\u5c06\u9884\u6d4b\u53d8\u91cf<\/span>\uff08\u7279\u5f81\uff09<\/span>\u8fdb\u884c\u805a\u7c7b\u5e76\u4e3a\u4e0d\u540c\u7684\u805a\u7c7b\u7c07\uff0c\u5176\u4e2d\u4e00\u4e9b\u805a\u7c7b\u7c07\u4ee3\u8868\u4e86\u5bf9\u76f8\u5e94\u53d8\u91cf\u5177\u6709\u56e0\u679c\u6548\u5e94\u7684\u7279\u5f81\u96c6\u5408\u3002\u5728\u5e73\u8861\u6df7\u6dc6\u53d8\u91cf\u65f6\u53ea\u9700\u8981\u9694\u79bb\u8fd9\u4e9b\u805a\u7c7b\u7c07\u5373\u53ef\u3002\u7531\u4e8e\u805a\u7c7b\u7c07\u7684\u6570\u91cf\u8fdc\u4f4e\u4e8e\u7279\u5f81\u7684\u7ef4\u5ea6\uff0c\u56e0\u6b64DVD\u4e0e\u6837\u672c\u52a0\u6743\u65b9\u6cd5DWR\u76f8\u6bd4\uff0c\u4fdd\u6301\u4e86\u66f4\u9ad8\u7684\u6709\u6548\u6837\u672c\u91cf\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe4\uff1a\u4e0e\u4e0d\u52a0\u533a\u5206\u5730\u53bb\u9664\u6240\u6709\u53d8\u91cf\u7684\u76f8\u5173\u6027\u7684\u65b9\u6cd5\uff08DWR\uff09\u76f8\u6bd4\uff0c\u533a\u5206\u6027\u53d8\u91cf\u53bb\u76f8\u5173\u65b9\u6cd5\uff08DVD\uff09\u5728\u76f8\u540c\u8bbe\u5b9a\u4e0b\u5177\u6709\u66f4\u5927\u7684\u6709\u6548\u6837\u672c\u91cf\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

2.2 \u5bf9\u6297\u7a33\u5b9a\u5b66\u4e60<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u7531\u4e8e\u4eba\u4eec\u603b\u662f\u60f3\u6700\u5927\u5316\u5730\u5229\u7528\u8bad\u7ec3\u6570\u636e\u4e2d\u53d1\u73b0\u7684\u6240\u6709\u76f8\u5173\u6027\uff0c\u5177\u6709\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u5728\u5206\u5e03\u53d8\u5316\u4e0b\u5f88\u5bb9\u6613\u53d7\u5230\u653b\u51fb\u3002\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u4e86\u7a33\u5b9a\u5bf9\u6297\u5b66\u4e60 <\/span>(Stable Adversarial Learning, SAL) <\/span>\u7b97\u6cd5\uff0c\u4ee5\u4e00\u79cd\u66f4\u6709\u539f\u5219\u548c\u7edf\u4e00\u7684\u65b9\u5f0f\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u8be5\u7b97\u6cd5\u5229\u7528\u5f02\u6784\u6570\u636e\u6e90\u6784\u5efa\u66f4\u5b9e\u7528\u7684\u4e0d\u786e\u5b9a\u6027\u96c6\u5e76\u8fdb\u884c\u5dee\u5f02\u5316\u9c81\u68d2\u6027\u4f18\u5316\uff0c\u5176\u4e2d\u534f\u53d8\u91cf\u6839\u636e\u5176\u4e0e\u76ee\u6807\u76f8\u5173\u6027\u7684\u7a33\u5b9a\u6027\u8fdb\u884c\u533a\u5206\u3002<\/span><\/p>\n


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

\u5177\u4f53\u6765\u8bf4\uff0c\u8be5\u65b9\u6cd5\u5c31\u662f\u91c7\u7528\u4e86Wasserstein\u5206\u5e03\u5f0f\u9c81\u68d2\u5b66\u4e60<\/span>\uff08Wasserstein distributionally robust learning, WDRL\uff09<\/span>\u7684\u6846\u67b6\u3002\u6839\u636e\u534f\u53d8\u91cf\u5728\u591a\u4e2a\u73af\u5883\u4e2d\u7684\u7a33\u5b9a\u6027\uff0c\u8fdb\u4e00\u6b65\u5c06\u4e0d\u786e\u5b9a\u6027\u96c6\u8868\u5f81\u4e3a\u5404\u5411\u5f02\u6027\uff0c\u8fd9\u4f1a\u7ed9\u4e0d\u7a33\u5b9a\u534f\u53d8\u91cf\u5e26\u6765\u6bd4\u7a33\u5b9a\u534f\u53d8\u91cf\u66f4\u5f3a\u7684\u5bf9\u6297\u6027\u6270\u52a8\u3002\u5e76\u4e14\u8bbe\u8ba1\u4e86\u4e00\u79cd\u534f\u540c\u7b97\u6cd5\u6765\u8054\u5408\u4f18\u5316\u534f\u53d8\u91cf\u7684\u5fae\u5206\u8fc7\u7a0b\u4ee5\u53ca\u6a21\u578b\u53c2\u6570\u7684\u5bf9\u6297\u8bad\u7ec3\u8fc7\u7a0b\u3002<\/span><\/p>\n


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

\u5728\u5b9e\u9a8c\u4e2d\uff0c\u5c06SAL\u7b97\u6cd5\u4e0e\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316<\/span> (Empirical Risk Minimization, ERM) <\/span>\u6846\u67b6\u3001Wasserstein\u5206\u5e03\u5f0f\u9c81\u68d2\u5b66\u4e60 <\/span>(Wasserstein distributionally robust learning, WDRL) <\/span>\u6846\u67b6\u3001\u4e0d\u53d8\u98ce\u9669\u6700\u5c0f\u5316<\/span> (Invariant Risk Minimization, IRM) <\/span>\u6846\u67b6\u8fdb\u884c\u6bd4\u8f83\uff1a<\/span><\/p>\n


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

<\/p>\n

\u56fe5\uff1a\u5b9e\u9a8c\u7ed3\u679c\u3002(a)\u5404\u73af\u5883\u4e0b\u7684\u6d4b\u8bd5\u6027\u80fd\u3002(b) \u5173\u4e8e\u534a\u5f84\u7684\u6d4b\u8bd5\u6027\u80fd\u3002(c) S\u548cV\u76f8\u5bf9\u4e8e\u534a\u5f84\u7684\u5b66\u4e60\u7cfb\u6570\u503c\u3002<\/span><\/p>\n


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

\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cSAL\u7b97\u6cd5\u5404\u5411\u5f02\u6027\u5730\u8003\u8651\u6bcf\u4e2a\u534f\u53d8\u91cf\u4ee5\u5b9e\u73b0\u66f4\u771f\u5b9e\u7684\u9c81\u68d2\u6027\u3002\u5e76\u4e14\u6784\u5efa\u4e86\u4e00\u4e2a\u66f4\u597d\u7684\u4e0d\u786e\u5b9a\u6027\u96c6\uff0c\u5728\u4e0d\u540c\u5206\u5e03\u7684\u6570\u636e\u4e0a\u53d6\u5f97\u4e86\u4e00\u81f4\u66f4\u597d\u7684\u8868\u73b0\uff0c\u9a8c\u8bc1\u4e86\u7b97\u6cd5\u7684\u6709\u6548\u6027\u3002<\/span><\/p>\n


<\/p>\n

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

\n
\n
\n

2.3 \u5f02\u8d28\u6027\u98ce\u9669\u6700\u5c0f\u5316<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u540c\u6837\u5730\uff0c\u5982\u679c\u60f3\u5b8c\u5168\u5229\u7528\u8bad\u7ec3\u6570\u636e\u4e2d\u53d1\u73b0\u7684\u6240\u6709\u76f8\u5173\u6027\uff0c\u5177\u6709\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u901a\u5e38\u8fd8\u4f1a\u5177\u6709\u8f83\u5dee\u7684\u6cdb\u5316\u6027\u80fd\uff0c\u8fd9\u4e9b\u76f8\u5173\u6027\u5728\u5206\u5e03\u53d8\u5316\u4e0b\u4e0d\u7a33\u5b9a\u3002\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u4e86\u5f02\u8d28\u6027\u98ce\u9669\u6700\u5c0f\u5316<\/span> (Heterogeneous Risk Minimization, HRM) <\/span>\u6846\u67b6\uff0c\u4ee5\u5b9e\u73b0\u5bf9\u6570\u636e\u4e4b\u95f4\u6f5c\u5728\u5f02\u8d28\u6027\u548c\u4e0d\u53d8\u5173\u7cfb\u7684\u8054\u5408\u5b66\u4e60\uff0c\u4ece\u800c\u5728\u5206\u5e03\u53d8\u5316\u7684\u60c5\u51b5\u4e0b\u5b9e\u73b0\u7a33\u5b9a\u7684\u9884\u6d4b\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe6\uff1aHRM\u6846\u67b6<\/span><\/p>\n


<\/p>\n

\u603b\u4f53\u6846\u67b6\u5982\u56fe6\u6240\u793a\uff0c\u8be5\u6846\u67b6\u5305\u542b\u4e24\u4e2a\u6a21\u5757\uff0c\u7528\u4e8e\u5f02\u8d28\u6027\u8bc6\u522b\u7684\u524d\u7aef\u548c\u7528\u4e8e\u4e0d\u53d8\u9884\u6d4b\u7684\u540e\u7aefMp<\/sub><\/span>\u3002\u7ed9\u5b9a\u5f02\u8d28\u6027\u6570\u636e\uff0c\u5b83\u4ece\u5f02\u8d28\u6027\u8bc6\u522b\u6a21\u5757M<\/span>c<\/sub><\/span>\u5f00\u59cb\uff0c\u5229\u7528\u5b66\u4e60\u7684\u53d8\u4f53\u03a8(X)\u8868\u793a\u751f\u6210\u5f02\u8d28\u6027\u73af\u5883\u03b5learn<\/sub><\/span>\u3002\u7136\u540e\u5206\u5e03\u5916\u6cdb\u5316\u9884\u6d4b\u6a21\u5757M<\/span>p<\/sub><\/span>\u4f7f\u7528\u5b66\u4e60\u5230\u7684\u73af\u5883\u6765\u5b66\u4e60MIP\u03a6(X)\u4ee5\u53ca\u4e0d\u53d8\u9884\u6d4b\u6a21\u578bf(\u03a6(X))<\/span><\/span>\u3002\u4e4b\u540e\uff0c\u6211\u4eec\u63a8\u5bfc\u51fa\u53d8\u4f53\u03a8(X)<\/span>\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u5757M<\/span>c<\/sub><\/span>\u3002\u81f3\u4e8e\u201c\u8f6c\u6362\u201d\u6b65\u9aa4\uff0c\u57fa\u4e8e\u6211\u4eec\u7684\u8bbe\u5b9a\uff0c\u6211\u4eec\u5728\u8fd9\u9879\u5de5\u4f5c\u4e2d\u91c7\u7528\u4e86\u7279\u5f81\u9009\u62e9\uff0c\u901a\u8fc7\u5b83\u53ef\u4ee5\u5728\u5b66\u4e60\u66f4\u591a\u4e0d\u53d8\u7684\u7279\u5f81\u65f6\u83b7\u5f97\u66f4\u591a\u7684\u53d8\u5f02\u7279\u5f81\u3002<\/span><\/p>\n


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

HRM\u662f\u4e00\u79cd\u4f18\u5316\u6846\u67b6\uff0c\u53ef\u5b9e\u73b0\u6570\u636e\u548c\u4e0d\u53d8\u9884\u6d4b\u5668\u4e4b\u95f4\u6f5c\u5728\u5f02\u8d28\u6027\u7684\u8054\u5408\u5b66\u4e60\u3002\u5c3d\u7ba1\u5206\u5e03\u53d1\u751f\u53d8\u5316\uff0c\u5728\u8be5\u6846\u67b6\u4e0b\u4ecd\u5177\u6709\u66f4\u597d\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/span><\/p>\n


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

\u4e3a\u4e86\u9a8c\u8bc1\u8be5\u6846\u67b6\u7684\u6709\u6548\u6027\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u5c06HRM\u6846\u67b6\u4e0e\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316 <\/span>(Empirical Risk Minimization, ERM) <\/span>\u6846\u67b6\u3001\u5206\u5e03\u9c81\u68d2\u4f18\u5316<\/span> (Distributionally Robust Optimization, DRO) <\/span>\u6846\u67b6\u3001\u4e0d\u53d8\u5b66\u4e60\u7684\u73af\u5883\u63a8\u7406<\/span> (Environment Inference for Invariant Learning, EIIL) <\/span>\u6846\u67b6\u3001\u5177\u6709\u73af\u5883\u03b5<\/span>tr<\/sub><\/span>\u6807\u7b7e\u7684\u4e0d\u53d8\u98ce\u9669\u6700\u5c0f\u5316<\/span> (IRM) <\/span>\u6846\u67b6\u8fdb\u884c\u6bd4\u8f83\u3002<\/span><\/p>\n


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

\u5b9e\u9a8c\u8868\u660e\uff0c\u4e0e\u57fa\u7ebf\u65b9\u6cd5\u76f8\u6bd4\uff0cHRM\u5728\u5e73\u5747\u6027\u80fd\u548c\u7a33\u5b9a\u6027\u65b9\u9762\u53d6\u5f97\u4e86\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u8868\u73b0\uff0c\u5c24\u5176\u662f\u8de8\u73af\u5883\u635f\u5931\u7684\u65b9\u5dee\u63a5\u8fd1\u4e8e0\u3002\u6b64\u5916\uff0cHRM\u4e0d\u9700\u8981\u73af\u5883\u6807\u7b7e\uff0c\u8fd9\u9a8c\u8bc1\u4e86\u6211\u4eec\u7684\u805a\u7c7b\u7b97\u6cd5\u53ef\u4ee5\u6316\u6398\u6570\u636e\u5185\u90e8\u6f5c\u5728\u7684\u5f02\u8d28\u6027\u3002<\/span><\/p>\n


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

\u7ee7\u7eed\u8fdb\u884c\u4e86\u4e09\u4e2a\u771f\u5b9e\u573a\u666f\u7684\u9884\u6d4b\uff0c\u5305\u62ec\u6c7d\u8f66\u4fdd\u9669\u9884\u6d4b\u3001\u4eba\u4eec\u6536\u5165\u9884\u6d4b\u548c\u623f\u4ef7\u9884\u6d4b\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe7\uff1a\u771f\u5b9e\u573a\u666f\u7684\u9884\u6d4b\u7ed3\u679c\u3002<\/span><\/p>\n

(a) \u6c7d\u8f66\u4fdd\u9669\u9884\u6d4b\u7684\u8bad\u7ec3\u548c\u6d4b\u8bd5\u51c6\u786e\u6027\u3002\u5de6\u5b50\u56fe\u663e\u793a\u4e865\u79cd\u8bbe\u7f6e\u7684\u8bad\u7ec3\u7ed3\u679c\uff0c\u53f3\u5b50\u56fe\u663e\u793a\u4e86\u5b83\u4eec\u5bf9\u5e94\u7684\u6d4b\u8bd5\u7ed3\u679c\u3002(b) \u6536\u5165\u9884\u6d4b\u7684\u8bef\u5206\u7c7b\u7387\u3002(c) \u623f\u4ef7\u9884\u6d4b\u7684\u9884\u6d4b\u8bef\u5dee\u3002<\/span><\/p>\n


<\/p>\n

\u4ece\u5b9e\u9a8c\u7ed3\u679c\u53ef\u4ee5\u770b\u51fa\uff0c\u5728\u6240\u6709\u4efb\u52a1\u548c\u51e0\u4e4e\u6240\u6709\u6d4b\u8bd5\u73af\u5883\u4e2d\uff0cHRM\u59cb\u7ec8\u4fdd\u6301\u6700\u4f73\u6027\u80fd\u3002HRM\u53ef\u4ee5\u6709\u6548\u5730\u63ed\u793a\u548c\u5145\u5206\u5229\u7528\u8bad\u7ec3\u6570\u636e\u7684\u5185\u5728\u5f02\u8d28\u6027\u8fdb\u884c\u4e0d\u53d8\u5b66\u4e60\u3002HRM\u653e\u5bbd\u4e86\u5bf9\u73af\u5883\u6807\u7b7e\u7684\u8981\u6c42\uff0c\u4e3a\u4e0d\u53d8\u5b66\u4e60\u5f00\u8f9f\u4e86\u65b0\u7684\u65b9\u5411\u3002\u5b83\u80fd\u591f\u6db5\u76d6\u5e7f\u6cdb\u7684\u5e94\u7528\uff0c\u4f8b\u5982\u533b\u7597\u4fdd\u5065\u3001\u91d1\u878d\u3001\u8425\u9500\u7b49\u3002<\/span><\/p>\n


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

\n
\n
\n

2.4 \u7a33\u5b9a\u5b66\u4e60\u7684\u7406\u8bba\u89e3\u91ca<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u534f\u53d8\u91cf\u504f\u79fb\u6cdb\u5316\u662f\u5206\u5e03\u5916\u6cdb\u5316<\/span> (out-of-distribution, OOD) <\/span>\u4e2d\u7684\u5178\u578b\u6848\u4f8b\uff0c\u5b83\u8981\u6c42\u5728\u672a\u77e5\u6d4b\u8bd5\u5206\u5e03\u4e0a\u5177\u6709\u826f\u597d\u7684\u6027\u80fd\uff0c\u5e76\u4e14\u8be5\u6d4b\u8bd5\u5206\u5e03\u4e0e\u8bad\u7ec3\u5206\u5e03\u5dee\u8ddd\u4f53\u73b0\u5728\u534f\u53d8\u91cf\u8fc1\u79fb\u4e0a\u3002\u5728\u6d89\u53ca\u56de\u5f52\u7b97\u6cd5\u548c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u51e0\u79cd\u5b66\u4e60\u6a21\u578b\u4e0a\uff0c\u7a33\u5b9a\u7684\u5b66\u4e60\u7b97\u6cd5\u5728\u5904\u7406\u534f\u53d8\u91cf\u79fb\u4f4d\u6cdb\u5316\u65b9\u9762\u5df2\u7ecf\u663e\u793a\u51fa\u4e00\u5b9a\u7684\u6709\u6548\u6027\u3002\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u901a\u8fc7\u5c06\u7a33\u5b9a\u5b66\u4e60\u7b97\u6cd5\u89e3\u91ca\u4e3a\u7279\u5f81\u9009\u62e9\u7684\u8fc7\u7a0b\uff0c\u5411\u7406\u8bba\u5206\u6790\u8fc8\u8fdb\u4e86\u4e00\u6b65\u3002<\/span><\/p>\n


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

\u5177\u4f53\u662f\u8fd9\u6837\u7684\uff0c\u9996\u5148\u5b9a\u4e49\u4e00\u7ec4\u53d8\u91cf\uff0c\u79f0\u4e3a\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6<\/span>\uff08minimal stable variable set\uff09<\/span>\uff0c\u5b83\u662f\u5904\u7406\u5e38\u89c1\u635f\u5931\u51fd\u6570<\/span>\uff08\u5305\u62ec\u5747\u65b9\u635f\u5931\u548c\u4e8c\u5143\u4ea4\u53c9\u71b5\u635f\u5931\uff09<\/span>\u4e0b\u7684\u534f\u53d8\u91cf\u8fc1\u79fb\u6cdb\u5316\u7684\u6700\u5c0f\u4e14\u6700\u4f18\u7684\u53d8\u91cf\u96c6\u5408\u3002\u7136\u540e\u8bc1\u660e\u4e86\u5728\u7406\u60f3\u6761\u4ef6\u4e0b\uff0c\u7a33\u5b9a\u7684\u5b66\u4e60\u7b97\u6cd5\u53ef\u4ee5\u8bc6\u522b\u51fa\u8fd9\u4e2a\u96c6\u5408\u4e2d\u7684\u53d8\u91cf\u3002\u8fd9\u4e9b\u7406\u8bba\u9610\u660e\u4e86\u4e3a\u4ec0\u4e48\u7a33\u5b9a\u5b66\u4e60\u9002\u7528\u4e8e\u534f\u53d8\u91cf\u8fc1\u79fb\u6cdb\u5316\u3002<\/span><\/p>\n


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

<\/p>\n


<\/p>\n

\u5178\u578b\u7a33\u5b9a\u5b66\u4e60\u7b97\u6cd5\u7684\u6846\u67b6\u5982\u7b97\u6cd51\u6240\u793a\u3002\u8be5\u7b97\u6cd5\u901a\u5e38\u5305\u62ec\u4e24\u4e2a\u6b65\u9aa4\uff0c\u5206\u522b\u662f\u91cd\u8981\u6027\u91c7\u6837\u548c\u52a0\u6743\u6700\u5c0f\u4e8c\u4e58\u3002\u5728\u7406\u60f3\u6761\u4ef6\u4e0b\uff0c\u7a33\u5b9a\u5b66\u4e60\u7b97\u6cd5\u53ef\u4ee5\u8bc6\u522b\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6\uff0c\u8fd9\u662f\u53ef\u4ee5\u5728\u534f\u53d8\u91cf\u504f\u79fb\u4e0b\u63d0\u4f9b\u826f\u597d\u9884\u6d4b\u7684\u6700\u5c0f\u53d8\u91cf\u96c6\u3002<\/span><\/p>\n


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

\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6\u4e0e\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u5bc6\u5207\u76f8\u5173\uff0c\u7a33\u5b9a\u5b66\u4e60\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u6709\u52a9\u4e8e\u8bc6\u522b\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u3002\u6b64\u5916\uff0c\u5982\u679c\u4ee5\u534f\u53d8\u91cf\u79fb\u4f4d\u6cdb\u5316\u4e3a\u76ee\u6807\uff0c\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u4e0d\u662f\u5fc5\u9700\u7684\uff0c\u800c\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6\u662f\u5145\u5206\u4e14\u6700\u4f18\u7684\u3002<\/span><\/p>\n


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

\u4e0e\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u76f8\u6bd4\uff0c\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6\u53ef\u4ee5\u5e26\u6765\u4e24\u4e2a\u4f18\u52bf\u3002<\/span><\/p>\n


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

\u2460 \u6761\u4ef6\u72ec\u7acb\u6027\u68c0\u9a8c\u662f\u7cbe\u786e\u53d1\u73b0\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u7684\u5173\u952e\u3002<\/span><\/p>\n


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

\u2461 \u5728\u51e0\u4e2a\u5e38\u89c1\u7684\u673a\u5668\u5b66\u4e60\u4efb\u52a1\u4e2d\uff0c\u5305\u62ec\u56de\u5f52\u548c\u4e8c\u5143\u5206\u7c7b\uff0c\u5e76\u4e0d\u662f\u6240\u6709\u7684\u53d8\u91cf\u90fd\u5728\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u3002\u6700\u5c0f\u7a33\u5b9a\u53d8\u91cf\u96c6\u88ab\u8bc1\u660e\u662f\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u7684\u5b50\u96c6\uff0c\u5b83\u6392\u9664\u4e86\u9a6c\u5c14\u53ef\u592b\u8fb9\u754c\u4e2d\u65e0\u7528\u7684\u53d8\u91cf\uff0c\u7528\u4e8e\u534f\u53d8\u91cf\u79fb\u4f4d\u6cdb\u5316\u3002<\/span><\/p>\n

\u200d<\/span><\/h2>\n

\n
\n


<\/p>\n

\n
\n

<\/section>\n

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

\u4e09\u3001\u7a33\u5b9a\u5b66\u4e60\u7684\u5e94\u7528<\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


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

\n
\n
\n

3.1 \u56fe\u4e0a\u7684\u7a33\u5b9a\u5b66\u4e60<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

3.1.1 \u5177\u6709\u9009\u62e9\u6027\u504f\u5dee\u7684\u591a\u4e2a\u73af\u5883\u4e2d\u5b66\u4e60\u7a33\u5b9a\u56fe<\/span><\/strong><\/span><\/h3>\n


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

\u5982\u4eca\uff0c\u56fe\u5df2\u6210\u4e3a\u4e00\u79cd\u901a\u7528\u4e14\u5f3a\u5927\u7684\u8868\u793a\uff0c\u901a\u8fc7\u5176\u7ed3\u6784\u4e2d\u7f16\u7801\u7684\u5e95\u5c42\u6a21\u5f0f\u6765\u63cf\u8ff0\u4e0d\u540c\u7c7b\u578b\u5b9e\u4f53\u4e4b\u95f4\u7684\u4e30\u5bcc\u5173\u7cfb\u3002\u7136\u800c\uff0c\u56fe\u751f\u6210\u7684\u6570\u636e\u6536\u96c6\u8fc7\u7a0b\u5145\u6ee1\u4e86\u5df2\u77e5\u6216\u672a\u77e5\u7684\u6837\u672c\u9009\u62e9\u6027\u504f\u5dee\uff0c\u5c24\u5176\u662f\u5728\u975e\u5e73\u7a33\u548c\u5f02\u6784\u7684\u73af\u5883\u4e2d\uff0c\u5b9e\u4f53\u4e4b\u95f4\u4f1a\u5b58\u5728\u865a\u5047\u5173\u8054\u3002\u9488\u5bf9\u4ece\u5177\u6709\u9009\u62e9\u6027\u504f\u5dee\u7684\u591a\u4e2a\u73af\u5883\u4e2d\u5b66\u4e60\u7a33\u5b9a\u56fe\u7684\u95ee\u9898\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u65e0\u76d1\u7763\u7684\u7a33\u5b9a\u56fe\u5b66\u4e60<\/span> (Stable Graph Learning, SGL) <\/span>\u6846\u67b6\uff0c\u7528\u4e8e\u4ece\u96c6\u5408\u6570\u636e\u4e2d\u5b66\u4e60\u7a33\u5b9a\u56fe\uff0c\u8be5\u6846\u67b6\u7531GCN <\/span>(Graph Convolutional Networks) <\/span>\u6a21\u5757\u548c\u9488\u5bf9\u9ad8\u7ef4\u7a00\u758f\u96c6\u5408\u6570\u636e\u7684E-VAE<\/span> (element-wise VAE) <\/span>\u6a21\u5757\u7ec4\u6210\u3002<\/span><\/p>\n


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

\u7a33\u5b9a\u56fe\u5b66\u4e60\u7684\u4efb\u52a1\u662f\u5b66\u4e60\u4e00\u4e2a\u8868\u793a\u65e0\u504f\u8fde\u63a5\u7ed3\u6784\u7684\u56feGs<\/sub><\/span><\/span>\uff0c\u56e0\u4e3a\u73af\u5883\u4e2d\u7684\u56fe\u662f\u4ece\u6570\u636e\u751f\u6210\u7684\uff0c\u5982\u679c\u6570\u636e\u7684\u6536\u96c6\u8fc7\u7a0b\u6765\u81ea\u5177\u6709\u9009\u62e9\u6027\u504f\u5dee\u7684\u73af\u5883\uff0c\u5219\u5143\u7d20\u4e4b\u95f4\u7684\u865a\u5047\u76f8\u5173\u6027\u4f1a\u5bfc\u81f4\u56fe\u5728\u5176\u4ed6\u73af\u5883\u4e2d\u8868\u73b0\u4e0d\u4f73\u3002SGL\u6846\u67b6\u80fd\u5f88\u597d\u5730\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0cSGL\u6846\u67b6\u53ef\u4ee5\u5206\u89e3\u4e3a\u4e24\u4e2a\u6b65\u9aa4\uff0c\u5305\u62ec\u57fa\u4e8e\u56fe\u7684\u96c6\u5408\u751f\u6210\u548c\u7a33\u5b9a\u56fe\u5b66\u4e60\u3002\u57fa\u4e8e\u56fe\u7684\u96c6\u5408\u751f\u6210\u7684\u8be6\u7ec6\u6846\u67b6\u5982\u56fe8\u6240\u793a\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe8\uff1a\u57fa\u4e8e\u56fe\u7684\u96c6\u5408\u751f\u6210\u7684\u8be6\u7ec6\u6846\u67b6\u3002<\/span><\/p>\n


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

\u7a33\u5b9a\u56fe\u5b66\u4e60\u8fc7\u7a0b\u56fe\u89e3\u5982\u56fe9\u6240\u793a\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe9\uff1a\u7a33\u5b9a\u56fe\u5b66\u4e60\u7684\u8fc7\u7a0b\u56fe\u89e3\u3002<\/span><\/p>\n


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

\u5728\u6a21\u62df\u5b9e\u9a8c\u4e2d\uff0c\u5982\u56fe10\u6240\u793a\uff0c\u5728\u51e0\u4e4e\u6240\u6709\u7684\u5b9e\u9a8c\u4e2d\uff0cSGL\u6846\u67b6\u7684\u6027\u80fd\u8981\u7a33\u5b9a\u5f97\u591a\uff0c\u7279\u522b\u662f\u5f53\u4e24\u4e2a\u73af\u5883\u4e4b\u95f4\u7684\u5dee\u5f02\u66f4\u663e\u7740\u65f6\uff0c\u5b83\u6bd4\u6240\u6709\u57fa\u7ebf\u65b9\u6cd5\u90fd\u8fbe\u5230\u66f4\u9ad8\u7684\u5e73\u5747\u51c6\u786e\u5ea6\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe10\uff1a\u6a21\u62df\u5b9e\u9a8c\u7ed3\u679c\u3002\u6bcf\u4e2a\u5b50\u56fe\u5bf9\u5e94\u4e00\u4e2a\u5b9e\u9a8c\uff0c\u7d2b\u8272\u66f2\u7ebf\u8868\u793aSGL\u6846\u67b6\u751f\u6210\u7684\u56feGs<\/sub><\/span>\u7684\u5b9e\u9a8c\u8868\u73b0\u3002<\/span><\/p>\n


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

\u800c\u76f8\u5e94\u5730\uff0c\u5728\u771f\u5b9e\u5b9e\u9a8c\u91cc\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u7814\u7a76\u4e86\u5546\u54c1\u63a8\u8350\u7684\u5e38\u89c1\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u7a33\u5b9a\u56fe\u7ed3\u6784\u95ee\u9898\u3002<\/span><\/p>\n


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

\u4ece\u88681\u53ef\u4ee5\u770b\u51fa\uff0cSGL\u6846\u67b6\u751f\u6210\u7684\u56feGs<\/sub><\/span>\u53ef\u4ee5\u5e73\u8861\u4e24\u79cd\u73af\u5883\u4e0b\u7684\u76f8\u5173\u6027\uff0c\u66f4\u7a33\u5b9a\u5730\u8fbe\u5230\u6700\u9ad8\u5e73\u5747\u9884\u6d4b\u7387\u3002<\/span><\/p>\n


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

<\/p>\n

\u88681\uff1a\u4f7f\u7528\u4ece\u5546\u54c1\u7f51\u7edc\u4e2d\u5b66\u4e60\u7684\u9879\u76ee\u5d4c\u5165\u8fdb\u884c\u5e26\u6709\u66dd\u5149\u504f\u5dee\u7684\u8d2d\u4e70\u884c\u4e3a\u9884\u6d4b\u3002<\/span><\/p>\n


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

\u5982\u88682\u6240\u793a\u3002SGL\u6846\u67b6\u53ef\u4ee5\u5f88\u597d\u5730\u5f25\u8865\u5355\u4e00\u73af\u5883\u4e0b\u7684\u4fe1\u606f\u635f\u5931\uff0c\u901a\u8fc7\u5b66\u4e60\u5546\u54c1\u4e4b\u95f4\u7684\u672c\u8d28\u5173\u7cfb\uff0c\u751f\u6210\u6574\u4f53\u6027\u80fd\u6700\u4f73\u7684\u56feG<\/span>s<\/sub><\/span>\u3002<\/span><\/p>\n


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

<\/p>\n

\u88682\uff1a\u4f7f\u7528\u4ece\u5546\u54c1\u7f51\u7edc\u4e2d\u5b66\u4e60\u5230\u7684\u9879\u76ee\u5d4c\u5165\u6765\u9884\u6d4b\u4e0d\u540c\u6027\u522b\u7fa4\u4f53\u7684\u8d2d\u4e70\u884c\u4e3a\u3002<\/span><\/p>\n


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

\u56fe\u751f\u6210\u7684\u6570\u636e\u9009\u62e9\u6027\u504f\u5dee\u53ef\u80fd\u5bfc\u81f4\u6709\u504f\u5dee\u7684\u56fe\u7ed3\u6784\u5728Non-I.I.D.\u573a\u666f\u4e2d\u6027\u80fd\u4e0d\u4f73\u3002\u9488\u5bf9\u8be5\u95ee\u9898\u63d0\u51fa\u7684SGL\u6846\u67b6\u53ef\u4ee5\u63d0\u9ad8\u5b66\u4e60\u56fe\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u5e76\u80fd\u5f88\u597d\u5730\u9002\u5e94\u4e0d\u540c\u7c7b\u578b\u7684\u56fe\u8868\u548c\u6536\u96c6\u7684\u6570\u636e\u3002<\/span><\/p>\n


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

3.1.2 \u5177\u6709\u4e0d\u53ef\u77e5\u5206\u5e03\u504f\u79fb\u7684\u56fe\u7684\u7a33\u5b9a\u9884\u6d4b<\/span><\/strong><\/span><\/h3>\n


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

\u56fe\u795e\u7ecf\u7f51\u7edc<\/span> (Graph Neural Networks, GNNs) <\/span>\u5df2\u88ab\u8bc1\u660e\u5728\u5177\u6709\u968f\u673a\u5206\u79bb\u7684\u8bad\u7ec3\u548c\u6d4b\u8bd5\u6570\u636e\u7684\u5404\u79cd\u56fe\u4efb\u52a1\u4e0a\u662f\u6709\u6548\u7684\u3002\u7136\u800c\uff0c\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u8bad\u7ec3\u56fe\u7684\u5206\u5e03\u53ef\u80fd\u4e0e\u6d4b\u8bd5\u56fe\u7684\u5206\u5e03\u4e0d\u540c\u3002\u6b64\u5916\uff0c\u5728\u8bad\u7ec3GNNs\u65f6\uff0c\u6d4b\u8bd5\u6570\u636e\u7684\u5206\u5e03\u59cb\u7ec8\u662f\u4e0d\u53ef\u77e5\u7684\u3002\u56e0\u6b64\uff0c\u5927\u5bb6\u9762\u4e34\u7740\u56fe\u5b66\u4e60\u8bad\u7ec3\u548c\u6d4b\u8bd5\u4e4b\u95f4\u7684\u4e0d\u53ef\u77e5\u5206\u5e03\u8f6c\u53d8\uff0c\u8fd9\u5c06\u5bfc\u81f4\u4f20\u7edfGNNs\u5728\u4e0d\u540c\u6d4b\u8bd5\u73af\u5883\u4e2d\u7684\u63a8\u7406\u4e0d\u7a33\u5b9a\u3002<\/span><\/p>\n


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

\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u6d59\u6c5f\u5927\u5b66\u51b5\u7428\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684GNNs\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\uff0c\u5b83\u5141\u8bb8\u5728\u56fe\u4e0a\u8fdb\u884c\u5c40\u90e8\u548c\u5168\u5c40\u7a33\u5b9a\u7684\u5b66\u4e60\u548c\u9884\u6d4b,\u53ef\u4ee5\u51cf\u5c11\u5f02\u6784\u73af\u5883\u4e2d\u7684\u8bad\u7ec3\u635f\u5931\uff0c\u4ece\u800c\u4f7fGNNs\u80fd\u591f\u5f88\u597d\u5730\u6cdb\u5316\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u662f\u4e3aGNNs\u8bbe\u8ba1\u4e86\u4e00\u79cd\u65b0\u7684\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\uff0c\u8be5\u6846\u67b6\u80fd\u6355\u83b7\u6bcf\u4e2a\u8282\u70b9\u7684\u7a33\u5b9a\u5c5e\u6027\uff0c\u5728\u6b64\u57fa\u7840\u4e0a\u5b66\u4e60\u8282\u70b9\u8868\u793a\u5e76\u8fdb\u884c\u9884\u6d4b<\/span>\uff08\u5c40\u90e8\u7a33\u5b9a\uff09<\/span>\uff0c\u5e76\u89c4\u8303GNNs\u5728\u5f02\u6784\u73af\u5883\u4e2d\u7684\u8bad\u7ec3<\/span>\uff08\u5168\u5c40\u7a33\u5b9a\uff09<\/span>\u3002\u8be5\u65b9\u6cd5\u7684\u672c\u8d28\u5982\u56fe11\u6240\u793a\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe11\uff1a\u6574\u4f53\u67b6\u6784\u3002\u7531\u4e24\u4e2a\u57fa\u672c\u7ec4\u6210\u90e8\u5206\u7ec4\u6210\uff0c\u5373\u5728\u6bcf\u4e2a\u76ee\u6807\u8282\u70b9\u7684\u8868\u793a\u5b66\u4e60\u4e2d\u6355\u83b7\u8de8\u73af\u5883\u7a33\u5b9a\u7684\u5c5e\u6027\u7684\u5c40\u90e8\u7a33\u5b9a\u5b66\u4e60\uff0c\u4ee5\u53ca\u663e\u5f0f\u5e73\u8861\u4e0d\u540c\u8bad\u7ec3\u7684\u5168\u5c40\u7a33\u5b9a\u5b66\u4e60\u73af\u5883\u3002<\/span><\/p>\n


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

\u5728\u56fe\u57fa\u51c6\u5b9e\u9a8c\u4e2d\uff0c\u6d59\u6c5f\u5927\u5b66\u51b5\u7428\u8001\u5e08\u56e2\u961f\u4f7f\u7528OGB\u6570\u636e\u96c6\u548c\u4f20\u7edf\u57fa\u51c6Citeseer\u6570\u636e\u96c6\uff0c\u6784\u5efa\u4e24\u5c42GCN\u548cGAT\u3002\u6240\u6709\u5176\u4ed6\u65b9\u6cd5<\/span>\uff08\u5305\u62ec\u6211\u4eec\u7684\u65b9\u6cd5\uff09<\/span>\u4e5f\u5305\u542b\u4e24\u4e2a\u56fe\u5f62\u5c42\u4ee5\u8fdb\u884c\u516c\u5e73\u6bd4\u8f83\u3002OGBarxiv\u7684\u6240\u6709\u65b9\u6cd5\u7684\u9690\u85cf\u5c42\u795e\u7ecf\u8282\u70b9\u4e2a\u6570\u4e3a250\uff0cCiteseer\u7684\u9690\u85cf\u5c42\u795e\u7ecf\u8282\u70b9\u4e2a\u6570\u4e3a64\uff0c\u5b66\u4e60\u7387\u4e3a0.002\u3002<\/span><\/p>\n


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

\u6d4b\u8bd5\u7ed3\u679c\u5982\u56fe12\u548c\u56fe13\u6240\u793a\u3002\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\u6709\u7740\u66f4\u7a33\u5b9a\u7684\u5b9e\u9a8c\u7ed3\u679c\u3002\u5f53\u6d4b\u8bd5\u5206\u5e03\u4e0e\u8bad\u7ec3\u5206\u5e03\u7684\u5dee\u5f02\u66f4\u5927\u65f6\uff0c\u5927\u591a\u6570GNNs\u4f1a\u906d\u53d7\u5206\u5e03\u53d8\u5316\u5e76\u4e14\u4ea7\u751f\u8f83\u5dee\u7684\u6027\u80fd<\/span>\uff08\u4f8b\u5982\uff0c\u56fe12a\u7684\u53f3\u4fa7\uff09<\/span>\u3002\u5c3d\u7ba1\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\u5728\u5206\u5e03\u66f4\u63a5\u8fd1\u8bad\u7ec3\u7684\u6d4b\u8bd5\u73af\u5883\u4e2d\u727a\u7272\u4e86\u4e00\u4e9b\u6027\u80fd<\/span>\uff08\u4f8b\u5982\uff0c\u56fe12a\u7684\u5de6\u4fa7\uff09<\/span>\uff0c\u4f46\u83b7\u5f97\u4e86\u663e\u7740\u66f4\u9ad8\u7684 Average_Score \u548c\u66f4\u4f4e\u7684 Stability_Error\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe12\uff1a\u5728OGB-Arxiv\u6570\u636e\u96c6\u4e0a\u7684\u6d4b\u8bd5\u7ed3\u679c\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe13\uff1a\u5728Citeseer\u6570\u636e\u96c6\u4e0a\u7684\u6d4b\u8bd5\u7ed3\u679c<\/span><\/p>\n


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

\u4e3a\u4e86\u8bc1\u660e\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u6709\u6548\u6027\uff0c\u6d59\u6c5f\u5927\u5b66\u51b5\u7428\u8001\u5e08\u56e2\u961f\u6536\u96c6\u771f\u5b9e\u4e16\u754c\u7684\u5608\u6742\u6570\u636e\u96c6\uff0c\u5bf9\u63a8\u8350\u7cfb\u7edf\u7684\u7528\u6237-\u9879\u76ee\u4e8c\u5206\u56fe\u8fdb\u884c\u4e86\u5b9e\u9a8c\u3002\u5982\u56fe14\u300115\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u7a33\u5b9a\u9884\u6d4b\u6846\u67b6\u6bd4\u5176\u4ed6\u57fa\u7ebf\u65b9\u6cd5\u53d6\u5f97\u4e86\u660e\u663e\u66f4\u7a33\u5b9a\u7684\u7ed3\u679c\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe14\uff1a\u5177\u6709\u7531\u8282\u70b9\u5c5e\u6027\u5f15\u8d77\u7684\u5206\u5e03\u504f\u79fb\u7684\u771f\u5b9e\u4e16\u754c\u63a8\u8350\u6570\u636e\u96c6\u7684\u7ed3\u679c\u3002<\/span><\/p>\n


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

\u56fe15\uff1a\u5177\u6709\u771f\u5b9e\u4e16\u754c\u73af\u5883\u7684\u63a8\u8350\u6570\u636e\u96c6\u7684\u7ed3\u679c\uff08\u6bcf\u5929\u4f5c\u4e3a\u4e00\u4e2a\u5355\u72ec\u7684\u73af\u5883\uff09\u3002<\/span><\/p>\n


<\/p>\n

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

\n
\n
\n

3.2 \u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u7a33\u5b9a\u5b66\u4e60<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n


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

\u57fa\u4e8e\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u65b9\u6cd5\u5728\u6d4b\u8bd5\u6570\u636e\u548c\u8bad\u7ec3\u6570\u636e\u5171\u4eab\u76f8\u4f3c\u5206\u5e03\u65f6\u53d6\u5f97\u4e86\u60ca\u4eba\u7684\u6027\u80fd\uff0c\u4f46\u6709\u65f6\u53ef\u80fd\u4f1a\u5931\u8d25\u3002\u56e0\u6b64\uff0c\u6d88\u9664\u8bad\u7ec3\u548c\u6d4b\u8bd5\u6570\u636e\u4e4b\u95f4\u5206\u5e03\u53d8\u5316\u7684\u5f71\u54cd\u5bf9\u4e8e\u6784\u5efa\u6027\u80fd\u6709\u5e0c\u671b\u7684\u6df1\u5ea6\u6a21\u578b\u81f3\u5173\u91cd\u8981\u3002\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u5efa\u8bae\u901a\u8fc7\u5b66\u4e60\u8bad\u7ec3\u6837\u672c\u7684\u6743\u91cd\u6765\u6d88\u9664\u7279\u5f81\u4e4b\u95f4\u7684\u4f9d\u8d56\u5173\u7cfb\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u8fd9\u6709\u52a9\u4e8e\u6df1\u5ea6\u6a21\u578b\u6446\u8131\u865a\u5047\u5173\u8054\uff0c\u8fdb\u800c\u66f4\u591a\u5730\u5173\u6ce8\u5224\u522b\u7279\u5f81\u548c\u6807\u7b7e\u4e4b\u95f4\u7684\u771f\u6b63\u8054\u7cfb\u3002<\/span><\/p>\n


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

\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u79cd\u79f0\u4e3aStableNet\u7684\u65b9\u6cd5\u3002\u8be5\u65b9\u6cd5\u901a\u8fc7\u5168\u5c40\u52a0\u6743\u6837\u672c\u6765\u89e3\u51b3\u5206\u5e03\u504f\u79fb\u95ee\u9898\uff0c\u4ee5\u76f4\u63a5\u5bf9\u6bcf\u4e2a\u8f93\u5165\u6837\u672c\u7684\u6240\u6709\u7279\u5f81\u8fdb\u884c\u53bb\u76f8\u5173\uff0c\u4ece\u800c\u6d88\u9664\u76f8\u5173\u548c\u4e0d\u76f8\u5173\u7279\u5f81\u4e4b\u95f4\u7684\u7edf\u8ba1\u76f8\u5173\u6027\u3002\u8fd9\u662f\u4e00\u79cd\u57fa\u4e8e\u968f\u673a\u5085\u7acb\u53f6\u7279\u5f81 <\/span>(Random Fourier Features, RFF) <\/span>\u7684\u65b0\u578b\u975e\u7ebf\u6027\u7279\u5f81\u53bb\u76f8\u5173\u65b9\u6cd5\uff0c\u5177\u6709\u7ebf\u6027\u8ba1\u7b97\u590d\u6742\u5ea6\u3002\u540c\u65f6\uff0c\u5b83\u4e5f\u662f\u6709\u6548\u7684\u4f18\u5316\u673a\u5236\uff0c\u901a\u8fc7\u8fed\u4ee3\u4fdd\u5b58\u548c\u91cd\u65b0\u52a0\u8f7d\u6a21\u578b\u7684\u7279\u5f81\u548c\u6743\u91cd\u6765\u5168\u5c40\u611f\u77e5\u548c\u6d88\u9664\u76f8\u5173\u6027\uff0c\u8fd8\u80fd\u5728\u8bad\u7ec3\u6570\u636e\u91cf\u5927\u65f6\u51cf\u5c11\u5b58\u50a8\u7684\u4f7f\u7528\u548c\u8ba1\u7b97\u6210\u672c\u3002\u6b64\u5916\uff0c\u5982\u56fe16\u6240\u793a\uff0cStableNet\u53ef\u4ee5\u6709\u6548\u5730\u5254\u9664\u4e0d\u76f8\u5173\u7684\u7279\u5f81<\/span>\uff08\u4f8b\u5982\uff0c\u6c34\uff09<\/span>\u5e76\u5229\u7528\u771f\u6b63\u76f8\u5173\u7684\u7279\u5f81\u8fdb\u884c\u9884\u6d4b\uff0c\u4ece\u800c\u5728\u91ce\u5916\u975e\u5e73\u7a33\u73af\u5883\u4e2d\u83b7\u5f97\u66f4\u7a33\u5b9a\u7684\u6027\u80fd\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe16\uff1a\u5f53\u8bc6\u522b\u72d7\u7684\u8bad\u7ec3\u56fe\u50cf\u5305\u542b\u5f88\u591a\u6c34\u65f6\uff0cStableNet\u6a21\u578b\u4e3b\u8981\u5173\u6ce8\u4e8e\u72d7\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe17\uff1aStableNet\u7684\u6574\u4f53\u67b6\u6784\u3002<\/span><\/p>\n


<\/p>\n

\u4e3a\u4e86\u6db5\u76d6\u66f4\u666e\u904d\u548c\u66f4\u5177\u6311\u6218\u6027\u7684\u5206\u5e03\u53d8\u5316\u6848\u4f8b\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u5728\u5b9e\u9a8c\u4e2d\u91c7\u7528\u5982\u4e0b\u56db\u79cd\u8bbe\u7f6e\uff1a\u975e\u5e73\u8861\u3001\u7075\u6d3b\u3001\u5bf9\u6297\u3001\u7ecf\u5178\u3002\u5728\u4e0d\u540c\u7684\u5b9e\u9a8c\u8bbe\u7f6e\u4e0b\uff0cStableNet\u90fd\u80fd\u4e0d\u540c\u7a0b\u5ea6\u5f97\u4f18\u4e8e\u5176\u4ed6\u65b9\u6cd5\u3002<\/span><\/p>\n


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

\u5728\u6d88\u878d\u7814\u7a76\u4e2d\uff0c\u901a\u8fc7\u968f\u673a\u9009\u62e9\u7528\u4e8e\u8ba1\u7b97\u5177\u6709\u4e0d\u540c\u6bd4\u7387\u7684\u4f9d\u8d56\u5173\u7cfb\u7684\u7279\u5f81\u6765\u8fdb\u4e00\u6b65\u964d\u4f4e\u7279\u5f81\u7ef4\u5ea6\u3002\u56fe18\u663e\u793a\u4e86\u5177\u6709\u4e0d\u540c\u7ef4\u5ea6\u968f\u673a\u5085\u91cc\u53f6\u7279\u5f81\u7684\u5b9e\u9a8c\u7ed3\u679c\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe18\uff1a\u6d88\u878d\u7814\u7a76\u7684\u7ed3\u679c\u3002<\/span><\/p>\n


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

\u56fe\u50cf\u5206\u7c7b\u6a21\u578b\u7684\u4e00\u79cd\u76f4\u89c2\u89e3\u91ca\u662f\u8bc6\u522b\u5bf9\u6700\u7ec8\u51b3\u7b56\u6709\u5f88\u5927\u5f71\u54cd\u7684\u50cf\u7d20\u3002\u6240\u4ee5\uff0c\u5728\u663e\u7740\u6027\u56fe\u50cf\u4e0a\uff0c\u4e3a\u4e86\u6f14\u793a\u6a21\u578b\u5728\u8fdb\u884c\u9884\u6d4b\u65f6\u662f\u5173\u6ce8\u5bf9\u8c61\u8fd8\u662f\u4e0a\u4e0b\u6587<\/span>\uff08\u57df\uff09<\/span>\uff0c\u5bf9\u7c7b\u522b\u5f97\u5206\u51fd\u6570\u76f8\u5bf9\u4e8e\u8f93\u5165\u50cf\u7d20\u7684\u68af\u5ea6\u8fdb\u884c\u4e86\u53ef\u89c6\u5316\u3002\u53ef\u89c6\u5316\u7ed3\u679c\u5982\u56fe19\u6240\u793a\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe19\uff1aStableNet\u7684\u663e\u7740\u6027\u56fe\u50cf\u3002\u50cf\u7d20\u8d8a\u4eae\uff0c\u5b83\u5bf9\u9884\u6d4b\u7684\u8d21\u732e\u5c31\u8d8a\u5927\u3002<\/span><\/p>\n


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

\u5404\u79cd\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cStableNet\u65b9\u6cd5\u53ef\u4ee5\u901a\u8fc7\u6837\u672c\u52a0\u6743\u6d88\u9664\u76f8\u5173\u548c\u4e0d\u76f8\u5173\u7279\u5f81\u4e4b\u95f4\u7684\u7edf\u8ba1\u76f8\u5173\u6027\uff0c\u8fdb\u800c\u6709\u6548\u5254\u9664\u4e0d\u76f8\u5173\u7684\u7279\u5f81\u5e76\u5229\u7528\u771f\u6b63\u76f8\u5173\u7684\u7279\u5f81\u8fdb\u884c\u9884\u6d4b\u3002<\/span><\/p>\n


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

\n
\n
\n

3.3 \u7a33\u5b9a\u5b66\u4e60\u4e0e\u516c\u5e73\u6027<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u5982\u4eca\uff0c\u516c\u5e73\u95ee\u9898\u5df2\u7ecf\u6210\u4e3a\u4e86\u51b3\u7b56\u7cfb\u7edf\u4e2d\u7684\u91cd\u8981\u95ee\u9898\u3002\u5df2\u7ecf\u6709\u5f88\u591a\u5b66\u8005\u63d0\u51fa\u4e86\u5404\u79cd\u516c\u5e73\u7684\u6982\u5ff5\u6765\u8861\u91cf\u7b97\u6cd5\u7684\u4e0d\u516c\u5e73\u7a0b\u5ea6\u3002\u73c0\u5c14\u7814\u7a76\u4e86\u4f2f\u514b\u5229\u5927\u5b66\u7814\u7a76\u751f\u5165\u5b66\u6027\u522b\u504f\u89c1\u7684\u6848\u4f8b\u3002\u6570\u636e\u663e\u793a\uff0c\u603b\u4f53\u800c\u8a00\uff0c\u7537\u6027\u7533\u8bf7\u4eba\u7684\u5165\u5b66\u7387\u8f83\u9ad8\uff0c\u4f46\u5728\u7814\u7a76\u9662\u7cfb\u9009\u62e9\u65f6\uff0c\u7ed3\u679c\u6709\u6240\u4e0d\u540c\u3002\u7531\u9662\u7cfb\u9009\u62e9\u5f15\u8d77\u7684\u504f\u5dee\u5e94\u8be5\u88ab\u8ba4\u4e3a\u662f\u516c\u5e73\u7684\uff0c\u4f46\u4f20\u7edf\u7684\u7fa4\u4f53\u516c\u5e73\u89c2\u5ff5\u7531\u4e8e\u6ca1\u6709\u8003\u8651\u9662\u7cfb\u9009\u62e9\u800c\u65e0\u6cd5\u5224\u65ad\u516c\u5e73\u3002\u53d7\u6b64\u542f\u53d1\uff0c\u57fa\u4e8e\u56e0\u679c\u5173\u7cfb\u7684\u516c\u5e73\u7406\u5ff5\u5e94\u8fd0\u800c\u751f\u3002\u5728\u8fd9\u4e9b\u8bba\u6587\u4e2d\uff0c\u4f5c\u8005\u9996\u5148\u5047\u8bbe\u4e86\u7279\u5f81\u4e4b\u95f4\u7684\u56e0\u679c\u56fe\uff0c\u7136\u540e\uff0c\u4ed6\u4eec\u53ef\u4ee5\u5c06\u654f\u611f\u5c5e\u6027\u5bf9\u7ed3\u679c\u7684\u4e0d\u516c\u5e73\u56e0\u679c\u6548\u5e94\u5b9a\u4e49\u4e3a\u4e00\u4e2a\u5ea6\u91cf\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u516c\u5e73\u6027\u6982\u5ff5\u9700\u8981\u975e\u5e38\u6709\u529b\u7684\u5047\u8bbe\uff0c\u800c\u4e14\u5b83\u4eec\u4e0d\u53ef\u6269\u5c55\u3002\u5728\u5b9e\u8df5\u4e2d\uff0c\u7ecf\u5e38\u5b58\u5728\u4e00\u7ec4\u6211\u4eec\u79f0\u4e4b\u4e3a\u516c\u5e73\u53d8\u91cf\u7684\u53d8\u91cf\uff0c\u5b83\u4eec\u662f\u51b3\u7b56\u524d\u7684\u534f\u53d8\u91cf\uff0c\u4f8b\u5982\u7528\u6237\u7684\u9009\u62e9\u3002<\/span><\/p>\n


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

\u516c\u5e73\u53d8\u91cf\u5e76\u4e0d\u4f1a\u5f71\u54cd\u8bc4\u4f30\u51b3\u7b56\u652f\u6301\u7b97\u6cd5\u7684\u516c\u5e73\u6027\u3002\u56e0\u6b64\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u901a\u8fc7\u8bbe\u7f6e\u516c\u5e73\u53d8\u91cf\u5c06\u6761\u4ef6\u516c\u5e73\u5b9a\u4e49\u4e3a\u66f4\u5408\u7406\u7684\u516c\u5e73\u5ea6\u91cf\u3002\u901a\u8fc7\u9009\u53d6\u4e0d\u540c\u7684\u516c\u5e73\u53d8\u91cf\uff0c\u5d14\u9e4f\u8001\u5e08\u56e2\u961f\u8bc1\u660e\u4e86\u4f20\u7edf\u7684\u516c\u5e73\u6982\u5ff5\uff0c\u4f8b\u5982\u7edf\u8ba1\u516c\u5e73\u548c\u673a\u4f1a\u5747\u7b49\uff0c\u662f\u6761\u4ef6\u516c\u5e73\u7b26\u53f7\u7684\u7279\u4f8b\u3002\u5e76\u4e14\u63d0\u51fa\u4e86\u4e00\u79cd\u53ef\u6c42\u5bfc\u7684\u6761\u4ef6\u516c\u5e73\u6b63\u5219\u5316\u5668<\/span>\uff08Derivable <\/span>Conditional Fairness Regularizer, DCFR\uff09<\/span><\/span>\uff0c\u5b83\u53ef\u4ee5\u96c6\u6210\u5230\u4efb\u4f55\u51b3\u7b56\u6a21\u578b\u4e2d\uff0c\u4ee5\u8ddf\u8e2a\u7b97\u6cd5\u51b3\u7b56\u7684\u7cbe\u5ea6\u548c\u516c\u5e73\u6027\u4e4b\u95f4\u7684\u6743\u8861\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe20\uff1aDCFR\u7684\u6846\u67b6\u3002<\/span><\/p>\n


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

\u4e3a\u4e86\u516c\u5e73\u6bd4\u8f83\uff0c\u5728\u5b9e\u9a8c\u4e2d\uff0c\u9009\u62e9\u4e5f\u4f7f\u7528\u5bf9\u6297\u6027\u8868\u793a\u5b66\u4e60\u7684\u65b9\u6cd5\u6765\u89e3\u51b3\u95ee\u9898\u7684\u516c\u5e73\u4f18\u5316\u7b97\u6cd5\u4f5c\u5bf9\u7167\u3002\u6709UNFAIR\u3001ALFR\u3001CFAIR\u548cLAFTR\uff0c\u4ee5\u53ca\u5b83\u7684\u53d8\u4f53LAFTR-DP\u548cLAFTR-EO\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe21\uff1a\u5404\u79cd\u6570\u636e\u96c6\uff08\u4ece\u4e0a\u5230\u4e0b\u4f9d\u6b21\u4e3a\u6536\u5165\u6570\u636e\u96c6\u3001\u8377\u5170\u4eba\u53e3\u666e\u67e5\u6570\u636e\u96c6\u3001COMPAS\u6570\u636e\u96c6\uff09\u4e0a\u4e0d\u540c\u516c\u5e73\u6027\u6307\u6807\uff08\u4ece\u5de6\u5230\u53f3\u4f9d\u6b21\u4e3a\u0394\ud835\udc37\ud835\udc43\u3001\u0394\ud835\udc38\ud835\udc42\u3001\u0394\ud835\udc36\ud835\udc39\uff09\u7684\u51c6\u786e\u6027-\u516c\u5e73\u6027\u6743\u8861\u66f2\u7ebf\u3002DCFR\u4ee5\u7c97\u7ebf\u663e\u793a\u3002<\/span><\/p>\n


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

\u5f88\u660e\u663e\uff0c\u5728\u5b9e\u9a8c\u4e2dDCFR\u66f4\u6709\u4f18\u52bf\uff0c\u5728\u51c6\u786e\u6027\u548c\u516c\u5e73\u6027\u4e0a\u8fbe\u5230\u66f4\u597d\u7684\u6743\u8861\u6548\u679c\u3002\u5bf9\u4e8e\u7edf\u8ba1\u516c\u5e73\u548c\u673a\u4f1a\u5747\u7b49\u4efb\u52a1\uff0cDCFR\u7684\u9000\u5316\u53d8\u4f53\u80fd\u6709\u4e0e\u4e13\u4e3a\u8fd9\u4e9b\u4efb\u52a1\u8bbe\u8ba1\u7684\u6700\u5148\u8fdb\u57fa\u7ebf\u65b9\u6cd5\u76f8\u5f53\u7684\u6027\u80fd\uff0c\u6709\u65f6\u751a\u81f3\u8fd8\u80fd\u6709\u66f4\u597d\u7684\u7ed3\u679c\u3002\u7efc\u4e0a\u6240\u8ff0\uff0cDCFR\u5728\u771f\u5b9e\u6570\u636e\u96c6\u4e0a\u975e\u5e38\u6709\u6548\uff0c\u5e76\u5728\u6761\u4ef6\u516c\u5e73\u76ee\u6807\u4e0a\u53d6\u5f97\u4e86\u826f\u597d\u7684\u6027\u80fd\u3002\u5e76\u4e14\u968f\u7740\u516c\u5e73\u53d8\u91cf\u7684\u6570\u91cf\u589e\u52a0\uff0c\u5176\u8868\u73b0\u4f1a\u66f4\u597d\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

3.4 \u7a33\u5b9a\u5b66\u4e60\u4e0e\u9886\u57df\u81ea\u9002\u5e94<\/span><\/strong><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u7a33\u5b9a\u5b66\u4e60\u6700\u521d\u7684\u5b9a\u4e49\u662f\u4e0d\u9700\u8981\u76ee\u6807\u57df\u4fe1\u606f\u7684\uff0c\u8fd9\u91cc\u7684\u9886\u57df\u81ea\u9002\u5e94\u662f\u4e00\u79cd\u5229\u7528\u4e86\u76ee\u6807\u57df\u4fe1\u606f\u7684\u505a\u6cd5\uff0c\u53ef\u4ee5\u7406\u89e3\u4e3a\u62d3\u5c55\u4e86\u6700\u521d\u7684\u7a33\u5b9a\u5b66\u4e60\u7684\u542b\u4e49\u3002<\/span><\/p>\n


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

\u7814\u7a76\u8868\u660e\uff0c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u5b66\u4e60\u5230\u7684\u8868\u5f81\u53ef\u4ee5\u8f6c\u79fb\u5230\u6211\u4eec\u6ca1\u6709\u5145\u8db3\u6807\u8bb0\u6570\u636e\u7684\u5176\u4ed6\u9886\u57df\u4e2d\uff0c\u5e76\u8fdb\u884c\u7c7b\u4f3c\u7684\u9884\u6d4b\u4efb\u52a1\u3002\u7136\u800c\uff0c\u5f53\u6211\u4eec\u8fc7\u6e21\u5230\u6a21\u578b\u4e2d\u7684\u66f4\u9ad8\u795e\u7ecf\u5c42\u65f6\uff0c\u8868\u5f81\u53d8\u5f97\u66f4\u52a0\u9002\u7528\u4e8e\u7279\u5b9a\u4efb\u52a1\u800c\u4e0d\u901a\u7528\u3002\u5173\u4e8e\u8fd9\u4e2a\u95ee\u9898\uff0c\u6df1\u5ea6\u57df\u9002\u5e94\u7684\u7814\u7a76\u63d0\u51fa\u901a\u8fc7\u5f3a\u5236\u6df1\u5ea6\u6a21\u578b\u5b66\u4e60\u66f4\u591a\u8de8\u57df\u53ef\u8fc1\u79fb\u7684\u8868\u5f81\u6765\u7f13\u89e3\u3002\u8fd9\u5176\u5b9e\u662f\u901a\u8fc7\u5c06\u57df\u9002\u5e94\u65b9\u6cd5\u6574\u5408\u5230\u6df1\u5ea6\u5b66\u4e60\u7ba1\u9053\u4e2d\u6765\u5b9e\u73b0\u7684\u3002\u7136\u800c\uff0c\u76f8\u5173\u6027\u5e76\u4e0d\u603b\u662f\u53ef\u8f6c\u79fb\u7684\u3002\u4e9a\u5229\u6851\u90a3\u5dde\u7acb\u5927\u5b66<\/span>\uff08Arizona State University\uff0cASU\uff09<\/span>\u5218\u6b22\u8001\u5e08\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u4e2a\u7528\u4e8e\u65e0\u76d1\u7763\u57df\u9002\u5e94<\/span> (Deep Causal Representation learning framework for unsupervised Domain Adaptation, DCDAN)<\/span> \u7684\u6df1\u5ea6\u56e0\u679c\u8868\u793a\u5b66\u4e60\u6846\u67b6\uff0c\u4ee5\u5b66\u4e60\u7528\u4e8e\u76ee\u6807\u57df\u9884\u6d4b\u7684\u53ef\u8fc1\u79fb\u7279\u5f81\u8868\u793a\uff0c\u5982\u56fe22\u6240\u793a\u3002\u5176\u5b9e\u5c31\u662f\u4f7f\u7528\u6765\u81ea\u6e90\u57df\u7684\u91cd\u65b0\u52a0\u6743\u6837\u672c\u6765\u6a21\u62df\u865a\u62df\u76ee\u6807\u57df\uff0c\u5e76\u4f30\u8ba1\u7279\u5f81\u5bf9\u7ed3\u679c\u7684\u56e0\u679c\u5f71\u54cd\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe22\uff1aDCDAN\u6982\u8ff0\u3002<\/span><\/p>\n


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

DCDAN\u7531\u4e00\u4e2a\u6b63\u5219\u5316\u9879\u7ec4\u6210\uff0c\u8be5\u6b63\u5219\u5316\u9879\u901a\u8fc7\u5e73\u8861\u4ece\u6570\u636e\u4e2d\u5b66\u4e60\u5230\u7684\u7279\u5f81\u8868\u793a\u7684\u5206\u5e03\u6765\u5b66\u4e60\u6e90\u6570\u636e\u7684\u5e73\u8861\u6743\u91cd\u3002\u8fd9\u4e9b\u6743\u91cd\u7684\u8bbe\u8ba1\u6709\u52a9\u4e8e\u6a21\u578b\u6355\u6349\u7279\u5f81\u5bf9\u76ee\u6807\u53d8\u91cf\u7684\u56e0\u679c\u5f71\u54cd\uff0c\u800c\u4e0d\u662f\u5b83\u4eec\u7684\u76f8\u5173\u6027\u3002\u6b64\u5916\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5305\u62ec\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u52a0\u6743\u635f\u5931\u51fd\u6570\uff0c\u5176\u4e2d\u6bcf\u4e2a\u6837\u672c\u7684\u6743\u91cd\u6765\u81ea\u6b63\u5219\u5316\u9879\uff0c\u635f\u5931\u51fd\u6570\u8d1f\u8d23\u5b66\u4e60\u9884\u6d4b\u57df\u4e0d\u53d8\u7279\u5f81\uff0c\u4ee5\u53ca\u5c06\u5b66\u4e60\u5230\u7684\u8868\u5f81\u6620\u5c04\u5230\u8f93\u51fa\u7684\u5206\u7c7b\u5668\u6216\u56e0\u679c\u673a\u5236\u3002\u5c06\u5b66\u4e60\u7ec4\u4ef6\u7684\u6837\u672c\u6743\u91cd\u5d4c\u5165\u5230\u6a21\u578b\u7684\u7ba1\u9053\u4e2d\uff0c\u5e76\u5c06\u8fd9\u4e9b\u6743\u91cd\u4e0e\u8868\u5f81\u8054\u5408\u5b66\u4e60\uff0c\u8fd9\u6837\u4e0d\u4ec5\u53ef\u4ee5\u4ece\u6df1\u5ea6\u6a21\u578b\u4e2d\u53d7\u76ca\uff0c\u8fd8\u80fd\u5b66\u4e60\u5bf9\u76ee\u6807\u5177\u6709\u53ef\u8f6c\u79fb\u6027\u548c\u826f\u597d\u9884\u6d4b\u6548\u679c\u7684\u56e0\u679c\u7279\u5f81\u3002<\/span><\/p>\n


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

\u56fe23\uff1aDCDAN\u751f\u6210\u7684\u6570\u636e\u96c6\u4e2d\u6837\u672c\u793a\u4f8b(EQ2)\u548c\u70ed\u56fe\u3002\u56fe23(a)\u663e\u793a\u4e86\u6765\u81ea\u6570\u636e\u7684\u793a\u4f8b\u56fe\u50cf\uff0c\u56fe23(b)\u663e\u793a\u4e86\u4eceVQA-X\u6570\u636e\u96c6\u4e2d\u63d0\u53d6\u7684\u56fe23(a)\u7684\u56e0\u679c\u7279\u5f81\u7684\u57fa\u672c\u4e8b\u5b9e\uff0c\u56fe23(c) \u663e\u793a\u4e86DCDAN\u4e3a\u56e0\u679c\u8868\u5f81\u751f\u6210\u7684\u70ed\u56fe<\/span><\/p>\n


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

\u4e3a\u4e86\u9a8c\u8bc1\u8be5\u6846\u67b6\u7684\u6709\u6548\u6027\uff0c\u4e9a\u5229\u6851\u90a3\u5dde\u7acb\u5927\u5b66<\/span>\uff08Arizona State University\uff0cASU\uff09<\/span>\u5218\u6b22\u8001\u5e08\u56e2\u961f\u5c06ResNet-50\u3001DDC\u3001DAN\u3001Deep CORAL\u3001DANN\u3001HAFN\u8bbe\u7f6e\u4e3a\u5bf9\u7167\u65b9\u6cd5\u6765\u8fdb\u884c\u5b9e\u9a8c\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe24\uff1aDCDAN\u548cResnet-50 \u5728VQA-X\u6570\u636e\u5b50\u96c6\u4e0a\u751f\u6210\u7684\u70ed\u56fe\u3002<\/span><\/p>\n


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

<\/p>\n

\u56fe25\uff1a\u5728\u4efb\u52a1\u4e0a\u5177\u6709\u4e0d\u540c\u8d85\u53c2\u6570\u3001\u548c\u7684DCDAN\u7684\u51c6\u786e\u6027\u3002<\/span><\/p>\n


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

\u5728\u5b9e\u9a8c\u4e2d\uff0cDCDAN\u5728\u8bb8\u591a\u60c5\u51b5\u4e0b\u4f18\u4e8e\u57fa\u7ebf\u65b9\u6cd5\uff0c\u7ed3\u679c\u8868\u660eDCDAN\u53ef\u4ee5\u6267\u884c\u65e0\u76d1\u7763\u7684\u57df\u81ea\u9002\u5e94\uff0c\u663e\u793a\u4e86\u5b83\u5728\u5b66\u4e60\u56e0\u679c\u8868\u793a\u65b9\u9762\u7684\u6709\u6548\u6027\u3002\u800c\u4e14\u8fd9\u8fd8\u9a8c\u8bc1\u4e86\u56e0\u679c\u7279\u5f81\u8868\u793a\u6709\u52a9\u4e8e\u5b66\u4e60\u8de8\u57df\u7684\u53ef\u8fc1\u79fb\u7279\u5f81\uff0c\u8fdb\u4e00\u6b65\u8bc1\u5b9e\u4e86\u56e0\u679c\u635f\u5931\u548c\u5206\u7c7b\u635f\u5931\u4e4b\u95f4\u7684\u826f\u597d\u6743\u8861\u53ef\u4ee5\u5bfc\u81f4\u5b66\u4e60\u66f4\u591a\u53ef\u8f6c\u79fb\u7684\u7279\u5f81\u3002<\/span><\/p>\n


<\/p>\n


<\/p>\n

\n
\n
\n

<\/section>\n

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

\u56db\u3001\u5c0f\u7ed3<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


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

\u672c\u7bc7\u63a8\u9001\u603b\u7ed3\u4e86\u7a33\u5b9a\u5b66\u4e60\u7684\u7814\u7a76\u8fdb\u5c55\uff0c\u4e3a\u8fd9\u4e2a\u7814\u7a76\u65b9\u5411\u7ed9\u51fa\u4e86\u4e00\u4e2a\u5927\u56fe\u666f\u3002\u603b\u7684\u6765\u8bf4\uff0c\u7a33\u5b9a\u5b66\u4e60\u53d7\u5230\u56e0\u679c\u63a8\u65ad\u7684\u542f\u53d1\uff0c\u8bd5\u56fe\u8bc6\u522b\u51fa\u5728\u4e0d\u540c\u73af\u5883\u4e0b\u90fd\u5177\u6709\u7a33\u5b9a\u6027\u7684\u9884\u6d4b\u53d8\u91cf\uff0c\u4ece\u800c\u514b\u670d\u865a\u5047\u76f8\u5173\u5e26\u6765\u7684\u5f71\u54cd\uff0c\u4f7f\u5f97\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5728\u4e0d\u540c\u73af\u5883\u4e0b\u7684\u8868\u73b0\u5728\u7406\u8bba\u4e0a\u5177\u6709\u7a33\u5b9a\u6027\u3002\u6700\u5f00\u59cb\u7684\u7a33\u5b9a\u5b66\u4e60\u65b9\u6cd5\u6539\u8fdb\u81ea\u56e0\u679c\u63a8\u65ad\u4e2d\u7684\u534f\u53d8\u91cf\u5e73\u8861\uff0c\u800c\u968f\u540e\u53d1\u5c55\u7684\u65b9\u6cd5\u8fd8\u5305\u62ec\u4e86\u7edf\u8ba1\u5b66\u4e60\u3001\u4f18\u5316\u8fc7\u7a0b\u7b49\u4e0d\u540c\u89c6\u89d2\uff0c\u56e0\u6b64\u8fdb\u4e00\u6b65\u4e30\u5bcc\u4e86\u7a33\u5b9a\u5b66\u4e60\u7684\u542b\u4e49\u3002\u7a33\u5b9a\u5b66\u4e60\u4e3a\u5f53\u4e0b\u673a\u5668\u5b66\u4e60\u548c\u9884\u6d4b\u5efa\u6a21\u63d0\u4f9b\u4e86\u4e00\u4e2a\u65b0\u7684\u5b66\u4e60\u8303\u5f0f\uff0c\u6765\u5f25\u8865\u76f8\u5173\u6027\u5b66\u4e60\u548c\u56e0\u679c\u5173\u7cfb\u4e4b\u95f4\u7684\u9e3f\u6c9f\uff0c\u4ece\u800c\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u514b\u670d\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5728\u7a33\u5b9a\u6027\u3001\u53ef\u89e3\u91ca\u6027\u548c\u516c\u5e73\u6027\u7b49\u65b9\u9762\u6240\u9762\u4e34\u7684\u98ce\u9669\u3002\u5728\u672a\u6765\uff0c\u7a33\u5b9a\u5b66\u4e60\u5c06\u7ee7\u7eed\u81f4\u529b\u4e8e\u5728\u4e0d\u727a\u7272\u9884\u6d4b\u80fd\u529b\u7684\u60c5\u51b5\u4e0b\u6784\u5efa\u66f4\u52a0\u53ef\u4fe1\u7684\u9884\u6d4b\u6a21\u578b\uff0c\u5e76\u5408\u7406\u5730\u653e\u677e\u5173\u4e8e\u56e0\u679c\u63a8\u65ad\u7684\u4e25\u683c\u5047\u8bbe\uff0c\u4ee5\u5339\u914d\u66f4\u591a\u73b0\u5b9e\u7684\u5e94\u7528\u573a\u666f\u3002<\/span><\/p>\n


<\/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

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

[1]. Peng, C. and A. Susan, Stable learning establishes some common ground between causal inference and machine learning. Nature machine intelligence, 2022. 4(2): p. 110-115.<\/span><\/p>\n

[2]. Kuang, K., et al. Stable Prediction across Unknown Environments. in ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2018.<\/span><\/p>\n

[3]. Shen, Z., et al. Stable Learning via Differentiated Variable Decorrelation. in 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: ACM.<\/span><\/p>\n

[4]. Shen, Z., et al., Stable Learning via Sample Reweighting. AAAI2020, 2019.<\/span><\/p>\n

[5]. Kuang, K., et al. Stable Prediction with Model Misspecification and Agnostic Distribution Shift. in Association for the Advancement of Artificial Intelligence. 2020.<\/span><\/p>\n

[6]. Liu, J., et al. Stable Adversarial Learning under Distributional Shifts. in Association for the Advancement of Artificial Intelligence, 2021. 2020.<\/span><\/p>\n

[7]. Liu, J., et al. Heterogeneous Risk Minimization. in International Conference on Machine Learning. 2021.<\/span><\/p>\n

[8]. Liu, J., et al. Kernelized Heterogeneous Risk Minimization. in 35th Conference on Neural Information Processing Systems (NeurIPS 2021). 2021.<\/span><\/p>\n

[9]. Xu, R., et al. Algorithmic Decision Making with Conditional Fairness. in ACM Conferences. 2020: ACM.<\/span><\/p>\n

[10]. Zhang, X., et al. Deep Stable Learning for Out-Of-Distribution Generalization. in 2021 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021.<\/span><\/p>\n

[11]. He, Y., et al. Learning Stable Graphs from Multiple Environments with Selection Bias. in ACM Conferences. 2020: ACM.<\/span><\/p>\n

[12]. Zhang, S., et al., Stable Prediction on Graphs with Agnostic Distribution Shift. 2021.<\/span><\/p>\n

[13]. Xu, R., et al., Why Stable Learning Works? A Theory of Covariate Shift Generalization. 2021.<\/span><\/p>\n

[14]. Moraffah, R., et al., Deep causal representation learning for unsupervised domain adaptation. 2019.<\/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<\/span><\/p>\n<\/section>\n<\/section>\n


<\/section>\n

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

\u56e0\u679c\u6d3e\u8bba\u575b\uff1a<\/strong><\/p>\n

\u56e0\u679c\u9769\u547d\u2014\u2014\u4e0b\u4e00\u4ee3\u53ef\u4fe1AI<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u96c6\u667a\u4ff1\u4e50\u90e8\u643a\u624b\u96f6\u7280\u79d1\u6280\u4e3a\u4fc3\u8fdb\u56e0\u679c\u79d1\u5b66\u5728\u4ea7\u4e1a\u843d\u5730\uff0c\u63a8\u51fa\u4e86\u56e0\u679c\u79d1\u5b66\u5b50\u54c1\u724c\u2014\u2014\u56e0\u679c\u6d3e\u3002\u4e3a\u52a0\u901f\u5b66\u754c\u548c\u4ea7\u4e1a\u754c\u5728\u56e0\u679c\u79d1\u5b66\u65b9\u9762\u7684\u878d\u5408\uff0c\u5145\u5206\u53d1\u6325\u5404\u81ea\u4f18\u52bf\uff0c\u7528\u79d1\u6280\u52a9\u529b\u7ecf\u6d4e\u793e\u4f1a\u53d1\u5c55\uff0c\u56e0\u679c\u6d3e\u4e5f\u5c06\u5c55\u5f00\u7cfb\u5217\u4e3b\u9898\u6d3b\u52a8\u3002<\/span><\/p>\n


<\/p>\n

7\u67082\u65e5\u7684\u56e0\u679c\u6d3e\u8bba\u575b\u9080\u8bf7\u5230\u4e86\u6e05\u534e\u5927\u5b66\u5d14\u9e4f\u8001\u5e08\u3001\u91cd\u5e86\u5927\u5b66\u5218\u793c\u8001\u5e08\u4ee5\u53ca\u96f6\u7280\u79d1\u6280\u9996\u5e2d\u79d1\u5b66\u5bb6\u5f20\u529b\u8fdb\u884c\u76f8\u5173\u4e3b\u9898\u62a5\u544a\uff0c\u63a2\u8ba8\u667a\u80fd\u670d\u52a1\u4e2d\u6240\u9762\u4e34\u7684\u56e0\u679c\u95ee\u9898\uff0c\u4ee5\u53ca\u5f53\u524d\u5b66\u672f\u754c\u524d\u6cbf\u7684\u5e94\u5bf9\u65b9\u6cd5\uff0c\u6b22\u8fce\u5404\u4f4d\u5b66\u8005\u3001\u4e1a\u754c\u7684\u4ece\u4e1a\u8005\u7b49\u52a0\u5165\u8ba8\u8bba\uff01<\/span><\/p>\n


<\/p>\n

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


<\/section>\n


<\/p>\n

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

\u56e0\u679c\u6d3e\u8bba\u575b\uff1a\u56e0\u679c\u9769\u547d\u2014\u2014\u4e0b\u4e00\u4ee3\u53ef\u4fe1AI | \u96f6\u7280\u79d1\u6280\u00d7\u96c6\u667a\u4ff1\u4e50\u90e8<\/strong><\/a><\/span>
<\/span><\/p>\n

\n


<\/p>\n

\u52a0\u5165\u96c6\u667a\u56e0\u679c\u79d1\u5b66\u793e\u533a\uff1a<\/span><\/p>\n

\u56e0\u679c+X\uff1a\u89e3\u51b3\u591a\u5b66\u79d1\u9886\u57df\u7684\u56e0\u679c\u95ee\u9898 | \u56e0\u679c\u79d1\u5b66\u8bfb\u4e66\u4f1a\u7b2c\u4e09\u5b63\u542f\u52a8<\/strong><\/a>
<\/p>\n


<\/p>\n


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

\n

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