{"id":1078461,"date":"2024-08-21T20:51:00","date_gmt":"2024-08-22T03:51:00","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-blog-post&#038;p=1078461"},"modified":"2024-09-25T04:25:40","modified_gmt":"2024-09-25T11:25:40","slug":"causal-discovery","status":"publish","type":"msr-blog-post","link":"https:\/\/www.microsoft.com\/en-us\/research\/articles\/causal-discovery\/","title":{"rendered":"\u5f00\u542f\u56e0\u679c\u53d1\u73b0\u65b0\u8303\u5f0f\uff01\u89e3\u5bc6\u590d\u6742\u7cfb\u7edf\u7684\u6838\u5fc3\u673a\u5236"},"content":{"rendered":"\n<p><em>\u4f5c\u8005\uff1aDKI\u7ec4<\/em><\/p>\n\n\n\n<p>\u7f16\u8005\u6309\uff1a\u5728\u5f53\u4eca\u6570\u636e\u9a71\u52a8\u7684\u4e16\u754c\uff0c\u7406\u89e3\u590d\u6742\u7cfb\u7edf\u4e2d\u7684\u56e0\u679c\u5173\u7cfb\u662f\u79d1\u5b66\u7814\u7a76\u548c\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u5173\u952e\u6311\u6218\u3002\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df\uff0c\u56e0\u679c\u63a8\u7406\u80fd\u529b\u66f4\u662f\u6210\u4e3a\u4e00\u4e2a\u70ed\u95e8\u8bdd\u9898\u3002\u5982\u4f55\u63ed\u793a\u6570\u636e\u80cc\u540e\u56e0\u679c\u673a\u5236\u7684\u5173\u952e\uff1f\u5982\u4f55\u5229\u7528\u6570\u636e\u5b9e\u73b0\u56e0\u679c\u53d1\u73b0\u7684\u7a81\u7834\uff1f\u4e3a\u56de\u7b54\u8fd9\u4e9b\u95ee\u9898\uff0c\u6765\u81ea\u5fae\u8f6f DKI\uff08Data, Knowledge and Intelligence\uff0c\u6570\u636e\u3001\u77e5\u8bc6\u4e0e\u667a\u80fd\uff09\u9886\u57df\u7684\u7814\u7a76\u5458\u4eec\u5728\u8fdb\u884c\u4e86\u6301\u7eed\u800c\u6df1\u5165\u7684\u63a2\u7d22\uff0c\u5176\u76f8\u5173\u6210\u679c\u53d1\u8868\u5728 AAAI 2020\u3001KDD 2022\u3001SDM 2023\u3001SIGMOD 2023 \u548c KDD 2024 \u7b49\u56fd\u9645\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae\u4e0a\u3002\u540c\u65f6\uff0c\u8be5\u7cfb\u5217\u5de5\u4f5c\u4e5f\u5728 Microsoft Power BI \u7b49\u4ea7\u54c1\u4e2d\u5f97\u5230\u5e94\u7528\u3002<\/p>\n\n\n\n<p>\u672c\u7bc7\u6587\u7ae0\u5c06\u4e3a\u5927\u5bb6\u4ecb\u7ecd\u7531\u5fae\u8f6f\uff08\u4e9a\u6d32\uff09\u4e92\u8054\u7f51\u5de5\u7a0b\u9662 DKI \u7ec4\u548c\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u3001\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff08CMU\uff09\u5408\u4f5c\u5b8c\u6210\u7684\u524d\u6cbf\u7814\u7a76\u9879\u76ee\uff0c\u8be5\u9879\u76ee\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u673a\u5668\u76d1\u7763\u5b66\u4e60\u7684\u56e0\u679c\u53d1\u73b0\u65b0\u8303\u5f0f\uff0c\u5305\u542b\u5bf9\u56e0\u679c\u5b9a\u5411\u3001\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\u3001\u5904\u7406\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u65f6\u7684\u56e0\u679c\u53d1\u73b0\u4e09\u4e2a\u5173\u952e\u95ee\u9898\u7684\u56de\u7b54\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u56e0\u679c\u53d1\u73b0\uff08causal discovery\uff09\u662f\u6570\u636e\u79d1\u5b66\u548c\u673a\u5668\u5b66\u4e60\u4e2d\u7684\u5173\u952e\u9886\u57df\uff0c\u5176\u4efb\u52a1\u662f\u901a\u8fc7\u7b97\u6cd5\u4ece\u6570\u636e\u4e2d\u63d0\u53d6\u56e0\u679c\u8054\u7cfb\uff0c\u5e76\u7528\u56e0\u679c\u56fe\u8fdb\u884c\u8868\u793a\u3002\u56e0\u679c\u53d1\u73b0\u5bf9\u4e8e\u7406\u89e3\u590d\u6742\u7cfb\u7edf\u3001\u652f\u6301\u653f\u7b56\u5236\u5b9a\u3001\u533b\u5b66\u548c\u793e\u4f1a\u79d1\u5b66\u7814\u7a76\u7b49\u9886\u57df\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002<\/p>\n\n\n\n<p>\u4f20\u7edf\u4e0a\uff0c\u57fa\u4e8e\u89c2\u6d4b\u6570\u636e\u7684\u56e0\u679c\u53d1\u73b0\u4e3b\u8981\u6709\u4e09\u7c7b\u65b9\u6cd5\uff1a\u57fa\u4e8e\u7ea6\u675f\u3001\u8bc4\u5206\u548c\u8fde\u7eed\u4f18\u5316\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u57fa\u4e8e\u7ea6\u675f\u7684\u65b9\u6cd5\uff1a\u901a\u8fc7\u7edf\u8ba1\u6d4b\u8bd5\u6765\u5224\u65ad\u53d8\u91cf\u4e4b\u95f4\u7684\u72ec\u7acb\u6027\u548c\u4f9d\u8d56\u6027\uff0c\u4ece\u800c\u6784\u5efa\u4e0e\u72ec\u7acb\u6027\u76f8\u5bb9\u7684\u56e0\u679c\u56fe\u3002\u8fd9\u7c7b\u65b9\u6cd5\u4f9d\u8d56\u6761\u4ef6\u72ec\u7acb\u6027\u6d4b\u8bd5\uff0c\u65e8\u5728\u8bbe\u8ba1\u66f4\u9c81\u68d2\u3001\u66f4\u5feb\u901f\u7684\u6d4b\u8bd5\u7b97\u6cd5\u6216\u7ed3\u5408\u641c\u7d22\u7b56\u7565\u63d0\u9ad8\u6548\u7387\u3002\u5e38\u7528\u7b97\u6cd5\u5305\u62ec PC \u548c FCI\u3002<\/li>\n\n\n\n<li>\u57fa\u4e8e\u8bc4\u5206\u7684\u65b9\u6cd5\uff1a\u901a\u8fc7\u4e3a\u6bcf\u4e2a\u53ef\u80fd\u7684\u56e0\u679c\u56fe\u5206\u914d\u8bc4\u5206\uff0c\u5e76\u5bfb\u627e\u5f97\u5206\u6700\u9ad8\u7684\u7ed3\u6784\u3002\u8bc4\u5206\u4f9d\u636e\u7684\u662f\u6570\u636e\u62df\u5408\u7a0b\u5ea6\u548c\u6a21\u578b\u590d\u6742\u6027\uff0c\u5176\u5e38\u7528\u51c6\u5219\u4e3a\u8d1d\u53f6\u65af\u4fe1\u606f\u51c6\u5219\uff08BIC\uff09\u548c\u8d1d\u53f6\u65af\u72c4\u5229\u514b\u96f7\u7b49\u4ef7\u5747\u5300\u51c6\u5219\uff08BDeu\uff09\uff0c\u7b97\u6cd5\u4e3b\u8981\u56f4\u7ed5\u7ec4\u5408\u4f18\u5316\u6c42\u89e3\u548c\u542f\u53d1\u5f0f\u641c\u7d22\uff0c\u6bd4\u5982 GES\u3002<\/li>\n\n\n\n<li>\u57fa\u4e8e\u8fde\u7eed\u4f18\u5316\u7684\u65b9\u6cd5\uff1a\u5c06\u79bb\u6563\u7684\u7ec4\u5408\u4f18\u5316\u95ee\u9898\u8f6c\u6362\u4e3a\u8fde\u7eed\u4f18\u5316\u95ee\u9898\uff0c\u5728\u8fd1\u5e74\u6210\u4e3a\u7814\u7a76\u70ed\u70b9\u3002\u4f8b\u5982\uff0cNOTEARS \u7b97\u6cd5\u5c06\u56e0\u679c\u53d1\u73b0\u95ee\u9898\u8f6c\u5316\u4e3a\u4e86\u8fde\u7eed\u4f18\u5316\u95ee\u9898\uff0c\u5e76\u4f7f\u7528\u68af\u5ea6\u4e0b\u964d\u7b49\u6280\u672f\u8fdb\u884c\u6c42\u89e3\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u8fd9\u4e9b\u4f20\u7edf\u65b9\u6cd5\u4f53\u73b0\u4e86\u56e0\u679c\u53d1\u73b0\u9886\u57df\u8fc7\u53bb\u7684\u4e3b\u6d41\u65b9\u6cd5\u8bba\uff0c\u5373\u5c06\u4eba\u7c7b\u7684\u667a\u6167\u7f16\u7801\u6210\u5de7\u5999\u7684\u7b97\u6cd5\uff0c\u6765\u63d0\u9ad8\u56e0\u679c\u53d1\u73b0\u4efb\u52a1\u7684\u6027\u80fd\u3002\u7136\u800c\uff0c\u6765\u81ea\u5fae\u8f6f\uff08\u4e9a\u6d32\uff09\u4e92\u8054\u7f51\u5de5\u7a0b\u9662 DKI \u7ec4\u3001\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u548c\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff08CMU\uff09\u7684\u7814\u7a76\u5458\u4eec\u5c1d\u8bd5\u7740\u6539\u53d8\u601d\u8def\uff0c<strong>\u4ece\u57fa\u4e8e\u5b66\u4e60\u7684\u65b9\u6cd5\u8bba\uff08learning-based\uff09\u5165\u624b\uff0c\u7740\u773c\u4e8e\u5b9e\u8bc1\u6027\u80fd\uff08empirical performance-driven\uff09\u6765\u63a2\u8ba8\u4e00\u79cd\u57fa\u4e8e\u76d1\u7763\u5b66\u4e60\u7684\u56e0\u679c\u53d1\u73b0\u65b0\u8303\u5f0f\uff08supervision-based causal discovery\uff09<\/strong>\u3002\u8be5\u8303\u5f0f\u805a\u7126\u8bad\u7ec3\u6570\u636e\u83b7\u53d6\u3001\u5b66\u4e60\u4efb\u52a1\u5b9a\u4e49\u3001\u5b66\u4e60\u76ee\u6807\u8bbe\u8ba1\u3001\u5b9e\u73b0\u7aef\u5230\u7aef\u7684\u56e0\u679c\u5173\u7cfb\u8bc6\u522b\uff0c\u540c\u65f6\u4e25\u8083\u5bf9\u5f85\u56e0\u679c\u53d1\u73b0\u7684\u6838\u5fc3\u6982\u5ff5\uff1a\u5206\u6790\u53ef\u5b66\u4e60\u6027 \uff08learnability\uff09 \u4e0e\u56e0\u679c\u7ed3\u6784\u53ef\u8fa8\u8bc6\u6027\uff08identifiability\uff09\u7684\u5173\u7cfb\u3002<\/p>\n\n\n\n<p><em>\u6ce8\uff1a\u57fa\u4e8e\u5b66\u4e60\u7684\u65b9\u6cd5\u8bba\u6210\u4e3a\u53ef\u80fd\u7684\u524d\u63d0\u662f\uff0c\u56e0\u679c\u53d1\u73b0\u4efb\u52a1\u5177\u5907\u201c\u81ea\u76d1\u7763\u5b66\u4e60\u201d\u7684\u7279\u8d28\uff1a\u53ef\u4ee5\u81ea\u52a8\u5408\u6210\u5927\u91cf <Gi, Di> \u914d\u5bf9\uff0c\u5176\u4e2d Gi \u662f\u56e0\u679c\u56fe\uff0cDi \u662f\u4e0e\u4e4b\u76f8\u5bb9\u7684\u89c2\u6d4b\u6570\u636e\u3002\u7136\u540e\u4ee4 Di \u4f5c\u4e3a\u6a21\u578b\u8f93\u5165\uff0cGi \u4f5c\u4e3a\u6807\u6ce8\u3002<\/em><\/p>\n\n\n\n<p>\u4e0b\u9762\uff0c\u6211\u4eec\u4e00\u8d77\u6765\u4e86\u89e3\u4e00\u4e0b\u5982\u4f55\u5728\u56e0\u679c\u5b9a\u5411\uff08causal orientation\uff09\u3001\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\uff08casual skeleton learning\uff09\u4ee5\u53ca\u5904\u7406\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u65f6\u7684\u56e0\u679c\u53d1\u73b0\uff08causal discovery in the presence of latent confounders\uff09\u9886\u57df\u7684\u4e00\u7cfb\u5217\u63a2\u7d22\u4e2d\u5b8c\u6210\u5bf9\u8be5\u8303\u5f0f\u7684\u6784\u5efa\u3002\u8fd9\u4e5f\u5c06\u4e3a\u672a\u6765\u8be5\u9886\u57df\u7684\u63a2\u7d22\u63d0\u4f9b\u5d2d\u65b0\u7684\u601d\u8def\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ml4c-\u4ece\u76d1\u7763\u5b66\u4e60\u51fa\u53d1\u8fdb\u884cv-\u7ed3\u6784\u56e0\u679c\u5b9a\u5411\">ML4C\uff1a\u4ece\u76d1\u7763\u5b66\u4e60\u51fa\u53d1\u8fdb\u884cV-\u7ed3\u6784\u56e0\u679c\u5b9a\u5411<\/h2>\n\n\n\n<p>\u53ef\u8bc6\u522b\u6027\u662f\u6307\u6839\u636e\u89c2\u5bdf\u6570\u636e\u786e\u5b9a\u552f\u4e00\u6a21\u578b\u7684\u80fd\u529b\uff0c\u662f\u56e0\u679c\u53d1\u73b0\u7684\u91cd\u4e2d\u4e4b\u91cd\u3002\u5728\u4f20\u7edf\u7684\u56e0\u679c\u53d1\u73b0\u7b97\u6cd5\u4e2d\uff0c\u4eba\u4eec\u901a\u5e38\u57fa\u4e8e\u5bf9\u6570\u636e\u751f\u6210\u8fc7\u7a0b\u7684\u67d0\u4e9b\u5047\u8bbe\uff0c\u6765\u8bbe\u8ba1\u5408\u9002\u7684\u7279\u5f81\u4ee5\u8bc6\u522b\u56e0\u679c\u7ed3\u6784\uff0c\u4f8b\u5982\u57fa\u4e8e\u4e0d\u5bf9\u79f0\u6027\u3002\u7136\u800c\uff0c\u7814\u7a76\u5458\u4eec\u9009\u62e9\u6539\u53d8\u601d\u8def\uff0c\u4ece\u673a\u5668\u5b66\u4e60\u7684\u89c6\u89d2\u51fa\u53d1\uff0c\u5728\u6570\u636e\u4e2d\u76f4\u63a5\u5b66\u5f97\u8fd9\u4e9b\u7279\u5f81\u6765\u5e2e\u52a9\u8bc6\u522b\u56e0\u679c\u7ed3\u6784\uff0c\u4ece\u800c\u63d0\u5347\u6027\u80fd\u3002<\/p>\n\n\n\n<p>\u8fd9\u5176\u4e2d\u7684\u5173\u952e\u5728\u4e8e\uff0c\u5b66\u4e60\u76ee\u6807\u5e94\u8be5\u5982\u4f55\u5b9a\u4e49\uff1f\u6362\u8a00\u4e4b\uff0c\u54ea\u4e9b\u6570\u636e\u7279\u5f81\u5e94\u4f5c\u4e3a\u8bc6\u522b\u5bf9\u8c61\u5e2e\u52a9\u5224\u65ad\u56e0\u679c\u7ed3\u6784\uff1f<\/p>\n\n\n\n<p>\u4e3a\u4e86\u56de\u7b54\u8fd9\u4e2a\u95ee\u9898\uff0c\u7814\u7a76\u5458\u4eec\u805a\u7126\u4e8e\u65e0\u9700\u53c2\u6570\u5316\u5047\u8bbe\u7684\u65b9\u6cd5\uff0c\u4ec5\u5229\u7528\u89c2\u6d4b\u6570\u636e\u7684\u6761\u4ef6\u72ec\u7acb\u5173\u7cfb\u6765\u63a8\u6d4b\u56e0\u679c\u7ed3\u6784\u3002\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u53d1\u73b0\u4e24\u4e2a\u56e0\u679c\u56fe\u5982 X\u2192Y\u2192Z \u548c X\u2190Y\u2190Z \u5b8c\u5168\u65e0\u6cd5\u533a\u5206\uff0c\u56e0\u4e3a\u5b83\u4eec\u8574\u542b\u7684\u6761\u4ef6\u72ec\u7acb\u5173\u7cfb\u5b8c\u5168\u76f8\u540c\uff08\u552f\u4e00\u7684\u72ec\u7acb\u5173\u7cfb\u662f X\u22a5Z\u2223Y\uff09\uff0c\u56e0\u6b64\uff0c\u7814\u7a76\u5458\u4eec\u6307\u51fa\uff0c\u4e0d\u53ef\u4ee5\u76f4\u63a5\u62ff\u56e0\u679c\u7684\u6709\u5411\u8fb9\u4f5c\u4e3a\u5b66\u4e60\u76ee\u6807\uff0c\u8fd9\u5c06\u4f7f\u76d1\u7763\u6a21\u578b\u65e0\u6cd5\u6536\u655b\u3002<\/p>\n\n\n\n<p>\u5728\u4e00\u822c\u60c5\u51b5\u4e0b\uff0c\u7814\u7a76\u5458\u4eec\u8ba4\u4e3a\u5176\u53ef\u8bc6\u522b\u7684\u76ee\u6807\u5206\u4e3a\u4e24\u7c7b\uff1a\u4e24\u4e2a\u53d8\u91cf\u662f\u5426\u90bb\u63a5\uff0c\u4ee5\u53ca\u786e\u5b9a\u90bb\u63a5\u5173\u7cfb\u540e\u5bf9\u53d8\u91cf\u8fdb\u884c\u5b9a\u5411\u3002\u524d\u8005\u4e0e\u9aa8\u67b6\u5b66\u4e60\u76f8\u5173\uff0c\u540e\u8005\u4e0e\u56e0\u679c\u5b9a\u5411\u76f8\u5173\u3002\u5bf9\u4e8e\u540e\u8005\uff0c\u7814\u7a76\u5458\u4eec\u53d1\u73b0\uff0c\u5bf9\u4e8e\u4e00\u4e2a\u672a\u906e\u6321\u4e09\u5143\u7ec4 X-Y-Z \u6765\u8bf4\uff0cX \u548c Y\u3001Y \u548c Z \u90bb\u63a5\uff0c\u800c X \u548c Z \u4e0d\u76f8\u8fde\uff0c\u56e0\u6b64\u4e00\u5b9a\u5b58\u5728\u4e00\u4e2a\u5269\u4f59\u53d8\u91cf\u7684\u5b50\u96c6 S\uff0c\u4f7f\u5f97 X\u22a5Z\u2223S\u3002\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 V-\u7ed3\u6784\uff08V-structure\uff0c\u5373 X\u2192Y\u2190Z\uff09\u7684\u5b9a\u5411\u89c4\u5219\uff1a\u5f53\u4e14\u4ec5\u5f53 Y \u4e0d\u5c5e\u4e8e X \u548c Z \u7684\u6761\u4ef6\u72ec\u7acb\u96c6 S\uff0c\u5e76\u5c06\u5176\u8bbe\u5b9a\u4e3a\u4e3b\u8981\u7684\u5b66\u4e60\u76ee\u6807\u3002<\/p>\n\n\n\n<p>\u7efc\u4e0a\uff0cV-\u7ed3\u6784\u7684\u8bc6\u522b\u770b\u4f3c\u7b80\u5355\uff0c\u53ea\u9700\u67e5\u770b\u4e00\u4e2a\u672a\u906e\u6321\u4e09\u5143\u7ec4 X-Y-Z \u7684\u4e2d\u95f4\u53d8\u91cf Y \u662f\u5426\u5c5e\u4e8e X \u548c Z \u7684\u4efb\u610f\u4e00\u4e2a\u6761\u4ef6\u72ec\u7acb\u96c6 S\u3002\u7136\u800c\uff0c\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u8fd9\u79cd\u5224\u65ad\u5bb9\u6613\u5bfc\u81f4\u8bc6\u522b\u9519\u8bef\uff0c\u8fdb\u800c\u4ea7\u751f\u7ea7\u8054\u6548\u5e94\u3002\u56e0\u6b64\uff0c\u5df2\u6709\u7814\u7a76\u5c1d\u8bd5\u4e86\u4e0d\u540c\u7684\u65b9\u6cd5\uff0c\u5982\u6bd4\u8f83 X \u548c Z \u5728\u7ed9\u5b9a S \u548c\u7ed9\u5b9a S\u222a{Y} \u65f6\u6761\u4ef6\u72ec\u7acb\u5ea6\u91cf\u7684\u533a\u522b\uff08\u66f4\u7cbe\u7ec6\u4e14\u5b9a\u91cf\u5316\u7684\u51c6\u5219\uff09\u3002\u53e6\u5916\uff0c\u8003\u8651\u5230\u4e0d\u6b62\u6709\u4e00\u4e2a\u8fd9\u6837\u7684\u6761\u4ef6\u72ec\u7acb\u96c6 S\uff0c\u53ef\u4ee5\u8ba1\u7b97\u6240\u6709\u6761\u4ef6\u72ec\u7acb\u96c6 S \u4e2d Y \u51fa\u73b0\u7684\u6bd4\u4f8b\u3002\u4f8b\u5982\uff0c\u4fdd\u5b88 PC\uff08conservative PC, CPC\uff09\u5224\u5b9a V-\u7ed3\u6784\u5f53\u4e14\u4ec5\u5f53 Y \u4e0d\u5728\u4efb\u4f55 S \u4e2d\u51fa\u73b0\uff0c\u800c\u7a33\u5b9a PC\uff08PC-stable\uff09\u5224\u5b9a V-\u7ed3\u6784\u5f53\u4e14\u4ec5\u5f53 Y \u5728\u5c11\u4e8e50%\u7684 S \u4e2d\u51fa\u73b0\u3002\u8fd9\u4e2a\u9608\u503c\u662f\u53ef\u4ee5\u8c03\u6574\u7684\u3002\u7814\u7a76\u5458\u4eec\u89c2\u5bdf\u5230\uff0c\u4e0d\u7ba1\u662f\u539f\u6709\u7684 PC\uff0c\u8fd8\u662f\u6539\u8fdb\u540e\u7684 CPC\u3001PC-stable\uff0c\u7406\u8bba\u4e0a\u90fd\u662f\u540c\u6837\u6b63\u786e\u7684\u3002\u4f46\u662f\uff0c\u540e\u4e24\u8005\u6bd4\u539f\u6709 PC \u663e\u5f97\u66f4\u201c\u7cbe\u7ec6\u201d\uff0c\u56e0\u4e3a\u5b83\u6709\u66f4\u590d\u6742\u7684\u5206\u7c7b\u673a\u5236\u3001\u7528\u5230\u4e86\u66f4\u591a\uff08\u4e00\u5b9a\u610f\u4e49\u4e0a\u66f4\u5168\u9762\uff09\u7684\u4fe1\u606f\u6765\u505a\u5224\u65ad\uff0c\u4e5f\u56e0\u6b64\u6709\u66f4\u597d\u7684\u5b9e\u9645\u8868\u73b0\u3002\u8fd9\u4e9b\u975e\u76d1\u7763\u7684\u3001\u4f9d\u8d56\u624b\u5de5\u786e\u5b9a\u7684\u65b9\u6cd5\u4e5f\u5b8c\u5168\u53ef\u4ee5\u770b\u4f5c\u662f\u7279\u5f81\uff0c\u4f46\u8fd9\u4e9b\u7279\u5f81\u4ece\u673a\u5668\u5b66\u4e60\u7684\u89d2\u5ea6\u6765\u770b\uff0c\u53c8\u8fc7\u4e8e\u7b80\u5355\u3002\u6240\u4ee5\uff0c\u7814\u7a76\u5458\u4eec\u5229\u7528\u76d1\u7763\u5b66\u4e60\uff0c\u4e0d\u518d\u91c7\u7528\u8fd9\u4e9b\u7b80\u5355\u7684\u624b\u5de5\u5206\u7c7b\u673a\u5236\uff0c\u800c\u662f\u76f4\u63a5\u4ece\u6570\u636e\u4e2d\u81ea\u52a8\u5b66\u5f97\u66f4\u7cbe\u7ec6\u3001\u590d\u6742\u3001\u5168\u9762\u7684\u5206\u7c7b\u673a\u5236\u2014\u2014\u5c3d\u7ba1\u8fd9\u4e9b\u673a\u5236\u53ef\u80fd\u4e0d\u5bb9\u6613\u89e3\u91ca\uff0c\u4f46\u7531\u4e8e\u5047\u8bbe\u5b66\u4e60\u76ee\u6807\u662f\u53ef\u8fa8\u8bc6\u7684\uff0c\u6240\u4ee5\u53ef\u4ee5\u653e\u5fc3\u4f7f\u7528\u76d1\u7763\u5e26\u6765\u7684\u6027\u80fd\u63d0\u5347\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u8bc6\u522b V-\u7ed3\u6784\uff0c\u7814\u7a76\u5458\u4eec\u8bbe\u8ba1\u4e86 ML4C \u7b97\u6cd5\uff0c\u5176\u4e0d\u518d\u91c7\u7528\u7b80\u5355\u7684\u624b\u5de5\u5206\u7c7b\u673a\u5236\uff0c\u800c\u662f\u5229\u7528\u76d1\u7763\u5b66\u4e60\u76f4\u63a5\u4ece\u6570\u636e\u4e2d\u81ea\u52a8\u5b66\u5f97\u66f4\u7cbe\u7ec6\u3001\u590d\u6742\u3001\u5168\u9762\u7684\u5206\u7c7b\u673a\u5236\u3002\u5177\u4f53\u800c\u8a00\uff0cML4C \u5047\u8bbe\u5df2\u83b7\u5f97\u6b63\u786e\u7684\u65e0\u5411\u56fe\uff0c\u7136\u540e\u5229\u7528\u76d1\u7763\u5b66\u4e60\u8fdb\u884c\u4e8c\u5206\u7c7b\u4efb\u52a1\uff0c\u4ee5\u5224\u65ad\u6bcf\u4e2a\u672a\u906e\u6321\u4e09\u5143\u7ec4\u662f\u5426\u4e3a V-\u7ed3\u6784\u3002\u4e00\u65e6\u8bc6\u522b\u51fa V-\u7ed3\u6784\uff0cML4C \u5c06\u5e94\u7528 Meek \u89c4\u5219\u751f\u6210\u6700\u7ec8\u7684 CPDAG\u3002\u66f4\u5177\u521b\u65b0\u7684\u662f\uff0cML4C \u80fd\u6269\u5c55\u7279\u5f81\u96c6\uff0c\u4e0d\u4ec5\u5173\u6ce8\u672a\u906e\u6321\u4e09\u5143\u7ec4\u7684\u53d8\u91cf\uff0c\u8fd8\u5305\u62ec\u90bb\u57df\u4e2d\u7684\u6761\u4ef6\u4f9d\u8d56\u5173\u7cfb\u548c\u7ed3\u6784\u7ea0\u7f20\uff0c\u4ece\u800c\u589e\u5f3a V-\u7ed3\u6784\u8bc6\u522b\u7684\u51c6\u786e\u6027\u3002<\/p>\n\n\n\n<p>\u901a\u8fc7\u751f\u6210\u6a21\u62df\u8bad\u7ec3\u6570\u636e\u5e76\u4f7f\u7528\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u8bad\u7ec3\u5206\u7c7b\u5668\uff0cML4C \u8fbe\u5230\u4e86\u6e10\u8fdb\u6b63\u786e\u6027\uff0c\u5e76\u8868\u73b0\u51fa\u9ad8\u9c81\u68d2\u6027\uff0c\u5728\u6027\u80fd\u8bc4\u4f30\u4e2d\u4e5f\u4f18\u4e8e\u4f20\u7edf\u7b97\u6cd5\uff0c\u5728\u591a\u4e2a\u6807\u51c6\u6570\u636e\u96c6\u4e0a\u8868\u73b0\u4f18\u5f02\uff0c\u5c24\u5176\u662f\u5728\u566a\u58f0\u8f83\u5927\u7684\u6570\u636e\u73af\u5883\u4e2d\u3002\u6df1\u5165\u5206\u6790\u8868\u660e\uff0cML4C \u4e0d\u4ec5\u786e\u4fdd\u4e86\u7406\u8bba\u4e0a\u7684\u53ef\u5b66\u4e60\u6027\uff0c\u8fd8\u5c55\u793a\u4e86\u5728\u4e0d\u540c\u6837\u672c\u91cf\u548c\u4e0d\u5b8c\u7f8e\u9aa8\u67b6\u8f93\u5165\u4e0b\u7684\u5bb9\u9519\u6027\u548c\u7a33\u5b9a\u6027\uff0c\u663e\u793a\u51fa\u5176\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u6f5c\u529b\u3002\u4f5c\u4e3a\u7814\u7a76\u5458\u4eec\u5728\u63a2\u7d22\u76d1\u7763\u56e0\u679c\u5b66\u4e60\u4e0a\u8fc8\u51fa\u7684\u7b2c\u4e00\u6b65\uff0cML4C \u521d\u8bd5\u950b\u8292\u4fbf\u53d6\u5f97\u4e86\u9f13\u821e\u4eba\u5fc3\u7684\u6210\u7ee9\u3002<\/p>\n\n\n\n<p><em>ML4C: Seeing Causality Through Latent Vicinity. Haoyue Dai, Rui Ding, Yuanyuan Jiang, Shi Han, Dongmei Zhang. SDM 2023<\/em><\/p>\n\n\n\n<p><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/epubs.siam.org\/doi\/10.1137\/1.9781611977653.ch26\"><em>https:\/\/epubs.siam.org\/doi\/10.1137\/1.9781611977653.ch26<\/em><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ml4s-\u4ece\u8bad\u7ec3\u6570\u636e\u51fa\u53d1\u6784\u5efa\u51c6\u786e\u7684\u56e0\u679c\u9aa8\u67b6\">ML4S\uff1a\u4ece\u8bad\u7ec3\u6570\u636e\u51fa\u53d1\u6784\u5efa\u51c6\u786e\u7684\u56e0\u679c\u9aa8\u67b6<\/h2>\n\n\n\n<p>\u5c3d\u7ba1\u5728\u8bc6\u522b V-\u7ed3\u6784\u4e2d\u83b7\u5f97\u4e86\u5353\u6709\u6210\u6548\u7684\u8fdb\u5c55\uff0c\u4f46\u5f53\u7814\u7a76\u5458\u4eec\u5c1d\u8bd5\u5c06\u7c7b\u4f3c\u7684\u601d\u8def\u5e94\u7528\u5230\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\u4efb\u52a1\u65f6\uff0c\u5374\u9047\u5230\u4e86\u76f8\u5f53\u5927\u7684\u963b\u788d\u3002\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\u65e8\u5728\u4ece\u89c2\u6d4b\u6570\u636e\u4e2d\u8bc6\u522b\u56e0\u679c\u56fe\u7684\u65e0\u5411\u56fe\uff08\u5373\u56e0\u679c\u9aa8\u67b6\uff0ccausal skeleton\uff09\uff0c\u8fd9\u662f\u56e0\u679c\u53d1\u73b0\u548c\u8bb8\u591a\u5176\u4ed6\u4e0b\u6e38\u5e94\u7528\u7684\u5173\u952e\u6b65\u9aa4\u3002\u7136\u800c\uff0cML4C \u4f9d\u8d56\u4e8e\u968f\u673a\u751f\u6210\u7684\u6570\u636e\uff0c\u5f80\u5f80\u4e0e test-time \u6570\u636e\u5b58\u5728\u8f83\u5927\u7684\u504f\u79bb\u3002\u4f8b\u5982\u5728\u56fe1\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u53d1\u73b0\u4e0d\u8bba\u662f\u57fa\u4e8e ER \u6216\u8005\u662f SF \u7b49\u968f\u673a\u6a21\u578b\u751f\u6210\u7684\u8bad\u7ec3\u6570\u636e\u4e0a\uff0c\u90fd\u96be\u4ee5\u7cbe\u51c6\u6062\u590d\u6b63\u786e\u7684\u56e0\u679c\u9aa8\u67b6\u7ed3\u6784\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"285\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1-1024x285.png\" alt=\"table\" class=\"wp-image-1078467\" style=\"width:736px;height:auto\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1-1024x285.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1-300x84.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1-768x214.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1-240x67.png 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-1.png 1164w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">\u56fe1\uff1aML4S \u6a21\u578b\u5728\u4e0d\u540c\u6d4b\u8bd5\u6570\u636e\uff08\u884c\uff09\u548c\u4e0d\u540c\u7684\u8bad\u7ec3\u6570\u636e\u5206\u5e03\uff08\u5217\uff09\u4e0b\u7684\u8868\u73b0\uff08F1-Score\uff09<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u5728 test-time training \u7b49\u5b66\u4e60\u8303\u5f0f\u7684\u542f\u53d1\u4e0b\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u57fa\u4e8e Vicinal \u56fe\u7684\u5b66\u4e60\u7b97\u6cd5\uff0c\u663e\u8457\u63d0\u9ad8\u4e86 ML4S \u6a21\u578b\u7684\u9884\u6d4b\u7cbe\u5ea6\u3002Vicinal \u56fe\u7684\u751f\u6210\u81f4\u529b\u4e8e\u521b\u5efa\u4e0e\u6d4b\u8bd5\u6570\u636e\uff08\u5373 test-time \u6570\u636e\uff09\u5206\u5e03\u76f8\u4f3c\u7684\u56fe\u7ed3\u6784\uff0c\u7528\u4e8e\u8bad\u7ec3\u76d1\u7763\u5b66\u4e60\u6a21\u578b\u3002\u8fd9\u4e9b\u56fe\u7ed3\u6784\u4e0d\u4ec5\u4fdd\u7559\u4e86\u539f\u59cb\u6d4b\u8bd5\u6570\u636e\u7684\u7edf\u8ba1\u7279\u6027\uff0c\u8fd8\u80fd\u6709\u6548\u5e94\u5bf9\u201c\u9886\u57df\u8f6c\u79fb\u201d\uff08domain shift\uff09\u95ee\u9898\uff0c\u4ece\u800c\u63d0\u9ad8\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\u7684\u51c6\u786e\u6027\u548c\u9c81\u68d2\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"831\" height=\"210\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-2.jpg\" alt=\"\u751f\u6210 Vicinal \u56fe\" class=\"wp-image-1078470\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-2.jpg 831w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-2-300x76.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-2-768x194.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-2-240x61.jpg 240w\" sizes=\"auto, (max-width: 831px) 100vw, 831px\" \/><figcaption class=\"wp-element-caption\">\u56fe2\uff1a\u751f\u6210 Vicinal \u56fe<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5728\u6280\u672f\u8bbe\u8ba1\u4e0a\uff0c\u7814\u7a76\u5458\u4eec\u91c7\u7528\u4e86\u73b0\u6709\u56e0\u679c\u53d1\u73b0\u7b97\u6cd5\uff08\u5982 PC \u7b97\u6cd5\uff09\u4ece test-time \u89c2\u6d4b\u6570\u636e\u4e2d\u5b66\u4e60\u4ee3\u7406\u56e0\u679c\u56fe\uff0c\u5e76\u7528\u6781\u5927\u4f3c\u7136\u4f30\u8ba1\uff08MLE\uff09\u65b9\u6cd5\u4f30\u7b97\u5176\u6bcf\u4e2a\u8282\u70b9\u7684\u6761\u4ef6\u6982\u7387\u8868\uff08Conditional Probability Table\uff0cCPT\uff09\u4ee5\u53c2\u6570\u5316\uff0c\u968f\u540e\uff0c\u901a\u8fc7\u5bf9\u8be5\u53c2\u6570\u5316\u56e0\u679c\u56fe\u6240\u4ee3\u8868\u7684\u8054\u5408\u6982\u7387\u5206\u5e03\u8fdb\u884c\u91c7\u6837\uff0c\u751f\u6210\u4e0e\u89c2\u6d4b\u6570\u636e\u5206\u5e03\u76f8\u4f3c\u7684\u5927\u91cf\u6837\u672c\u3002\u540c\u65f6\uff0c\u4e3a\u4e86\u51cf\u5c11\u4ee3\u7406\u56e0\u679c\u56fe\u7684\u504f\u5dee\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u901a\u8fc7\u63d2\u5165\u548c\u5220\u9664\u8fb9\u7b49\u53d8\u5f02\uff08mutation\uff09\u7b56\u7565\u751f\u6210\u4e86\u591a\u6837\u6027\u7684\u90bb\u8fd1\u56fe\uff0c\u5e76\u91cd\u65b0\u5206\u914d CPT \u4ee5\u4fdd\u6301\u5206\u5e03\u76f8\u4f3c\u6027\u3002\u8fd9\u79cd\u6d4b\u8bd5\u65f6\u8bad\u7ec3\u6837\u672c\u7684\u83b7\u53d6\u548c\u5b66\u4e60\uff08test-time training\uff09\u7684\u6280\u672f\uff0c\u65e8\u5728\u514b\u670d\u7531\u4e8e\u6d4b\u8bd5\u6837\u672c\u4e0d\u670d\u4ece\u8bad\u7ec3\u6570\u636e\u5206\u5e03\u800c\u5bfc\u81f4\u7684\u673a\u5668\u5b66\u4e60\u6cdb\u5316\u95ee\u9898\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u4f18\u5316\u4e86\u4efb\u52a1\u76ee\u6807\u3001\u6a21\u578b\u67b6\u6784\u548c\u7279\u5f81\u63d0\u53d6\uff0c\u4f8b\u5982\u63d0\u51fa\u4f7f\u7528\u987a\u5e8f\u7ea7\u8054\u5206\u7c7b\u5668\uff08cascade classifier\uff09\u67b6\u6784\u6765\u9884\u6d4b\u56e0\u679c\u9aa8\u67b6\u4e2d\u4e24\u4e2a\u8282\u70b9\u7684\u90bb\u63a5\u5173\u7cfb\uff0c\u8bbe\u8ba1\u5b9a\u91cf\u4e14\u7ed3\u6784\u611f\u77e5\u7684\u7279\u5f81\u63d0\u53d6\u65b9\u6cd5\u6765\u83b7\u5f97\u66f4\u7ec6\u7c92\u5ea6\u7684\u6570\u636e\u4fe1\u606f\uff08\u5177\u4f53\u6280\u672f\u7ec6\u8282\u53ef\u89c1\u8bba\u6587\u539f\u6587\uff09\u3002<\/p>\n\n\n\n<p>\u56fe3\u5448\u73b0\u4e86 ML4S \u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0b\u4e0e SOTA \u7b97\u6cd5\u7684\u4f18\u52a3\u6bd4\u8f83\uff0c\u5176\u4e2d\u84dd\u8272\u683c\u5b50\u8868\u793a ML4S \u5728\u8be5\u6570\u636e\u96c6\u4e0b\u6240\u751f\u6210\u7684\u56e0\u679c\u9aa8\u67b6\u7684 F1 Score \u9ad8\u4e8e\u8be5\u7b97\u6cd5\uff0c\u800c\u7ea2\u8272\/\u6a59\u8272\u683c\u5b50\u8868\u793a ML4S \u7684 F1 Score \u4f4e\u4e8e\u6240\u6bd4\u8f83\u7684\u7b97\u6cd5\u3002\u8be5\u56fe\u8868\u660e\uff0cML4S \u5728\u7edd\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u5747\u4f18\u4e8e\u73b0\u6709\u7684\u65b9\u6cd5\uff0c\u5c24\u5176\u662f\u5728\u5904\u7406\u8282\u70b9\u6570\u91cf\u8f83\u591a\u7684\u6570\u636e\u96c6\u65f6\uff0c\u8868\u73b0\u5c24\u4e3a\u7a81\u51fa\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"646\" height=\"428\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-3.jpg\" alt=\"ML4S \u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0b\u4e0e SOTA \u7b97\u6cd5\u7684\u6bd4\u8f83\" class=\"wp-image-1078473\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-3.jpg 646w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-3-300x199.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-3-240x159.jpg 240w\" sizes=\"auto, (max-width: 646px) 100vw, 646px\" \/><figcaption class=\"wp-element-caption\">\u56fe3\uff1aML4S \u5728\u4e0d\u540c\u6570\u636e\u96c6\u4e0b\u4e0e SOTA \u7b97\u6cd5\u7684\u6bd4\u8f83<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p><em>ML4S: Learning Causal Skeleton from Vicinal Graphs. Pingchuan Ma, Rui Ding, Haoyue Dai, Yuanyuan Jiang, Shuai Wang, Shi Han, and Dongmei Zhang. KDD 2022<\/em><\/p>\n\n\n\n<p><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/dl.acm.org\/doi\/10.1145\/3534678.3539447\"><em>https:\/\/dl.acm.org\/doi\/10.1145\/3534678.3539447<\/em><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"spot-\u56e0\u679c\u9aa8\u67b6\u5f15\u5bfc\u7684\u53ef\u5fae\u56e0\u679c\u53d1\u73b0\">SPOT\uff1a\u56e0\u679c\u9aa8\u67b6\u5f15\u5bfc\u7684\u53ef\u5fae\u56e0\u679c\u53d1\u73b0<\/h2>\n\n\n\n<p>\u5728\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u5b58\u5728\u7684\u60c5\u51b5\u4e0b\uff0c\u56e0\u679c\u53d1\u73b0\u66f4\u5177\u6311\u6218\u6027\u3002\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u5e7f\u6cdb\u5b58\u5728\u4e8e\u771f\u5b9e\u6570\u636e\u4e2d\uff0c\u5176\u8868\u793a\u4e86\u4e00\u4e2a\u6570\u636e\u96c6\u5185\u7684\u4e24\u4e2a\u53d8\u91cf\u540c\u65f6\u53d7\u5230\u540c\u4e00\u5916\u90e8\u53d8\u91cf\u7684\u5f71\u54cd\u3002\u7814\u7a76\u5458\u4eec\u6307\u51fa\uff0c\u867d\u7136\u73b0\u6709\u7684\u53ef\u5fae\u5206\u56e0\u679c\u53d1\u73b0\u65b9\u6cd5\u5728\u5b66\u4e60\u6709\u5411\u65e0\u73af\u56fe\uff08DAGs\uff09\u65b9\u9762\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\uff0c\u4f46\u5728\u5904\u7406\u5305\u542b\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u7684\u5b9e\u9645\u6570\u636e\u96c6\u65f6\u4ecd\u7136\u53d7\u5230\u9650\u5236\u3002\u8fd9\u662f\u56e0\u4e3a\u73b0\u6709\u7684\u65b9\u6cd5\u96be\u4ee5\u6269\u5c55\u5230\u8f83\u5927\u7684\u6570\u636e\u96c6\uff08\u5982\u5305\u542b50\u4e2a\u4ee5\u4e0a\u53d8\u91cf\u7684\u6570\u636e\u96c6\uff09\u4e0a\u3002\u5728 ML4S \u7684\u57fa\u7840\u4e0a\uff0c\u7814\u7a76\u5458\u4eec\u53d1\u73b0\uff0c\u56e0\u679c\u9aa8\u67b6\u5728\u63d0\u9ad8\u51c6\u786e\u6027\u548c\u51cf\u5c11\u4f18\u5316\u8fc7\u7a0b\u7684\u641c\u7d22\u7a7a\u95f4\u65b9\u9762\u5177\u6709\u8f83\u5927\u6f5c\u529b\uff0c\u5982\u679c\u80fd\u591f\u4e8b\u5148\u77e5\u9053\u6570\u636e\u5bf9\u5e94\u7684\u56e0\u679c\u9aa8\u67b6\uff0c\u5c31\u53ef\u4ee5\u5c06\u53ef\u5fae\u56e0\u679c\u53d1\u73b0\u7684\u4f18\u5316\u8fc7\u7a0b\u52a0\u5165\u989d\u5916\u7684\u7b49\u5f0f\u7ea6\u675f\uff0c\u4ece\u800c\u63d0\u5347\u53ef\u5fae\u5206\u56e0\u679c\u53d1\u73b0\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p>\u5728\u8fd9\u4e2a\u601d\u8def\u4e0b\uff0c\u7814\u7a76\u5458\u4eec\u8bbe\u8ba1\u4e86 SPOT \u65b9\u6cd5\uff0c\u5176\u5173\u952e\u601d\u60f3\u662f\u5229\u7528\u9aa8\u67b6\u540e\u9a8c\uff08skeleton posterior\uff09\u6765\u5f15\u5bfc\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u4e0b\u7684\u53ef\u5fae\u5206\u56e0\u679c\u53d1\u73b0\u3002\u4e0e\u4f20\u7edf\u7684\u201c\u70b9\u4f30\u8ba1\u201d\uff08point estimation\uff09\u4e0d\u540c\uff0cSPOT \u65e8\u5728\u4f30\u8ba1\u7ed9\u5b9a\u6570\u636e\u96c6\u7684\u56e0\u679c\u9aa8\u67b6\u7684\u540e\u9a8c\u5206\u5e03 P(S\u2223D)\uff0c\u5e76\u6839\u636e P(S\u2223D) \u7684\u6570\u503c\uff0c\u5bf9\u56e0\u679c\u56fe\u90bb\u63a5\u77e9\u9635\u4e2d\u54cd\u5e94\u6570\u503c\u7684\u68af\u5ea6\u8fdb\u884c\u540e\u5904\u7406\uff0c\u4ece\u800c\u5f15\u5bfc\u4f18\u5316\u8fc7\u7a0b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"992\" height=\"218\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-4.jpg\" alt=\"\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u4e0b\u7684\u56e0\u679c\u56fe\u7ed3\u6784\u8868\u793a\uff0c(a) \u4e2d Z \u662f\u4e00\u4e2a\u53ef\u89c2\u6d4b\u7684\u6df7\u6742\u53d8\u91cf\uff0c (b) \u4e2d Z \u662f\u4e00\u4e2a\u4e0d\u53ef\u9884\u6d4b\u7684\u6f5c\u5728\u6df7\u6742\u53d8\u91cf\uff0c\u56e0\u6b64\u9700\u8981\u4f7f\u7528\u53cc\u5411\u7bad\u5934 \u2194 \u8868\u793a X \u548c Y \u4e4b\u95f4\u7531\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u5f15\u8d77\u7684\u5173\u8054\" class=\"wp-image-1078476\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-4.jpg 992w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-4-300x66.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-4-768x169.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-4-240x53.jpg 240w\" sizes=\"auto, (max-width: 992px) 100vw, 992px\" \/><figcaption class=\"wp-element-caption\">\u56fe4\uff1a\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u4e0b\u7684\u56e0\u679c\u56fe\u7ed3\u6784\u8868\u793a\uff0c(a) \u4e2d Z \u662f\u4e00\u4e2a\u53ef\u89c2\u6d4b\u7684\u6df7\u6742\u53d8\u91cf\uff0c (b) \u4e2d Z \u662f\u4e00\u4e2a\u4e0d\u53ef\u9884\u6d4b\u7684\u6f5c\u5728\u6df7\u6742\u53d8\u91cf\uff0c\u56e0\u6b64\u9700\u8981\u4f7f\u7528\u53cc\u5411\u7bad\u5934 \u2194 \u8868\u793a X \u548c Y \u4e4b\u95f4\u7531\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u5f15\u8d77\u7684\u5173\u8054<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5728 SPOT \u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u5f15\u5165\u4e86\u52a8\u6001\u548c\u9759\u6001\u6570\u636e\u751f\u6210\u6a21\u5f0f\uff0c\u4ee5\u5e94\u5bf9\u539f\u6709 ML4S \u65b9\u6cd5\u4e2d\u53ef\u80fd\u51fa\u73b0\u7684\u6a21\u578b\u504f\u5dee\u3001\u6df7\u6742\u56e0\u7d20\u60c5\u51b5\u4e0d\u9002\u7528\u7684\u95ee\u9898\u3002\u540c\u65f6\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u8bbe\u8ba1\u4e86\u4e00\u79cd\u968f\u673a\u4f18\u5316\u7b56\u7565\uff0c\u5176\u4f1a\u6839\u636e P(S_ij\u2223D)\u7684\u503c\uff0c\u4ee5\u7279\u5b9a\u7684\u6982\u7387\u5c06 \u0394_ij \u8d4b\u503c\u4e3a\u96f6\uff0c\u6765\u5bf9\u4f18\u5316\u53d8\u91cf\u8fdb\u884c\u6bcf\u4e00\u6b65\u66f4\u65b0\u3002<\/p>\n\n\n\n<p>\u56fe5\u5c55\u793a\u4e86\u4e0d\u540c\u6a21\u5f0f\u5bf9\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u4e0b\u53ef\u5fae\u56e0\u679c\u53d1\u73b0\u7684\u6027\u80fd\u63d0\u5347\u3002\u57fa\u7ebf\u65b9\u6cd5 ABIC \u5728\u4e09\u4e2a\u6307\u6807\u4e0a\u7684\u8868\u73b0\u76f8\u5bf9\u8f83\u4f4e\uff0c\u8fd9\u8868\u660e\u4e0d\u5229\u7528\u4efb\u4f55\u9aa8\u67b6\u540e\u9a8c\u4fe1\u606f\u7684\u65b9\u6cd5\u5728\u5904\u7406\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u65f6\u6548\u679c\u6709\u9650\u3002\u4f7f\u7528\u771f\u5b9e\u9aa8\u67b6\u8fdb\u884c\u4f18\u5316\u7684\u65b9\u6cd5\uff08w\/ True Skeleton\uff09\u5728\u6240\u6709\u6307\u6807\u4e0a\u90fd\u8fbe\u5230\u4e86\u6700\u9ad8\u503c\uff08\u7406\u60f3\u5316\u7684\u4e0a\u9650\u6027\u80fd\uff09\uff0c\u663e\u8457\u4f18\u4e8e\u5176\u4ed6\u65b9\u6cd5\uff0c\u8868\u660e\u51c6\u786e\u7684\u9aa8\u67b6\u4fe1\u606f\u5bf9\u56e0\u679c\u53d1\u73b0\u6027\u80fd\u7684\u63d0\u5347\u662f\u975e\u5e38\u91cd\u8981\u7684\u3002\u5229\u7528 FCI \u7b97\u6cd5\u4f30\u8ba1\u7684\u9aa8\u67b6\u8fdb\u884c\u4f18\u5316\u7684\u65b9\u6cd5\uff08w\/ FCI\uff09\uff0c\u76f8\u6bd4 ABIC \u7565\u6709\u63d0\u5347\uff0c\u4f46\u63d0\u5347\u5e45\u5ea6\u6709\u9650\uff0c\u53ef\u80fd\u662f\u56e0\u4e3a FCI \u7b97\u6cd5\u5728\u5904\u7406\u590d\u6742\u6570\u636e\u65f6\u7684\u9aa8\u67b6\u4f30\u8ba1\u5e76\u4e0d\u5b8c\u5168\u51c6\u786e\u3002\u5229\u7528\u9aa8\u67b6\u540e\u9a8c\u6307\u5bfc\u4f18\u5316\u8fc7\u7a0b\u7684 SPOT \u65b9\u6cd5\u5728\u6240\u6709\u6307\u6807\u4e0a\u5747\u4f18\u4e8e ABIC \u548c w\/ FCI\uff0c\u63a5\u8fd1\u7406\u60f3\u5316\u7684 w\/ True Skeleton\uff0c\u8fd9\u8868\u660e\u9aa8\u67b6\u540e\u9a8c\u7684\u5f15\u5bfc\u80fd\u591f\u6709\u6548\u63d0\u5347\u56e0\u679c\u53d1\u73b0\u7684\u51c6\u786e\u6027\uff0c\u5c24\u5176\u662f\u5728\u5904\u7406\u590d\u6742\u548c\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u7684\u60c5\u51b5\u4e0b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1004\" height=\"268\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-5.png\" alt=\"table\" class=\"wp-image-1078479\" style=\"width:721px;height:auto\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-5.png 1004w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-5-300x80.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-5-768x205.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2024\/08\/causal-discovery-5-240x64.png 240w\" sizes=\"auto, (max-width: 1004px) 100vw, 1004px\" \/><figcaption class=\"wp-element-caption\">\u56fe5\uff1a\u4e0d\u540c\u6a21\u5f0f\u5bf9\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u4e0b\u53ef\u5fae\u56e0\u679c\u53d1\u73b0\u7684\u6027\u80fd\u63d0\u5347<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u7efc\u4e0a\u6240\u8ff0\uff0c\u56fe5\u7684\u7ed3\u679c\u8868\u660e\uff0c\u4f7f\u7528\u9aa8\u67b6\u540e\u9a8c\u4fe1\u606f\uff08\u5982 SPOT \u65b9\u6cd5\uff09\u663e\u8457\u63d0\u9ad8\u4e86\u56e0\u679c\u53d1\u73b0\u7684\u6027\u80fd\uff0c\u7279\u522b\u662f\u5728\u6709\u5411\u8fb9\u7bad\u5934\u548c\u5c3e\u90e8\u7684\u51c6\u786e\u6027\u4e0a\u3002\u867d\u7136\u7406\u60f3\u5316\u7684\u771f\u5b9e\u9aa8\u67b6\u63d0\u4f9b\u4e86\u6700\u597d\u7684\u6027\u80fd\uff0c\u4f46 SPOT \u65b9\u6cd5\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u5c55\u793a\u4e86\u5176\u6709\u6548\u6027\u548c\u4f18\u8d8a\u6027\uff0c\u8868\u660e\u5728\u56e0\u679c\u53d1\u73b0\u4efb\u52a1\u4e2d\u5f15\u5165\u9aa8\u67b6\u540e\u9a8c\u662f\u4e00\u79cd\u975e\u5e38\u6709\u524d\u666f\u7684\u7b56\u7565\u3002<\/p>\n\n\n\n<p><em>Scalable Differentiable Causal Discovery in the Presence of Latent Confounders with Skeleton Posterior. Pingchuan Ma, Rui Ding, Qiang Fu, Jiaru Zhang, Shuai Wang, Shi Han, and Dongmei Zhang. KDD 2024<\/em><\/p>\n\n\n\n<p><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/arxiv.org\/abs\/2406.10537\"><em>https:\/\/arxiv.org\/abs\/2406.10537<\/em><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u968f\u7740\u6570\u636e\u79d1\u5b66\u548c\u673a\u5668\u5b66\u4e60\u6280\u672f\u7684\u8fc5\u731b\u53d1\u5c55\uff0c\u6570\u636e\u9a71\u52a8\u7684\u56e0\u679c\u53d1\u73b0\u5728\u79d1\u5b66\u7814\u7a76\u548c\u5b9e\u9645\u5e94\u7528\u4e2d\u5c55\u73b0\u51fa\u5de8\u5927\u7684\u6f5c\u529b\u3002\u7136\u800c\uff0c\u5f53\u524d\u7684\u56e0\u679c\u53d1\u73b0\u65b9\u6cd5\u4ecd\u7136\u9762\u4e34\u8bb8\u591a\u6311\u6218\uff0c\u5305\u62ec\u9ad8\u7ef4\u6570\u636e\u7684\u5904\u7406\u3001\u590d\u6742\u56e0\u679c\u7ed3\u6784\u7684\u8bc6\u522b\u4ee5\u53ca\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u7684\u5f71\u54cd\u3002\u901a\u8fc7\u5f15\u5165\u5148\u8fdb\u7684\u673a\u5668\u5b66\u4e60\u6280\u672f\u548c\u521b\u65b0\u7684\u65b9\u6cd5\uff0c\u7814\u7a76\u5458\u4eec\u6b63\u5728\u9010\u6b65\u7a81\u7834\u8fd9\u4e9b\u74f6\u9888\uff0c\u5f00\u542f\u6570\u636e\u9a71\u52a8\u56e0\u679c\u53d1\u73b0\u7684\u65b0\u8303\u5f0f\u3002\u8fd9\u4e0d\u4ec5\u80fd\u4e3a\u56e0\u679c\u5206\u6790\u63d0\u4f9b\u65b0\u7684\u601d\u8def\uff0c\u4e5f\u5c06\u4f1a\u4e3a\u79d1\u5b66\u7814\u7a76\u548c\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u56e0\u679c\u53d1\u73b0\u5e26\u6765\u65b0\u7684\u53ef\u80fd\u6027\uff0c\u5bf9\u672a\u6765\u7684\u56e0\u679c\u7814\u7a76\u4ea7\u751f\u6df1\u8fdc\u5f71\u54cd\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4f5c\u8005\uff1aDKI\u7ec4 \u7f16\u8005\u6309\uff1a\u5728\u5f53\u4eca\u6570\u636e\u9a71\u52a8\u7684\u4e16\u754c\uff0c\u7406\u89e3\u590d\u6742\u7cfb\u7edf\u4e2d\u7684\u56e0\u679c\u5173\u7cfb\u662f\u79d1\u5b66\u7814\u7a76\u548c\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u5173\u952e\u6311\u6218\u3002\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df\uff0c\u56e0\u679c\u63a8\u7406\u80fd\u529b\u66f4\u662f\u6210\u4e3a\u4e00\u4e2a\u70ed\u95e8\u8bdd\u9898\u3002\u5982\u4f55\u63ed\u793a\u6570\u636e\u80cc\u540e\u56e0\u679c\u673a\u5236\u7684\u5173\u952e\uff1f\u5982\u4f55\u5229\u7528\u6570\u636e\u5b9e\u73b0\u56e0\u679c\u53d1\u73b0\u7684\u7a81\u7834\uff1f\u4e3a\u56de\u7b54\u8fd9\u4e9b\u95ee\u9898\uff0c\u6765\u81ea\u5fae\u8f6f DKI\uff08Data, Knowledge and Intelligence\uff0c\u6570\u636e\u3001\u77e5\u8bc6\u4e0e\u667a\u80fd\uff09\u9886\u57df\u7684\u7814\u7a76\u5458\u4eec\u5728\u8fdb\u884c\u4e86\u6301\u7eed\u800c\u6df1\u5165\u7684\u63a2\u7d22\uff0c\u5176\u76f8\u5173\u6210\u679c\u53d1\u8868\u5728 AAAI 2020\u3001KDD 2022\u3001SDM 2023\u3001SIGMOD 2023 \u548c KDD 2024 \u7b49\u56fd\u9645\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae\u4e0a\u3002\u540c\u65f6\uff0c\u8be5\u7cfb\u5217\u5de5\u4f5c\u4e5f\u5728 Microsoft Power BI \u7b49\u4ea7\u54c1\u4e2d\u5f97\u5230\u5e94\u7528\u3002 \u672c\u7bc7\u6587\u7ae0\u5c06\u4e3a\u5927\u5bb6\u4ecb\u7ecd\u7531\u5fae\u8f6f\uff08\u4e9a\u6d32\uff09\u4e92\u8054\u7f51\u5de5\u7a0b\u9662 DKI \u7ec4\u548c\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u3001\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff08CMU\uff09\u5408\u4f5c\u5b8c\u6210\u7684\u524d\u6cbf\u7814\u7a76\u9879\u76ee\uff0c\u8be5\u9879\u76ee\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u673a\u5668\u76d1\u7763\u5b66\u4e60\u7684\u56e0\u679c\u53d1\u73b0\u65b0\u8303\u5f0f\uff0c\u5305\u542b\u5bf9\u56e0\u679c\u5b9a\u5411\u3001\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\u3001\u5904\u7406\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u65f6\u7684\u56e0\u679c\u53d1\u73b0\u4e09\u4e2a\u5173\u952e\u95ee\u9898\u7684\u56de\u7b54\u3002 \u56e0\u679c\u53d1\u73b0\uff08causal discovery\uff09\u662f\u6570\u636e\u79d1\u5b66\u548c\u673a\u5668\u5b66\u4e60\u4e2d\u7684\u5173\u952e\u9886\u57df\uff0c\u5176\u4efb\u52a1\u662f\u901a\u8fc7\u7b97\u6cd5\u4ece\u6570\u636e\u4e2d\u63d0\u53d6\u56e0\u679c\u8054\u7cfb\uff0c\u5e76\u7528\u56e0\u679c\u56fe\u8fdb\u884c\u8868\u793a\u3002\u56e0\u679c\u53d1\u73b0\u5bf9\u4e8e\u7406\u89e3\u590d\u6742\u7cfb\u7edf\u3001\u652f\u6301\u653f\u7b56\u5236\u5b9a\u3001\u533b\u5b66\u548c\u793e\u4f1a\u79d1\u5b66\u7814\u7a76\u7b49\u9886\u57df\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002 \u4f20\u7edf\u4e0a\uff0c\u57fa\u4e8e\u89c2\u6d4b\u6570\u636e\u7684\u56e0\u679c\u53d1\u73b0\u4e3b\u8981\u6709\u4e09\u7c7b\u65b9\u6cd5\uff1a\u57fa\u4e8e\u7ea6\u675f\u3001\u8bc4\u5206\u548c\u8fde\u7eed\u4f18\u5316\u3002 \u8fd9\u4e9b\u4f20\u7edf\u65b9\u6cd5\u4f53\u73b0\u4e86\u56e0\u679c\u53d1\u73b0\u9886\u57df\u8fc7\u53bb\u7684\u4e3b\u6d41\u65b9\u6cd5\u8bba\uff0c\u5373\u5c06\u4eba\u7c7b\u7684\u667a\u6167\u7f16\u7801\u6210\u5de7\u5999\u7684\u7b97\u6cd5\uff0c\u6765\u63d0\u9ad8\u56e0\u679c\u53d1\u73b0\u4efb\u52a1\u7684\u6027\u80fd\u3002\u7136\u800c\uff0c\u6765\u81ea\u5fae\u8f6f\uff08\u4e9a\u6d32\uff09\u4e92\u8054\u7f51\u5de5\u7a0b\u9662 DKI \u7ec4\u3001\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u548c\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff08CMU\uff09\u7684\u7814\u7a76\u5458\u4eec\u5c1d\u8bd5\u7740\u6539\u53d8\u601d\u8def\uff0c\u4ece\u57fa\u4e8e\u5b66\u4e60\u7684\u65b9\u6cd5\u8bba\uff08learning-based\uff09\u5165\u624b\uff0c\u7740\u773c\u4e8e\u5b9e\u8bc1\u6027\u80fd\uff08empirical performance-driven\uff09\u6765\u63a2\u8ba8\u4e00\u79cd\u57fa\u4e8e\u76d1\u7763\u5b66\u4e60\u7684\u56e0\u679c\u53d1\u73b0\u65b0\u8303\u5f0f\uff08supervision-based causal discovery\uff09\u3002\u8be5\u8303\u5f0f\u805a\u7126\u8bad\u7ec3\u6570\u636e\u83b7\u53d6\u3001\u5b66\u4e60\u4efb\u52a1\u5b9a\u4e49\u3001\u5b66\u4e60\u76ee\u6807\u8bbe\u8ba1\u3001\u5b9e\u73b0\u7aef\u5230\u7aef\u7684\u56e0\u679c\u5173\u7cfb\u8bc6\u522b\uff0c\u540c\u65f6\u4e25\u8083\u5bf9\u5f85\u56e0\u679c\u53d1\u73b0\u7684\u6838\u5fc3\u6982\u5ff5\uff1a\u5206\u6790\u53ef\u5b66\u4e60\u6027 \uff08learnability\uff09 \u4e0e\u56e0\u679c\u7ed3\u6784\u53ef\u8fa8\u8bc6\u6027\uff08identifiability\uff09\u7684\u5173\u7cfb\u3002 \u6ce8\uff1a\u57fa\u4e8e\u5b66\u4e60\u7684\u65b9\u6cd5\u8bba\u6210\u4e3a\u53ef\u80fd\u7684\u524d\u63d0\u662f\uff0c\u56e0\u679c\u53d1\u73b0\u4efb\u52a1\u5177\u5907\u201c\u81ea\u76d1\u7763\u5b66\u4e60\u201d\u7684\u7279\u8d28\uff1a\u53ef\u4ee5\u81ea\u52a8\u5408\u6210\u5927\u91cf \u914d\u5bf9\uff0c\u5176\u4e2d Gi \u662f\u56e0\u679c\u56fe\uff0cDi \u662f\u4e0e\u4e4b\u76f8\u5bb9\u7684\u89c2\u6d4b\u6570\u636e\u3002\u7136\u540e\u4ee4 Di \u4f5c\u4e3a\u6a21\u578b\u8f93\u5165\uff0cGi \u4f5c\u4e3a\u6807\u6ce8\u3002 \u4e0b\u9762\uff0c\u6211\u4eec\u4e00\u8d77\u6765\u4e86\u89e3\u4e00\u4e0b\u5982\u4f55\u5728\u56e0\u679c\u5b9a\u5411\uff08causal orientation\uff09\u3001\u56e0\u679c\u9aa8\u67b6\u5b66\u4e60\uff08casual skeleton learning\uff09\u4ee5\u53ca\u5904\u7406\u6f5c\u5728\u6df7\u6742\u56e0\u7d20\u65f6\u7684\u56e0\u679c\u53d1\u73b0\uff08causal discovery in the presence of latent confounders\uff09\u9886\u57df\u7684\u4e00\u7cfb\u5217\u63a2\u7d22\u4e2d\u5b8c\u6210\u5bf9\u8be5\u8303\u5f0f\u7684\u6784\u5efa\u3002\u8fd9\u4e5f\u5c06\u4e3a\u672a\u6765\u8be5\u9886\u57df\u7684\u63a2\u7d22\u63d0\u4f9b\u5d2d\u65b0\u7684\u601d\u8def\u3002 \u53ef\u8bc6\u522b\u6027\u662f\u6307\u6839\u636e\u89c2\u5bdf\u6570\u636e\u786e\u5b9a\u552f\u4e00\u6a21\u578b\u7684\u80fd\u529b\uff0c\u662f\u56e0\u679c\u53d1\u73b0\u7684\u91cd\u4e2d\u4e4b\u91cd\u3002\u5728\u4f20\u7edf\u7684\u56e0\u679c\u53d1\u73b0\u7b97\u6cd5\u4e2d\uff0c\u4eba\u4eec\u901a\u5e38\u57fa\u4e8e\u5bf9\u6570\u636e\u751f\u6210\u8fc7\u7a0b\u7684\u67d0\u4e9b\u5047\u8bbe\uff0c\u6765\u8bbe\u8ba1\u5408\u9002\u7684\u7279\u5f81\u4ee5\u8bc6\u522b\u56e0\u679c\u7ed3\u6784\uff0c\u4f8b\u5982\u57fa\u4e8e\u4e0d\u5bf9\u79f0\u6027\u3002\u7136\u800c\uff0c\u7814\u7a76\u5458\u4eec\u9009\u62e9\u6539\u53d8\u601d\u8def\uff0c\u4ece\u673a\u5668\u5b66\u4e60\u7684\u89c6\u89d2\u51fa\u53d1\uff0c\u5728\u6570\u636e\u4e2d\u76f4\u63a5\u5b66\u5f97\u8fd9\u4e9b\u7279\u5f81\u6765\u5e2e\u52a9\u8bc6\u522b\u56e0\u679c\u7ed3\u6784\uff0c\u4ece\u800c\u63d0\u5347\u6027\u80fd\u3002 \u8fd9\u5176\u4e2d\u7684\u5173\u952e\u5728\u4e8e\uff0c\u5b66\u4e60\u76ee\u6807\u5e94\u8be5\u5982\u4f55\u5b9a\u4e49\uff1f\u6362\u8a00\u4e4b\uff0c\u54ea\u4e9b\u6570\u636e\u7279\u5f81\u5e94\u4f5c\u4e3a\u8bc6\u522b\u5bf9\u8c61\u5e2e\u52a9\u5224\u65ad\u56e0\u679c\u7ed3\u6784\uff1f \u4e3a\u4e86\u56de\u7b54\u8fd9\u4e2a\u95ee\u9898\uff0c\u7814\u7a76\u5458\u4eec\u805a\u7126\u4e8e\u65e0\u9700\u53c2\u6570\u5316\u5047\u8bbe\u7684\u65b9\u6cd5\uff0c\u4ec5\u5229\u7528\u89c2\u6d4b\u6570\u636e\u7684\u6761\u4ef6\u72ec\u7acb\u5173\u7cfb\u6765\u63a8\u6d4b\u56e0\u679c\u7ed3\u6784\u3002\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u53d1\u73b0\u4e24\u4e2a\u56e0\u679c\u56fe\u5982 [&hellip;]<\/p>\n","protected":false},"author":34512,"featured_media":1078482,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1078461","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1078461","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":4,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1078461\/revisions"}],"predecessor-version":[{"id":1078500,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1078461\/revisions\/1078500"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1078482"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1078461"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1078461"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1078461"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1078461"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}