{"id":27714,"date":"2021-07-23T18:47:19","date_gmt":"2021-07-23T10:47:19","guid":{"rendered":"https:\/\/swarma.org\/?p=27714"},"modified":"2021-07-23T18:47:19","modified_gmt":"2021-07-23T10:47:19","slug":"%e8%80%bf%e7%9b%b4%ef%bc%9a%e5%9b%a0%e6%9e%9c%e4%bd%9c%e7%94%a8%e8%af%84%e4%bb%b7%e4%b8%8e%e5%9b%a0%e6%9e%9c%e7%bd%91%e7%bb%9c%e5%ad%a6%e4%b9%a0%e5%8f%8a%e5%85%b6%e7%bb%93%e5%90%88","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=27714","title":{"rendered":"\u803f\u76f4\uff1a\u56e0\u679c\u4f5c\u7528\u8bc4\u4ef7\u4e0e\u56e0\u679c\u7f51\u7edc\u5b66\u4e60\u53ca\u5176\u7ed3\u5408"},"content":{"rendered":"
\n
\n
\n

<\/p>\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

\u56e0\u679c\u63a8\u65ad\u672c\u8d28\u662f\u6d88\u9664\u6df7\u6742\u56e0\u7d20\u5f15\u8d77\u7684\u865a\u5047\u76f8\u5173\u5173\u7cfb\u3002\u5176\u57fa\u672c\u4efb\u52a1\u5305\u62ec\uff1a\u5224\u65adPearl\u7684\u4e09\u4e2a\u5c42\u7ea7\u95ee\u9898\u3001\u6df7\u6742\u56e0\u7d20\u7684\u51c6\u5219\uff0c\u66ff\u4ee3\u6307\u6807\u6096\u8bba\uff0c\u56e0\u679c\u4f5c\u7528\u548c\u76f8\u5173\u5ea6\u91cf\u7684\u53ef\u4f20\u9012\u6027\u3001\u56e0\u679c\u7f51\u7edc\u7ed3\u6784\u7684\u5206\u89e3\u5b66\u4e60\u3001\u4e3b\u52a8\u5b66\u4e60\u548c\u5c40\u90e8\u5b66\u4e60\u65b9\u6cd5\u3002\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u56e0\u679c\u4f5c\u7528\u8bc4\u4ef7\u4e0e\u56e0\u679c\u7f51\u7edc\u5c40\u90e8\u7ed3\u6784\u5b66\u4e60\u7684\u7ed3\u5408\u6b63\u5728\u5d2d\u9732\u5934\u89d2\u3002<\/strong><\/p>\n


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

7\u670824\u65e5\uff08\u5468\u516d\uff0910:00-12:00\uff0c\u6211\u4eec\u7279\u522b\u9080\u8bf7\u5230\u5317\u4eac\u5927\u5b66\u6570\u5b66\u79d1\u5b66\u5b66\u9662\u7684\u803f\u76f4\u6559\u6388\u5f00\u5c55\u8bb2\u5ea7\uff0c\u4ecb\u7ecd\u56e0\u679c\u4f5c\u7528\u8bc4\u4ef7\u4e0e\u56e0\u679c\u5b66\u4e60\u7f51\u7edc\u76f8\u5173\u7814\u7a76\u3002\u96c6\u667a\u4ff1\u4e50\u90e8\u56e0\u679c\u793e\u533a<\/a><\/strong>\u6c47\u96c6\u4e86\u8d85\u8fc7600\u4f4d\u56e0\u679c\u79d1\u5b66\u76f8\u5173\u9886\u57df\u7814\u7a76\u8005\uff0c\u7ec4\u7ec7\u5404\u7c7b\u7ebf\u4e0a\u7ebf\u4e0b\u6d3b\u52a8\uff0c\u803f\u76f4\u8001\u5e08\u662f\u56e0\u679c\u793e\u533a\u6210\u5458\u3002\u672c\u6b21\u6d3b\u52a8\u4e3a\u56e0\u679c\u79d1\u5b66\u793e\u533a\u5185\u90e8\u7684\u7ebf\u4e0b\u6d3b\u52a8\uff0c\u6211\u4eec\u540c\u6b65\u5728\u7ebf\u4e0a\u76f4\u64ad\uff0c\u6587\u672b\u62a5\u540d\u5373\u53ef\u83b7\u5f97\u76f4\u64ad\u5730\u5740\uff0c\u6b22\u8fce\u5927\u5bb6\u53c2\u4e0e\u3002<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

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


<\/p>\n

<\/p>\n

\u803f\u76f4<\/span><\/strong>\uff0c<\/span>\u5317\u4eac\u5927\u5b66\u6570\u5b66\u79d1\u5b66\u5b66\u9662\u6559\u6388\u3002<\/span>1982\u5e74\u83b7\u4e0a\u6d77\u4ea4\u901a\u5927\u5b66\u5b66\u58eb\u5b66\u4f4d\uff0c1989\u5e74\u83b7\u65e5\u672c\u4e5d\u5dde\u5927\u5b66\u535a\u58eb\u5b66\u4f4d\u3002<\/span>1989\u5e74\u81f3\u4eca\u5728\u5317\u4eac\u5927\u5b66\u6982\u7387\u7edf\u8ba1\u7cfb\u4efb\u6559\u3002<\/span>\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u4e3a\u56e0\u679c\u63a8\u65ad\u3001\u6570\u7406\u7edf\u8ba1\u3001\u751f\u7269\u533b\u5b66\u7edf\u8ba1\u3001\u56e0\u679c\u7f51\u7edc\u3001\u8d1d\u53f6\u65af\u7f51\u7edc\u3001\u56fe\u6a21\u578b\u7b49\u3002<\/span>1998\u5e74\u5f53\u9009\u56fd\u9645\u7edf\u8ba1\u5b66\u4f1a\u63a8\u9009\u4f1a\u5458\uff0c1998\u5e74\u83b7\u56fd\u5bb6\u6770\u51fa\u9752\u5e74\u57fa\u91d1\u8d44\u52a9\u9879\u76ee\u3002<\/span>\u66fe\u4efb\u4e2d\u56fd\u6570\u5b66\u4f1a\u6982\u7387\u7edf\u8ba1\u5b66\u4f1a\u7406\u4e8b\u957f\u3001IMS-China\u4e3b\u5e2d\u3001\u56fd\u52a1\u9662\u5b66\u4f4d\u59d4\u5458\u4f1a\u7edf\u8ba1\u5b66\u5b66\u79d1\u8bc4\u8bae\u7ec4\u6210\u5458\u7b49\u804c\u52a1\u3002<\/span>\u8363\u83b7\u56fd\u5bb6\u6559\u59d4\u79d1\u6280\u8fdb\u6b65\u5956\u4e8c\u7b49\u5956\u3001\u5168\u56fd\u7edf\u8ba1\u79d1\u5b66\u7814\u7a76\u4f18\u79c0\u6210\u679c\u5956\u4e00\u7b49\u5956\u7b49\u591a\u9879\u5956\u52b1\u3002<\/span><\/section>\n
<\/span><\/section>\n


<\/p>\n


<\/p>\n

\u56e0\u679c\u63a8\u65ad\u65e8\u5728\u5229\u7528\u8bd5\u9a8c\u6027\u7814\u7a76\u548c\u89c2\u5bdf\u6027\u7814\u7a76\u7684\u6570\u636e\uff0c\u8bc4\u4ef7\u53d8\u91cf\u4e4b\u95f4\u7684\u56e0\u679c\u4f5c\u7528\u4ee5\u53ca\u8bc6\u522b\u56e0\u679c\u7f51\u7edc\u7684\u7ed3\u6784\u3002\u5728\u8bc4\u4f30\u56e0\u679c\u6548\u5e94\u65f6\uff0c\u7814\u7a76\u4eba\u5458\u5e94\u5bf9\u6df7\u6742\u56e0\u7d20\u8fdb\u884c\u8bc6\u522b\u548c\u8c03\u6574\u3002\u57fa\u4e8e\u5206\u914d\u673a\u5236\u7684\u53ef\u5ffd\u7565\u6027\u5047\u8bbe\uff0c\u4f8b\u5982\u5339\u914d\u65b9\u6cd5\u3001\u4fee\u526a\u6cd5\u3001\u9006\u6982\u52a0\u6743\u548c\u56de\u5f52\u65b9\u6cd5\u3001\u53cc\u7a33\u5065\u4f30\u8ba1\u65b9\u6cd5\u7b49\u53ef\u4ee5\u5b9e\u73b0\u5bf9\u6df7\u6742\u56e0\u7d20\u7684\u8c03\u6574\u3002\u5bf9\u4e8e\u672a\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\uff0c\u5de5\u5177\u53d8\u91cf\u65b9\u6cd5\u548c\u9634\u6027\u5bf9\u7167\u53d8\u91cf\u65b9\u6cd5\u53ef\u4ee5\u6709\u6548\u6d88\u9664\u4f30\u8ba1\u504f\u501a\u3002\u5f53\u611f\u5174\u8da3\u7684\u7ed3\u5c40\u53d8\u91cf\u96be\u4ee5\u89c2\u6d4b\u65f6\uff0c\u7814\u7a76\u4eba\u5458\u5e38\u5e38\u9009\u62e9\u89c2\u6d4b\u66ff\u4ee3\u6307\u6807\uff0c\u66ff\u4ee3\u6307\u6807\u51c6\u5219\u7684\u5efa\u7acb\u53ef\u4ee5\u907f\u514d\u66ff\u4ee3\u6307\u6807\u6096\u8bba\u73b0\u8c61\u3002\u5728\u56e0\u679c\u7f51\u7edc\u6a21\u578b\u4e2d\uff0c\u56e0\u679c\u4f5c\u7528\u7684\u53ef\u8bc6\u522b\u6027\u4e0e\u56e0\u679c\u7f51\u7edc\u7684\u5b66\u4e60\u7b97\u6cd5\u5df2\u6210\u4e3a\u8be5\u9886\u57df\u7684\u7814\u7a76\u70ed\u70b9\u3002<\/span><\/p>\n

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


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

\n
\n
\n
\n
\n

<\/section>\n

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

1. \u6df7\u6742\u56e0\u7d20\u7684\u8bc6\u522b\u53ca\u8c03\u6574\u65b9\u6cd5<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

\n
\n
\n

\u4ec0\u4e48\u662f\u6df7\u6742\u56e0\u7d20<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

Yule-Simpson\u6096\u8bba\u8868\u660e\uff0c\u5728\u56e0\u679c\u6548\u5e94\u8bc4\u4f30\u4e2d\u6709\u5fc5\u8981\u5bf9\u6df7\u6742\u56e0\u7d20\u8fdb\u884c\u8bc6\u522b\u548c\u8c03\u6574\u3002<\/span>\u57fa\u4e8e\u6d41\u884c\u75c5\u5b66\u7814\u7a76\u4e2d\u7684\u5b9e\u4f8b\uff0cMiettinen\u548cCook\u8ba4\u4e3a\u6df7\u6742\u56e0\u7d20V\u5fc5\u987b\u6ee1\u8db3\u4ee5\u4e0b\u4e24\u4e2a\u6761\u4ef6\uff1a<\/span>\u5176\u4e00\uff0c\u5b83\u53ef\u4ee5\u9884\u6d4b\u672a\u66b4\u9732\u4eba\u7fa4\u7684\u98ce\u9669\uff1b<\/span>\u5176\u4e8c\uff0c\u5b83\u5728\u66b4\u9732\u4eba\u7fa4\u548c\u672a\u66b4\u9732\u4eba\u7fa4\u4e2d\u7684\u5206\u5e03\u4e0d\u540c\u3002<\/span><\/p>\n


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

\u57fa\u4e8e\u6b64\uff0c\u540e\u7eed\u7684\u6587\u732e\u4e2d\u4f7f\u7528\u4e24\u7c7b\u6807\u51c6\u5bf9\u6df7\u6742\u56e0\u7d20\u8fdb\u884c\u4e25\u683c\u7684\u5b9a\u4e49\uff1a<\/span><\/p>\n


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

1. \u53ef\u6bd4\u8f83\u6027\u51c6\u5219<\/span> (The comparability-based criterion)<\/span>\uff1a\u5982\u679c\u66b4\u9732\u4eba\u7fa4\u548c\u672a\u66b4\u9732\u4eba\u7fa4\u7684\u6f5c\u5728\u7ed3\u679c\u5206\u5e03\u56e0\u5ffd\u7565\u534f\u53d8\u91cf\u800c\u4e0d\u540c\uff0c\u5219\u8be5\u534f\u53d8\u91cf\u662f\u6df7\u6742\u56e0\u7d20\uff1b<\/span><\/p>\n


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

2. \u53ef\u538b\u7f29\u6027\u51c6\u5219<\/span> (The collapsibility-based criterion)<\/span>\uff1a\u5982\u679c\u5173\u8054\u5ea6\u91cf\u6216\u611f\u5174\u8da3\u7684\u53c2\u6570\u53d7\u5230\u5ffd\u7565\u534f\u53d8\u91cf\u7684\u5f71\u54cd(\u5373\u5728\u534f\u53d8\u91cf\u4e0a\u4e0d\u53ef\u6298\u53e0)\uff0c\u5219\u8be5\u534f\u53d8\u91cf\u662f\u6df7\u6742\u56e0\u7d20\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

\u5206\u914d\u673a\u5236\u7684\u53ef\u5ffd\u7565\u6027\u5047\u8bbe<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u5728\u89c2\u5bdf\u6027\u7814\u7a76\u4e2d\uff0c\u5224\u65ad\u6df7\u6742\u56e0\u7d20\u548c\u63a8\u65ad\u56e0\u679c\u4f5c\u7528\u7684\u4e00\u4e2a\u91cd\u8981\u5047\u5b9a\u662f\u5904\u7406\u5206\u914d\u673a\u5236\u7684\u53ef\u5ffd\u7565\u6027<\/span> (ignorability of treatment assignment mechanism)<\/span>\u3002\u4ee4 V \u8868\u793a\u89c2\u6d4b\u7684\u534f\u53d8\u91cf\uff0c\u5982\u679c\u6ee1\u8db3 (i) (Y1<\/sub>, Y0<\/sub>)\u548cX\u5728V\u7684\u6761\u4ef6\u4e0b\u72ec\u7acb \u548c (ii) 0 < pr(X = 1|V ) < 1\uff0c\u90a3\u4e48\u79f0\u5904\u7406\u5206\u914d\u673a\u5236\u662f\u53ef\u5ffd\u7565\u7684\u3002\u5728\u53ef\u5ffd\u7565\u6027\u5047\u8bbe\u4e2d\uff0c\u6bcf\u4e00\u5c42\u5747\u53ef\u88ab\u89c6\u4e3a\u662f\u968f\u673a\u5bf9\u7167\u8bd5\u9a8c\uff0c\u4ece\u800c\u5728\u6bcf\u4e00\u5c42\u90fd\u80fd\u5f97\u5230\u8be5\u5c42\u5e73\u5747\u56e0\u679c\u4f5c\u7528\u7684\u76f8\u5408\u4f30\u8ba1\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

\u57fa\u4e8e\u53ef\u5ffd\u7565\u6027\u5047\u8bbe\u7684\u6df7\u6dc6\u56e0\u7d20\u8c03\u6574\u65b9\u6cd5<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u5339\u914d\u65b9\u6cd5<\/span><\/strong><\/h5>\n


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

\u4e3a\u4e86\u6d88\u9664\u534f\u53d8\u91cf\u7684\u5206\u5e03\u5728\u5904\u7406\u7ec4\u4e0e\u5bf9\u7167\u7ec4\u4e4b\u95f4\u7684\u5dee\u5f02\uff0c\u5339\u914d<\/span> (matching)<\/span> \u65b9\u6cd5\u7ecf\u5e38\u7528\u5728\u89c2\u5bdf\u6027\u7814\u7a76\u4e2d\u3002\u5339\u914d\u65b9\u6cd5\u7684\u76ee\u7684\u662f\u5bf9\u6bcf\u4e00\u4e2a\u4e2a\u4f53\u5339\u914d\u4e00\u4e2a\u5177\u6709\u76f8\u540c\u6216\u76f8\u8fd1\u534f\u53d8\u91cf\u53d6\u503c\u7684\u4e2a\u4f53\u96c6\u5408\uff0c\u4f7f\u5f97\u5339\u914d\u5f97\u5230\u7684\u6570\u636e\u5728\u5904\u7406\u7ec4\u548c\u5bf9\u7167\u7ec4\u6709\u76f8\u540c\u7684\u534f\u53d8\u91cf\u5206\u5e03\uff0c\u7136\u540e\u6839\u636e\u5339\u914d\u6570\u636e\u63a8\u65ad\u56e0\u679c\u4f5c\u7528\u3002\u9a6c\u6c0f\u5ea6\u91cf\u5339\u914d\u8bd5\u56fe\u76f4\u63a5\u5e73\u8861\u6240\u6709\u534f\u53d8\u91cf\uff0c\u800c\u503e\u5411\u5f97\u5206\u5339\u914d <\/span>(propensity score matching) <\/span>\u53ef\u4ee5\u4f9d\u636e\u4e00\u7ef4\u7684\u503e\u5411\u5f97\u5206\u6784\u9020\u5339\u914d\u96c6\u5408\uff0c\u4ee5\u5b9e\u73b0\u5bf9\u968f\u673a\u5316\u7684\u6a21\u62df\uff0c\u76ee\u524d\u662f\u89c2\u5bdf\u6027\u7814\u7a76\u4e2d\u5e38\u4f7f\u7528\u7684\u5339\u914d\u65b9\u6cd5\u3002<\/span><\/p>\n


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

\u9006\u6982\u52a0\u6743\u548c\u56de\u5f52\u65b9\u6cd5<\/span><\/strong><\/span><\/h5>\n


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

\u6b64\u5916\uff0c\u7ed9\u5b9a\u53ef\u5ffd\u7565\u6027\u5047\u5b9a\u548c\u4e00\u5b9a\u7684\u6b63\u5219\u6761\u4ef6\uff0c\u9006\u6982\u52a0\u6743\u4f30\u8ba1<\/span> (inverse probability weighted estimation) <\/span>\u548c\u56de\u5f52\u4f30\u8ba1<\/span> (regression-based estimator) <\/span>\u5206\u522b\u901a\u8fc7\u62df\u5408\u4e00\u4e2a\u503e\u5411\u5f97\u5206\u6a21\u578b\u548c\u5efa\u7acb\u4e00\u4e2a\u5bf9\u7ed3\u679c\u53d8\u91cf\u7684\u56de\u5f52\u6a21\u578b\u5b9e\u73b0\u5bf9\u5e73\u5747\u56e0\u679c\u6548\u5e94\u7684\u4f30\u8ba1\u3002\u5982\u679c\u56de\u5f52\u6a21\u578b\u548c\u503e\u5411\u5f97\u5206\u6a21\u578b\u5206\u522b\u6b63\u786e\uff0c\u5219\u53ef\u4ee5\u901a\u8fc7\u5e7f\u4e49\u77e9\u4f30\u8ba1\u65b9\u6cd5\u5b9e\u73b0\u5e73\u5747\u56e0\u679c\u6548\u5e94\u7684\u76f8\u5408\u4f30\u8ba1\uff0c\u4f46\u662f\u4e0a\u8ff0\u4e24\u79cd\u65b9\u6cd5\u5bf9\u6a21\u578b\u7279\u522b\u654f\u611f\uff0c\u5f53\u6307\u5b9a\u7684\u56de\u5f52\u6a21\u578b\u6216\u8005\u503e\u5411\u5f97\u5206\u6a21\u578b\u4e0d\u6b63\u786e\u65f6\uff0c\u5f80\u5f80\u4f1a\u5bfc\u81f4\u8f83\u5927\u504f\u5dee\u3002<\/span><\/p>\n


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

\u53cc\u7a33\u5065\u4f30\u8ba1\u65b9\u6cd5<\/span><\/strong><\/span><\/h5>\n


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

Robins\u5728\u7814\u7a76\u7f3a\u5931\u6570\u636e\u65f6\u63d0\u51fa\u4e86\u53cc\u7a33\u5065\u4f30\u8ba1\u65b9\u6cd5<\/span> (doubly robust estimation)<\/span>\u3002\u8fd9\u79cd\u65b9\u6cd5\u628a\u56de\u5f52\u4f30\u8ba1\u548c\u9006\u6982\u52a0\u6743\u4f30\u8ba1\u7ed3\u5408\u8d77\u6765\uff0c\u5e76\u5177\u6709\u53cc\u7a33\u5065\u6027\u8d28\uff1a\u53ea\u8981\u56de\u5f52\u6a21\u578b\u548c\u503e\u5411\u5f97\u5206\u6a21\u578b\u4e2d\u7684\u4e00\u4e2a\u6a21\u578b\u6b63\u786e\uff0c\u90a3\u4e48\u53cc\u7a33\u5065\u4f30\u8ba1\u5c31\u6709\u76f8\u5408\u6027\u3002\u76f8\u6bd4\u4e8e\u56de\u5f52\u4f30\u8ba1\u548c\u503e\u5411\u5f97\u5206\u4f30\u8ba1\uff0c\u53cc\u7a33\u5065\u4f30\u8ba1\u63d0\u4f9b\u4e86\u66f4\u591a\u51cf\u5c11\u4f30\u8ba1\u504f\u5dee\u7684\u673a\u4f1a\u3002\u7531\u4e8e\u53cc\u7a33\u5065\u4f30\u8ba1\u80fd\u6709\u6548\u5730\u51cf\u5c0f\u6a21\u578b\u9519\u8bef\u5bfc\u81f4\u7684\u504f\u5dee\uff0c\u8fd9\u79cd\u65b9\u6cd5\u8d8a\u6765\u8d8a\u5e7f\u6cdb\u5e94\u7528\u5728\u7f3a\u5931\u6570\u636e\u5206\u6790\u548c\u56e0\u679c\u63a8\u65ad\u4e2d\u3002<\/span><\/p>\n


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

\n
\n
\n

\u672a\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\u8c03\u6574\u65b9\u6cd5<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u5728\u5b9e\u9645\u7814\u7a76\u4e2d\uff0c\u5982\u679c\u6709\u91cd\u8981\u80cc\u666f\u53d8\u91cf\u672a\u88ab\u89c2\u6d4b\u3001\u6d4b\u91cf\u8bef\u5dee\u6216\u8005\u9009\u62e9\u504f\u5dee\uff0c\u5c31\u6709\u6f5c\u5728\u7684\u672a\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\uff0c\u53ef\u5ffd\u7565\u6027\u5047\u5b9a\u53ef\u80fd\u4e0d\u6210\u7acb\uff0c\u524d\u4e00\u8282\u4ecb\u7ecd\u7684\u7edf\u8ba1\u63a8\u65ad\u65b9\u6cd5\u5728\u51fa\u73b0\u672a\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\u65f6\u5c31\u6709\u504f\u5dee\u3002<\/span>\u5f53\u5b58\u5728\u672a\u88ab\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\u65f6\uff0c\u66f4\u5408\u7406\u7684\u5047\u5b9a\u662f\u6f5c\u5728\u53ef\u5ffd\u7565\u6027\uff1a<\/span>\u5b58\u5728\u672a\u88ab\u89c2\u6d4b\u7684\u53d8\u91cfU\u6ee1\u8db3<\/span>Y<\/span>x<\/sub>\u548cX\u5728 (U, V )\u7684\u6761\u4ef6\u4e0b\u72ec\u7acb\uff0c\u5176\u4e2d V \u4e3a\u89c2\u6d4b\u7684\u6df7\u6742\u56e0\u7d20\u3002<\/span><\/span><\/p>\n


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

\u5728\u6f5c\u5728\u53ef\u5ffd\u7565\u6027\u5047\u5b9a\u4e0b\uff0c\u8f85\u52a9\u53d8\u91cf\u7ecf\u5e38\u88ab\u7528\u6765\u5e2e\u52a9\u8bc6\u522b\u56e0\u679c\u4f5c\u7528\u548c\u6d88\u9664\u6df7\u6742\u504f\u501a\u3002\u8f85\u52a9\u53d8\u91cf\u901a\u5e38\u53ea\u4e0e (X, Y, U) \u4e09\u4e2a\u53d8\u91cf\u7684\u4e00\u4e2a\u5b50\u96c6\u76f8\u5173\uff0c\u56e0\u6b64\u5f15\u5165\u4e00\u4e9b\u6761\u4ef6\u72ec\u7acb\u6027\u5e2e\u52a9\u8bc6\u522b\u56e0\u679c\u4f5c\u7528\u3002\u672c\u8282\u4ecb\u7ecd\u5728\u6f5c\u5728\u53ef\u5ffd\u7565\u6027\u5047\u5b9a\u4e0b\u7528\u6765\u6d88\u9664\u6df7\u6742\u504f\u5dee\u7684\u4e24\u79cd\u65b9\u6cd5\uff0c\u4e00\u79cd\u662f\u5e38\u7528\u7684\u5de5\u5177\u53d8\u91cf<\/span> (instrumental variable) <\/span>\u65b9\u6cd5\uff0c\u53e6\u4e00\u79cd\u662f\u6700\u8fd1\u5f15\u8d77\u4eba\u4eec\u6ce8\u610f\u7684\u9634\u6027\u5bf9\u7167\u53d8\u91cf<\/span> (negative control variable) <\/span>\u65b9\u6cd5\u3002<\/span><\/p>\n


<\/p>\n


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

\n
\n
\n
\n
\n

<\/section>\n

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

2. \u66ff\u4ee3\u6307\u6807\u6096\u8bba\u4e0e\u66ff\u4ee3\u6307\u6807\u51c6\u5219<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

\n
\n
\n

\u66ff\u4ee3\u6307\u6807\u6096\u8bba<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u5728\u79d1\u5b66\u7814\u7a76\u4e2d\uff0c\u7279\u522b\u662f\u5728\u533b\u5b66\u548c\u751f\u7269\u5b66\u8bd5\u9a8c\u4e2d\uff0c\u5f53\u611f\u5174\u8da3\u7684\u7ec8\u70b9\u6307\u6807<\/span> (endpoint) <\/span>\u96be\u4ee5\u89c2\u6d4b\u65f6\uff0c\u5e38\u4f1a\u53d6\u800c\u4ee3\u4e4b\u89c2\u6d4b\u66ff\u4ee3\u6307\u6807<\/span> (surrogate) <\/span>\u6216\u6807\u8bb0\u7269<\/span> (marker)<\/span>\u3002\u66ff\u4ee3\u6307\u6807\u6096\u8bba <\/span>(surrogate paradox)<\/span> \u6307\u51fa\uff0c\u5f53\u5904\u7406<\/span> (\u6216\u79f0\u6cbb\u7597) <\/span>\u5bf9\u66ff\u4ee3\u6307\u6807\u6709\u6b63\u7684\u56e0\u679c\u4f5c\u7528\uff0c\u5e76\u4e14\u66ff\u4ee3\u6307\u6807\u5bf9\u7ec8\u70b9\u6307\u6807\u4e5f\u6709\u6b63\u7684\u56e0\u679c\u4f5c\u7528,\u8be5\u5904\u7406\u5bf9\u7ec8\u70b9\u6307\u6807\u4e5f\u8bb8\u4f1a\u4ea7\u751f\u8d1f\u7684\u56e0\u679c\u4f5c\u7528\u3002\u4f8b\u5982\uff0c\u4efb\u4f55\u4e00\u4e2a\u4eba\u5047\u82e5\u5fc3\u5f8b\u6b63\u5e38\u4e00\u5b9a\u6bd4\u5fc3\u5f8b\u4e0d\u6b63\u5e38\u6d3b\u5f97\u66f4\u957f\uff0c\u67d0\u79cd\u836f\u53ef\u4ee5\u663e\u8457\u7ea0\u6b63\u5fc3\u5f8b\u5931\u5e38\uff0c\u4f46\u662f\u8fd9\u79cd\u836f\u53cd\u800c\u51cf\u5c11\u60a3\u8005\u7684\u5bff\u547d\u3002\u56e0\u6b64\uff0c\u5de5\u5177\u53d8\u91cf\u4f30\u8ba1\u53ef\u80fd\u51fa\u73b0\u6b63\u8d1f\u7b26\u53f7\u7684\u6096\u8bba\u73b0\u8c61\uff0c\u5229\u7528\u4e2d\u95f4\u53d8\u91cf\u7684\u7edf\u8ba1\u7ed3\u8bba\u4e0d\u5177\u6709\u4f20\u9012\u6027\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

\u66ff\u4ee3\u6307\u6807\u51c6\u5219<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u4e3a\u4e86\u907f\u514d\u66ff\u4ee3\u6307\u6807\u6096\u8bba\u73b0\u8c61\uff0c\u5e94\u5efa\u7acb\u66ff\u4ee3\u6307\u6807\u51c6\u5219<\/span> (criteria for consistent surrogates) <\/span>\uff0c\u5373\u4e0d\u5fc5\u89c2\u6d4b\u7ec8\u70b9\u6307\u6807Y\uff0c\u800c\u53ea\u9700\u89c2\u6d4b\u66ff\u4ee3\u6307\u6807S\uff0c\u5c31\u53ef\u4ee5\u7528\u5904\u7406 X\u5bf9\u89c2\u6d4b\u7684\u66ff\u4ee3\u6307\u6807 S \u7684\u56e0\u679c\u4f5c\u7528\u7684\u6b63\u8d1f\u7b26\u53f7\u6765\u9884\u6d4b\u5904\u7406 X \u5bf9\u672a\u89c2\u6d4b\u7684\u7ec8\u70b9\u6307\u6807 Y \u7684\u56e0\u679c\u4f5c\u7528\u7684\u6b63\u8d1f\u7b26\u53f7\u3002<\/span><\/p>\n


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

\u66ff\u4ee3\u6307\u6807\u7684\u51c6\u5219\u53ef\u4ee5\u5b9a\u4e49\u4e3a\uff1a\u82e5\u7ed9\u5b9a\u4efb\u610f U = u \u7684\u6761\u4ef6\u4e0b\u66ff\u4ee3\u6307\u6807 S \u90fd\u662f\u7ec8\u70b9\u6307\u6807 Y \u7684\u4e00\u4e2a\u5371\u9669\u56e0\u7d20<\/span> (\u6216\u4fdd\u62a4\u56e0\u7d20)<\/span>\uff0c\u4e14\u7ed9\u5b9a\u4efb\u610fU = u\u7684\u6761\u4ef6\u4e0b\u5904\u7406 X \u5bf9\u66ff\u4ee3\u6307\u6807S\u6709\u76f8\u540c\u7b26\u53f7\u7684\u5206\u5e03\u56e0\u679c\u4f5c\u7528<\/span> (distributional causal effect, DCE)<\/span>\uff0c\u5219\u6839\u636eX\u5bf9S\u7684\u975e\u8d1f<\/span>\uff08\u975e\u6b63\uff09<\/span>DCE\u80fd\u9884\u6d4bX\u5bf9Y\u7684\u975e\u8d1f<\/span>\uff08\u975e\u6b63\uff09<\/span>\u5e73\u5747\u56e0\u679c\u4f5c\u7528<\/span> (average causal effect, ACE)<\/span>\uff0c\u5e76\u4e14\u6839\u636eX\u5bf9S\u7684\u96f6DCE\u80fd\u9884\u6d4bX\u5bf9Y\u7684\u96f6ACE\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

\u76f8\u5173\u5ea6\u91cf\u7684\u53ef\u4f20\u9012\u6027<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n


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

\u79f0\u4e09\u4e2a\u53d8\u91cf{X\uff0cY\uff0cZ}\u4e4b\u95f4\u7684\u5173\u8054\u5ea6\u91cf\u7684\u7b26\u53f7\u662f\u53ef\u4f20\u9012\u7684\uff0c\u5982\u679c\u53d8\u91cfX\u548c\u4e2d\u95f4\u53d8\u91cfY\u4e4b\u95f4\u7684\u6b63\u5173\u8054\u5ea6\u91cf\u4ee5\u53caY\u548c\u7ed3\u5c40\u53d8\u91cfZ\u4e4b\u95f4\u7684\u6b63\u5173\u8054\u5ea6\u91cf\u53ef\u4ee5\u8868\u660eX\u548cZ\u4e4b\u95f4\u7684\u6b63\u5173\u8054\u5ea6\u91cf\u3002\u5728\u201cQualitative evaluation of associations and validation of surrogacy by association transitivity\u201d\u4e00\u6587\u4e2d\uff0c\u901a\u8fc7\u5f15\u5165\u56db\u4e2a\u5177\u6709\u4e0d\u540c\u4e25\u683c\u6027\u7684\u5173\u8054\u5ea6\u91cf\uff1a\u5bc6\u5ea6\u5ea6\u91cf <\/span>(density association)<\/span>\uff0c\u5206\u5e03\u5ea6\u91cf<\/span> (distribution association)<\/span>\uff0c\u671f\u671b\u5ea6\u91cf<\/span> (expectation association)<\/span>\uff0c\u548c\u76f8\u5173\u6027\u5ea6\u91cf<\/span> (correlation coefficient)<\/span>\uff0c\u5e76\u8ba8\u8bba\u4e86\u8fd9\u4e9b\u5173\u8054\u5ea6\u91cf\u7684\u76f8\u5bf9\u4e25\u683c\u6027\u3002\u6b64\u5916\uff0c\u5728\u5047\u8bbeY\u9075\u5faa\u6307\u6570\u65cf\u5206\u5e03\u4e14Y\u548cZ\u4e4b\u95f4\u7684\u5173\u8054\u662f\u975e\u8d1f\u7684\u60c5\u51b5\u4e0b\uff0cX\u548cY\u4e4b\u95f4\u7684\u5173\u8054\u5ea6\u91cf\u7684\u7b26\u53f7\u4e0eX\u548cZ\u4e4b\u95f4\u7684\u5173\u8054\u5ea6\u91cf\u7684\u7b26\u53f7\u4e4b\u95f4\u7684\u5177\u6709\u7b49\u4ef7\u5173\u7cfb\u3002<\/span><\/p>\n

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


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

\n
\n
\n
\n
\n

<\/section>\n

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

3. \u56e0\u679c\u7f51\u7edc\u7684\u53ef\u8bc6\u522b\u6027\u4e0e\u5b66\u4e60\u7b97\u6cd5<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

Judea Pearl\u57fa\u4e8e\u8d1d\u53f6\u65af\u7f51\u7edc\u63d0\u51fa\u4e86\u5916\u90e8\u5e72\u9884\u7684\u6982\u5ff5\uff0c\u7528\u5916\u90e8\u5e72\u9884\u7684\u6982\u5ff5\u5bf9\u56e0\u679c\u5173\u7cfb\u7ed9\u51fa\u4e86\u4e00\u79cd\u5f62\u5f0f\u5316\u65b9\u6cd5\uff0c\u5efa\u7acb\u4e86\u56e0\u679c\u7f51\u7edc\u6a21\u578b\u3002<\/span>\u56e0\u679c\u7f51\u7edc\u63cf\u8ff0\u591a\u4e2a\u53d8\u91cf\u4e4b\u95f4\u76f8\u4e92\u7684\u56e0\u679c\u5173\u7cfb\uff0c\u7f51\u7edc\u56fe\u4e2d\u6bcf\u4e2a\u8282\u70b9\u8868\u793a\u4e00\u4e2a\u53d8\u91cf\uff0c\u8282\u70b9\u4e4b\u95f4\u7684\u6709\u5411\u7bad\u5934\u8868\u793a\u7531\u539f\u56e0\u5230\u7ed3\u679c\u7684\u56e0\u679c\u5173\u7cfb\uff0c\u6216\u8005\u53d8\u91cf\u4e4b\u95f4\u7684\u6570\u636e\u751f\u6210\u8fc7\u7a0b\u3002<\/span>
<\/p>\n

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

Pearl\u7684\u56e0\u679c\u5143\u6a21\u578b\u6d89\u53ca\u4e09\u4e2a\u5c42\u6b21\u7684\u62bd\u8c61\uff0c\u88ab\u79f0\u4e3a\u662f\u201c\u56e0\u679c\u4e4b\u68af\u201d\u3002\u7b2c\u4e00\u5c42\u7ea7\u662f\u89c2\u5bdf\u80fd\u529b\uff0c\u65e8\u5728\u53d1\u6398\u6570\u636e\u4e4b\u95f4\u7684\u76f8\u5173\u4e0e\u9884\u6d4b\u6027\uff1b\u7b2c\u4e8c\u5c42\u7ea7\u662f\u884c\u52a8\u80fd\u529b\uff0c\u5141\u8bb8\u7814\u7a76\u4eba\u5458\u8fdb\u884c\u56e0\u679c\u73b0\u8c61\u7684\u8ba8\u8bba\u4e0e\u51b3\u7b56\uff1b\u7b2c\u4e09\u5c42\u7ea7\u662f\u60f3\u8c61\u80fd\u529b\uff0c\u4f9d\u636e\u53cd\u4e8b\u5b9e\u63a8\u7406\u7684\u7ed3\u679c\uff0c\u53ef\u4ee5\u9884\u6d4b\u5728\u5c1a\u672a\u7ecf\u5386\u751a\u81f3\u672a\u66fe\u8bbe\u60f3\u8fc7\u7684\u60c5\u51b5\u4e0b\u4f1a\u53d1\u751f\u4ec0\u4e48\u3002\u6982\u7387\u80fd\u5c06\u6211\u4eec\u5bf9\u9759\u6001\u4e16\u754c\u7684\u4fe1\u5ff5\u8fdb\u884c\u7f16\u7801\uff0c\u800c\u56e0\u679c\u8bba\u5219\u544a\u8bc9\u6211\u4eec\uff0c\u65e0\u8bba\u6539\u53d8\u662f\u901a\u8fc7\u5e72\u9884\u8fd8\u662f\u901a\u8fc7\u60f3\u8c61\u5b9e\u73b0\u7684\uff0c\u6982\u7387\u662f\u5426\u4f1a\u53d1\u751f\u6539\u53d8\u4ee5\u53ca\u5982\u4f55\u53d1\u751f\u6539\u53d8\u3002<\/span><\/p>\n


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

\u5728\u56e0\u679c\u7f51\u7edc\u7684\u6846\u67b6\u4e0b\uff0c\u7814\u7a76\u4e24\u7c7b\u95ee\u9898\uff1a\u5176\u4e00\u662f\u56e0\u679c\u4f5c\u7528\u7684\u53ef\u8bc6\u522b\u6027\uff1b\u53e6\u4e00\u4e2a\u662f\u56e0\u679c\u7f51\u7edc\u7684\u5b66\u4e60\u3002\u8bc6\u522b\u56e0\u679c\u4f5c\u7528\u7684\u76ee\u7684\u4e0e\u6f5c\u5728\u7ed3\u679c\u7684\u56e0\u679c\u6a21\u578b\u7684\u76ee\u7684\u662f\u4e00\u81f4\u7684\uff0c\u5373\u63a2\u8ba8\u5224\u65ad\u6df7\u6742\u56e0\u7d20\u7684\u51c6\u5219\u548c\u7814\u7a76\u56e0\u679c\u4f5c\u7528\u7684\u53ef\u8bc6\u522b\u6027\u3002\u56e0\u679c\u7f51\u7edc\u7ed3\u6784\u5b66\u4e60\u6709\u4e24\u7c7b\u65b9\u6cd5\uff1a\u57fa\u4e8e\u8bc4\u5206\u7684\u641c\u7d22\u65b9\u6cd5\u548c\u57fa\u4e8e\u6761\u4ef6\u72ec\u7acb\u68c0\u9a8c\u7684\u65b9\u6cd5\u3002<\/span><\/p>\n


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

\n
\n
\n

\u5206\u89e3\u5b66\u4e60\u65b9\u6cd5<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u9488\u5bf9\u5927\u89c4\u6a21\u7f51\u7edc\u548c\u591a\u4e2a\u4e0d\u5b8c\u5168\u6570\u636e\u5e93\uff0c\u5206\u89e3\u5b66\u4e60\u65b9\u6cd5\u7684\u601d\u60f3\u662f\uff1a\u5728\u6709\u53d8\u91cf\u95f4\u6761\u4ef6\u72ec\u7acb\u7684\u5148\u9a8c\u77e5\u8bc6\u60c5\u5f62\u4e0b\u53ef\u4ee5\u4e0d\u5fc5\u540c\u65f6\u89c2\u6d4b\u6240\u6709\u7684\u53d8\u91cf\uff0c\u5229\u7528\u591a\u4e2a\u4e0d\u5b8c\u5168\u53d8\u91cf\u7684\u6570\u636e\u5e93\uff0c\u9996\u5148\u5404\u81ea\u5b66\u4e60\u5c40\u90e8\u7f51\u7edc\uff0c\u7136\u540e\u5408\u5e76\u4e3a\u4e00\u4e2a\u5b8c\u6574\u7684\u5927\u7f51\u7edc\u3002\u5728\u201cA recursive method for structural learning of directed acyclic graphs\u201d\u4e00\u6587\u4e2d\uff0c\u63d0\u51fa\u4e86\u7f51\u7edc\u7ed3\u6784\u7684\u9012\u5f52\u5b66\u4e60\u7b97\u6cd5\uff0c\u5c06\u4e00\u4e2a\u5927\u7f51\u7edc\u7ed3\u6784\u5b66\u4e60\u5206\u89e3\u4e3a\u4e24\u4e2a\u5c40\u90e8\u5c0f\u7f51\u7edc\u7684\u7ed3\u6784\u5b66\u4e60\uff0c\u7136\u540e\u9012\u5f52\u5730\u5c06\u6bcf\u4e2a\u5c40\u90e8\u7f51\u7edc\u7684\u7ed3\u6784\u5b66\u4e60\u518d\u5206\u89e3\u4e3a\u66f4\u5c0f\u7f51\u7edc\u7684\u7ed3\u6784\u5b66\u4e60\u3002<\/span><\/p>\n


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

\n
\n
\n

\u4e3b\u52a8\u5b66\u4e60\u65b9\u6cd5<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u7814\u7a76\u4eba\u5458\u65e0\u6cd5\u4ece\u89c2\u5bdf\u6570\u636e\u4e2d\u533a\u5206\u9a6c\u5c14\u53ef\u592b\u94fe\u4e2d\u4e0d\u540c\u7684\u6709\u5411\u65e0\u73af\u56fe\uff0c\u56e0\u6b64\u5fc5\u987b\u4ece\u5916\u90e8\u5e72\u9884\u7684\u5b9e\u9a8c\u4e2d\u6536\u96c6\u66f4\u591a\u5173\u4e8e\u56e0\u679c\u7ed3\u6784\u7684\u4fe1\u606f\u3002\u5728\u201cActive learning of causal networks with intervention experiments and optimal designs\u201d\u4e00\u6587\u4e2d\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u53d1\u73b0\u56e0\u679c\u7ed3\u6784\u7684\u4e3b\u52a8\u5b66\u4e60\u65b9\u6cd5\uff0c\u9996\u5148\u4ece\u89c2\u6d4b\u6570\u636e\u4e2d\u627e\u5230\u4e00\u4e2a\u9a6c\u5c14\u53ef\u592b\u7b49\u4ef7\u7c7b\uff0c\u7136\u540e\u901a\u8fc7\u5e72\u9884\u5b9e\u9a8c\u5206\u522b\u786e\u5b9a\u6bcf\u4e2a\u94fe\u6210\u5206\u4e2d\u7684\u65e0\u5411\u8fb9\u7684\u65b9\u5411\u3002<\/span><\/p>\n


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

\u5728\u5b9e\u9a8c\u4e2d\uff0c\u4e00\u4e9b\u53d8\u91cf\u901a\u8fc7\u5916\u90e8\u5e72\u9884\u6765\u64cd\u7eb5\u3002\u6700\u4f73\u5e72\u9884\u8bbe\u8ba1\u65e8\u5728\u4f7f\u7528\u6700\u5c11\u7684\u5e72\u9884\uff0c\u786e\u5b9a\u6240\u6709\u56e0\u679c\u65b9\u5411\u3002\u57fa\u4e8e\u6700\u5c0f\u6700\u5927\u51c6\u5219\u548c\u6700\u5927\u71b5\u51c6\u5219\uff0c\u8003\u8651\u4e24\u4e2a\u6700\u4f18\u5b9e\u9a8c\u8bbe\u8ba1\uff1a\u6210\u6279\u5e72\u9884\u65b9\u6cd5\u548c\u9010\u6b65\u5e72\u9884\u65b9\u6cd5\uff0c\u4ee5\u6700\u5c0f\u5316\u64cd\u7eb5\u53d8\u91cf\u7684\u6570\u91cf\u548c\u5019\u9009\u7ed3\u6784\u7684\u96c6\u5408\u3002\u5e76\u4e14\u4ece\u7406\u8bba\u4e0a\u8bc1\u660e\u4e86\u7ed3\u6784\u5b66\u4e60\u53ef\u4ee5\u5728\u94fe\u7ec4\u5206\u7684\u5b50\u56fe\u4e2d\u5c40\u90e8\u5b8c\u6210\uff0c\u800c\u4e0d\u9700\u8981\u68c0\u67e5\u6574\u4e2a\u56e0\u679c\u7f51\u7edc\u4e2d\u4e0d\u5408\u6cd5\u7684v-\u7ed3\u6784\u548c\u5faa\u73af\uff0c\u5e76\u4e14\u5728\u6bcf\u6b21\u5e72\u9884\u540e\u83b7\u5f97\u7684\u9a6c\u5c14\u53ef\u592b\u7b49\u4ef7\u5b50\u7c7b\u4ecd\u7136\u53ef\u4ee5\u88ab\u63cf\u8ff0\u4e3a\u94fe\u56fe\u3002<\/span><\/p>\n


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

\n
\n
\n

\u5c40\u90e8\u5b66\u4e60\uff1aMB-by-MB\u7b97\u6cd5<\/span><\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u5f53\u7814\u7a76\u76ee\u7684\u662f\u4e3a\u4e86\u5bfb\u627e\u548c\u786e\u5b9a\u67d0\u4e2a\u7ed9\u5b9a\u7684\u76ee\u6807\u53d8\u91cf\u7684\u5c40\u90e8\u56e0\u679c\u5173\u7cfb\uff0c\u53d1\u73b0\u5b83\u7684\u539f\u56e0\u662f\u4ec0\u4e48\u548c\u7ed3\u679c\u662f\u4ec0\u4e48\uff0c\u6216\u8005\u76ee\u7684\u662f\u505a\u5e72\u9884\u51b3\u7b56\u65f6\uff0c\u7814\u7a76\u4eba\u5458\u53ea\u9700\u5229\u7528\u6570\u636e\u53d1\u73b0\u8be5\u76ee\u6807\u53d8\u91cf\u7684\u5c40\u90e8\u56e0\u679c\u5173\u7cfb\uff0c\u6ca1\u5fc5\u8981\u5b66\u4e60\u6574\u4e2a\u7f51\u7edc\u7684\u7ed3\u6784\u3002\u5728\u201cDiscovering and orienting the edges connected to a target variable in a DAG via a sequential local learning approach\u201d\u4e00\u6587\u4e2d\uff0c\u63d0\u51fa\u4e86\u9010\u6b65\u6784\u5efa\u76ee\u6807\u53d8\u91cf Y \u7684\u5c40\u90e8\u7f51\u7edc\u7684\u65b9\u6cd5\uff0c\u79f0\u4e3a MB-by-MB <\/span>(Markov blanket by Markov blanket) <\/span>\u7b97\u6cd5\u3002<\/span><\/p>\n


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

\u8be5\u7b97\u6cd5\u4ee5\u76ee\u6807\u53d8\u91cfY\u4e3a\u4e2d\u5fc3\uff0c\u9010\u6b65\u5411\u5916\u6269\u6563\u5730\u6784\u5efa\u5c40\u90e8\u7f51\u7edc\u7ed3\u6784\u3002\u9996\u5148\u5bfb\u627e\u76ee\u6807\u53d8\u91cf Y \u7684 Markov\u8fb9\u754c MB(Y)\uff0c\u6784\u5efa MB(Y) \u7684\u5c40\u90e8\u7f51\u7edc\u7ed3\u6784; \u7136\u540e\u518d\u5bfb\u627eMB(Y)\u4e2d\u6bcf\u4e2aX<\/span>i<\/sub>\u7684MB(X<\/span>i<\/sub>)\uff0c\u6784\u5efa MB(Xi<\/sub>) \u7684\u5c40\u90e8\u7f51\u7edc\u7ed3\u6784\uff1b\u91cd\u590d\u8fd9\u4e2a\u8fc7\u7a0b\u76f4\u81f3\u80fd\u591f\u533a\u522b\u8c01\u662fY\u7684\u539f\u56e0\uff0c\u8c01\u662fY\u7684\u7ed3\u679c\u4e3a\u6b62\u3002<\/span><\/p>\n


<\/p>\n

\n
\n
\n

\u5c40\u90e8\u7f51\u7edc\u5b66\u4e60\u4e0e\u4f5c\u7528\u7684\u7ed3\u5408<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n


<\/p>\n

\u5728\u89c2\u5bdf\u6027\u7814\u7a76\u4e2d\uff0c\u7814\u7a76\u4eba\u5458\u611f\u5174\u8da3\u7684\u4e0d\u53ea\u662f\u8bc4\u4f30\u603b\u4f53\u6548\u679c\uff0c\u8fd8\u8981\u8bc4\u4f30\u6cbb\u7597\u53d8\u91cf\u5bf9\u53cd\u5e94\u53d8\u91cf\u7684\u76f4\u63a5\u548c\u95f4\u63a5\u5f71\u54cd\u3002\u5728\u56e0\u679c\u7f51\u7edc\u7684\u5c40\u90e8\u7ed3\u6784\u5b66\u4e60\u65b9\u9762\uff0c\u8bd5\u56fe\u627e\u5230\u6240\u6709\u53ef\u80fd\u7684\u603b\u56e0\u679c\u6548\u5e94\u548c\u76f4\u63a5\u56e0\u679c\u6548\u5e94\u5bf9\uff0c\u8fd9\u53ef\u4ee5\u8fdb\u4e00\u6b65\u7528\u4e8e\u8ba1\u7b97\u95f4\u63a5\u56e0\u679c\u6548\u5e94\u3002<\/span><\/p>\n


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

\u4e00\u79cd\u76f4\u89c2\u7684\u5168\u5c40\u5b66\u4e60\u65b9\u6cd5\u662f\u9996\u5148\u5728\u8868\u793a\u6240\u6709\u9a6c\u5c14\u53ef\u592b\u7b49\u4ef7\u56e0\u679c\u7f51\u7edc\u7684\u6240\u6709\u53d8\u91cf\u4e0a\u627e\u5230\u4e00\u4e2a\u672c\u8d28\u56fe\uff0c\u7136\u540e\u679a\u4e3e\u6240\u6709\u7b49\u4ef7\u7f51\u7edc\uff0c\u5e76\u4f30\u8ba1\u6bcf\u4e2a\u7f51\u7edc\u7684\u4e00\u5bf9\u603b\u6548\u5e94\u548c\u76f4\u63a5\u6548\u5e94\u3002\u7136\u800c\uff0c\u5f53\u771f\u6b63\u7684\u56e0\u679c\u56fe\u5f88\u5927\u65f6\uff0c\u5b66\u4e60\u4e00\u4e2a\u57fa\u672c\u56fe\u548c\u5217\u4e3e\u7b49\u4ef7\u7f51\u7edc\u53ef\u80fd\u662f\u4f4e\u6548\u7684\u3002<\/span><\/p>\n


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

\u56e0\u6b64\uff0c\u4e00\u7c7b\u5c40\u90e8\u5b66\u4e60\u65b9\u6cd5\u88ab\u63d0\u51fa\uff1a\u9996\u5148\u5728\u5c40\u90e8\u5b66\u4e60\u5305\u542b\u6cbb\u7597\u6216\u54cd\u5e94\u7684\u94fe\u7ec4\u4ef6\uff0c\u7136\u540e\u5c40\u90e8\u679a\u4e3e\u9a6c\u5c14\u53ef\u592b\u7b49\u4ef7\u7c7b\u4e2d\u6cbb\u7597\u7684\u7236\u8282\u70b9\u548c\u54cd\u5e94\u7684\u7236\u8282\u70b9\u7684\u6240\u6709\u53ef\u80fd\u914d\u5bf9\u3002\u4e3a\u4e86\u68c0\u67e5\u6cbb\u7597\u548c\u54cd\u5e94\u7684\u90bb\u5c45\u7684\u4efb\u4f55\u5b9a\u5411\u7ed3\u6784\u7684\u6709\u6548\u6027\uff0c\u5f15\u5165\u4ec5\u4f9d\u8d56\u4e8e\u6cbb\u7597\u548c\u54cd\u5e94\u7684\u76f8\u90bb\u8282\u70b9\u4e0a\u5b66\u4e60\u7684\u94fe\u7ec4\u4ef6\u7684\u5b50\u56fe\u7684\u5c40\u90e8\u51c6\u5219\u3002<\/span><\/p>\n


<\/p>\n

\u53c2\u8003\u6587\u732e<\/span><\/strong><\/span><\/h3>\n

1.  Geng, Z., Liu, Y., Liu, C. C. and Miao, W. (2019) Evaluation of Causal Effects and Local Structure Learning of Causal Networks. Ann. Rev. Statist. & Appl. 6, 103-124.<\/span><\/p>\n

2.  Liu, Y., Fang, Z. Y., He, Y. B., Geng, Z. and Liu, C. C. (2021) Local causal network learning for finding pairs of total and direct effects. To appear in J Mach Learn Res.<\/span><\/p>\n

3.  He, Y. B., & Geng, Z. (2008). Active learning of causal networks with intervention experiments and optimal designs. Journal of Machine Learning Research, 9(Nov), 2523-2547.<\/span><\/p>\n

4.  Jiang, Z., Ding, P., & Geng, Z. (2015). Qualitative evaluation of associations by the transitivity of the association signs. Statistica Sinica, 1065-1079.<\/span><\/p>\n

5.  Wang, C., Zhou, Y., Zhao, Q., & Geng, Z. (2014). Discovering and orienting the edges connected to a target variable in a DAG via a sequential local learning approach. Computational Statistics & Data Analysis, 77, 252-266.<\/span><\/p>\n

6.  Xie, X., & Geng, Z. (2008). A recursive method for structural learning of directed acyclic graphs. The Journal of Machine Learning Research, 9, 459-483.<\/span><\/p>\n

7.  Pearl, J. and Mackenzie, D. (2018) The Book of Why. \u4e3a\u4ec0\u4e48\uff1a\u5173\u4e8e\u56e0\u679c\u5173\u7cfb\u7684\u65b0\u79d1\u5b66, \u4e2d\u4fe1\u51fa\u7248\u96c6\u56e2\uff0c\u4e2d\u8bd1\u7248\uff0c2019<\/span><\/p>\n

8.  \u82d7\u65fa\uff0c\u5218\u6625\u8fb0\uff0c\u803f\u76f4 (2018) \u56e0\u679c\u4f5c\u7528\u4e0e\u56e0\u679c\u7f51\u7edc. \u4e2d\u56fd\u79d1\u5b66-\u6570\u5b66, 48, 1753-1778.<\/span><\/p>\n


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


<\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

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

\u76f4\u64ad\u62a5\u540d<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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

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

\u56e0\u679c\u63a8\u65ad\u672c\u8d28\u662f\u6d88\u9664\u6df7\u6742\u56e0\u7d20\u5f15\u8d77\u7684\u865a\u5047\u76f8\u5173\u5173\u7cfb\u3002\u672c\u62a5\u544a\u5c06\u4ecb\u7ecd\u5224\u65adPearl\u7684\u4e09\u4e2a\u5c42\u7ea7\u95ee\u9898\u3001\u6df7\u6742\u56e0\u7d20\u7684\u51c6\u5219\uff0c\u66ff\u4ee3\u6307\u6807\u6096\u8bba\uff0c\u56e0\u679c\u4f5c\u7528\u548c\u76f8\u5173\u5ea6\u91cf\u7684\u53ef\u4f20\u9012\u6027\u3001\u56e0\u679c\u7f51\u7edc\u7ed3\u6784\u7684\u5206\u89e3\u5b66\u4e60\u3001\u4e3b\u52a8\u5b66\u4e60\u548c\u5c40\u90e8\u5b66\u4e60\u65b9\u6cd5\u3002\u8fdb\u4e00\u6b65\u4ecb\u7ecd\u56e0\u679c\u4f5c\u7528\u8bc4\u4ef7\u4e0e\u56e0\u679c\u7f51\u7edc\u5c40\u90e8\u7ed3\u6784\u5b66\u4e60\u7ed3\u5408\u7684\u65b9\u6cd5\u3002<\/span>
<\/p>\n


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

\u76f4\u64ad\u65f6\u95f4\uff1a<\/span><\/strong>7\u670824\u65e5<\/span>\uff08\u5468\u516d\uff09<\/span>10:00-12:00<\/span><\/p>\n


<\/p>\n

\u53c2\u4e0e\u65b9\u5f0f\uff1a<\/span><\/strong><\/span><\/section>\n

1. B\u7ad9\u76f4\u64ad<\/span><\/p>\n

2. \u7ebf\u4e0b\u53c2\u4e0e<\/span>\uff08\u9700\u53c2\u52a0\u56e0\u679c\u8bfb\u4e66\u4f1a\uff0c\u5750\u6807\u5317\u4eac\uff0c\u5c11\u91cf\u540d\u989d\uff0c\u5148\u5230\u5148\u5f97\uff09<\/span><\/p>\n

3. <\/span>\u817e\u8baf\u4f1a\u8bae<\/span>\uff08\u9700\u53c2\u52a0\u56e0\u679c\u8bfb\u4e66\u4f1a\uff09<\/span><\/section>\n


<\/p>\n

\u62a5\u540d\u5730\u5740<\/span><\/strong>\uff08\u83b7\u53d6B\u7ad9\u76f4\u64ad\u95f4\u5730\u5740\uff09<\/span>\uff1a<\/span><\/p>\n


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

<\/p>\n

\u626b\u63cf\u4e8c\u7ef4\u7801\uff0c\u70b9\u51fb\u201c\u6211\u8981\u542c\u201d\u9884\u7ea6\uff0c\u5185\u9644 B \u7ad9\u76f4\u64ad\u5730\u5740\uff0c\u53ef\u83b7\u5f97\u6d3b\u52a8\u63d0\u9192\uff0c\u548c\u56de\u653e\u4e0a\u7ebf\u901a\u77e5\u3002<\/span><\/section>\n


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

\n
\n
\n
\n
\n

\u6b22\u8fce\u52a0\u5165\u56e0\u679c\u79d1\u5b66\u793e\u533a<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/section>\n<\/h2>\n

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

\u56e0\u679c\u63a8\u65ad\u4e0e\u673a\u5668\u5b66\u4e60\u9886\u57df\u7684\u7ed3\u5408\u5df2\u7ecf\u5438\u5f15\u4e86\u8d8a\u6765\u8d8a\u591a\u6765\u81ea\u5b66\u754c\u4e1a\u754c\u7684\u5173\u6ce8\uff0c\u4e3a\u6df1\u5165\u63a2\u8ba8\u3001\u666e\u53ca\u63a8\u5e7f\u56e0\u679c\u79d1\u5b66\u8bae\u9898\uff0c\u667a\u6e90\u793e\u533a\u643a\u624b\u96c6\u667a\u4ff1\u4e50\u90e8\u8fde\u7eed\u4e3e\u529e\u4e86\u4e24\u5b63\u300c\u56e0\u679c\u79d1\u5b66\u4e0eCausalAI\u8bfb\u4e66\u4f1a<\/a>\u300d\uff0c\u7814\u8bfb\u91cd\u8981\u8bba\u6587\u4e0e\u7ecf\u5178\u6559\u6750\u3002\u672c\u6b21\u6d3b\u52a8\u662f\u8bfb\u4e66\u4f1a\u7b2c\u4e8c\u5b63\u7684\u6536\u5b98\u8bb2\u5ea7\uff0c\u6d3b\u52a8\u7ed3\u675f\u540e\u8bfb\u4e66\u4f1a\u4ecd\u5c06\u7ee7\u7eed\u5f00\u653e\u62a5\u540d\u3002<\/span><\/p>\n


<\/p>\n

\u76ee\u524d\u56e0\u679c\u79d1\u5b66\u8bfb\u4e66\u4f1a<\/span>\uff08\u5373\u56e0\u679c\u79d1\u5b66\u793e\u533a\uff09<\/span>\uff0c\u5df2\u7ecf\u6709\u8d85\u8fc7600\u4f4d\u6d77\u5185\u5916\u76f8\u5173\u79d1\u7814\u5de5\u4f5c\u8005\u4ee5\u53ca\u4e92\u8054\u7f51\u4e00\u7ebf\u4ece\u4e1a\u4eba\u5458\u53c2\u4e0e\uff0c\u5982\u679c\u4f60\u4e5f\u5bf9\u8fd9\u4e2a\u4e3b\u9898\u611f\u5174\u8da3\uff0c\u60f3\u8981\u6df1\u5ea6\u5730\u53c2\u4e0e\uff0c\u5c31\u5feb\u52a0\u5165\u6211\u4eec\u5427\uff01<\/span><\/p>\n


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

<\/section>\n

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


<\/p>\n


<\/section>\n
\n

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