{"id":23269,"date":"2021-01-15T23:05:05","date_gmt":"2021-01-15T15:05:05","guid":{"rendered":"https:\/\/swarma.org\/?p=23269"},"modified":"2021-01-15T23:05:05","modified_gmt":"2021-01-15T15:05:05","slug":"%e5%9b%be%e7%bd%91%e7%bb%9c%e6%9c%aa%e6%9d%a5%e5%87%a0%e4%bd%95%ef%bc%9f12%e4%bd%8d%e5%ad%a6%e8%80%85%e5%af%84%e6%9c%9b2021%e5%a4%a7%e7%88%86%e5%8f%91","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=23269","title":{"rendered":"\u56fe\u7f51\u7edc\u672a\u6765\u51e0\u4f55\uff1f12\u4f4d\u5b66\u8005\u5bc4\u671b2021\u5927\u7206\u53d1"},"content":{"rendered":"
<\/p>\n

\u5bfc\u8bed<\/p>\n

\u8fd1\u6bb5\u65f6\u95f4\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\uff08GNN\uff09\u4f5c\u4e3a\u673a\u5668\u5b66\u4e60\u6700\u70ed\u95e8\u7684\u9886\u57df\u4e4b\u4e00\uff0c\u5531\u76db\u6709\u4e4b\uff0c\u5531\u8870\u4e5f\u4e0d\u5c11\u3002\u5c31\u5728\u524d\u4e24\u5929\uff0c\u4e00\u4f4d\u6570\u636e\u79d1\u5b66\u5bb6\u53d1\u8868\u4e86\u4e00\u7bc7\u9898\u4e3a\u201c\u4e3a\u4ec0\u4e48\u6211\u5bf9GNN\u4e0d\u611f\u5192\u201d\u7684\u6587\u7ae0\uff0c\u5386\u6570 GNN \u7684\u591a\u9879\u5f0a\u7aef\uff0c\u5728reddit\u4e0a\u5f15\u53d1\u5e7f\u6cdb\u8ba8\u8bba\u3002\u4f46\u5c3d\u7ba1\u5982\u6b64\uff0cGNN\u4f5c\u4e3a\u4e00\u9879\u65b9\u5174\u672a\u827e\u7684\u7814\u7a76\u9886\u57df\uff0c\u5728 2020\u5e74\u4f9d\u7136\u53d6\u5f97\u4e86\u8bf8\u591a\u8fdb\u5c55\uff0c\u57282021\u5e74\u4e5f\u4f9d\u7136\u5b58\u5728\u8bb8\u591a\u4e9f\u5f85\u6316\u6398\u7684\u8bfe\u9898\u3002

 Michael Bronstein | \u4f5c\u8005
\u667a\u6e90\u793e\u533a | \u6765\u6e90<\/p>\n

\u8fd1\u6bb5\u65f6\u95f4\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\uff08GNN\uff09\u4f5c\u4e3a\u673a\u5668\u5b66\u4e60\u6700\u70ed\u95e8\u7684\u9886\u57df\u4e4b\u4e00\uff0c\u5531\u76db\u6709\u4e4b\uff0c\u5531\u8870\u4e5f\u4e0d\u5c11\u3002 \u5c31\u5728\u524d\u4e24\u5929\uff0c\u4e00\u4f4d\u6570\u636e\u79d1\u5b66\u5bb6\u53d1\u8868\u4e86\u4e00\u7bc7\u9898\u4e3a\u201c\u4e3a\u4ec0\u4e48\u6211\u5bf9GNN\u4e0d\u611f\u5192\u201d\u7684\u6587\u7ae0\uff0c\u5386\u6570 GNN \u7684\u591a\u9879\u5f0a\u7aef\uff0c\u5728reddit\u4e0a\u5f15\u53d1\u5e7f\u6cdb\u8ba8\u8bba\u3002 \u8fc7\u53bb5\u5e74\u4e2d\uff0c\u5927\u591a\u6570GNN\u8bba\u6587\u63d0\u4f9b\u7684\u7ed3\u679c\uff0c\u5bf9\u4ece\u4e1a\u8005\u6ca1\u6709\u592a\u591a\u5e2e\u52a9\u3002\u7cdf\u7cd5\u7684\u56fe\u6570\u636e\u7ed3\u6784\u5168\u5c40\u65b9\u6cd5\uff0c\u662f\u4e2a\u6b7b\u80e1\u540c….. \u4f46\u5c3d\u7ba1\u5982\u6b64\uff0cGNN\u4f5c\u4e3a\u4e00\u9879\u65b9\u5174\u672a\u827e\u7684\u7814\u7a76\u9886\u57df\uff0c\u5728 2020\u5e74\u4f9d\u7136\u53d6\u5f97\u4e86\u8bf8\u591a\u8fdb\u5c55\uff0c\u57282021\u5e74\u4e5f\u4f9d\u7136\u5b58\u5728\u8bb8\u591a\u4e9f\u5f85\u6316\u6398\u7684\u8bfe\u9898\u3002 \u5728\u65b0\u5e74\u4f0a\u59cb\uff0c\u4ee5\u8272\u5217\u8ba1\u7b97\u673a\u79d1\u5b66\u5bb6 Michael Bronstein \u5f81\u6c42\u4e86 12 \u4f4d\u5728Graph ML\u53ca\u5176\u5e94\u7528\u9886\u57df\u505a\u51fa\u6770\u51fa\u5de5\u4f5c\u7684\u7814\u7a76\u4eba\u5458\u7684\u610f\u89c1\uff0c\u5bf9\u8fc7\u53bb\u4e00\u5e74 GML \u9886\u57df\u7684\u4eae\u70b9\u8fdb\u884c\u4e86\u603b\u7ed3\uff0c\u5e76\u9884\u6d4b\u4e862021\u5e74\u7684\u53d1\u5c55\uff0c\u5185\u5bb9\u6d89\u53caGNN\u7814\u7a76\u7684\u65b9\u65b9\u9762\u9762\u3002<\/p>\n

1.\u6d88\u606f\u4f20\u9012\u7684\u9650\u5236<\/p>\n


 Will Hamilton\uff0c\u9ea6\u5409\u5c14\u5927\u5b66\u7684\u52a9\u7406\u6559\u6388\uff0c\u8499\u7279\u5229\u5c14 CIFAR \u4e3b\u5e2d\uff0cGraphSAGE \u4f5c\u8005\u3002<\/p>\n

Will Hamilton\uff1a
2020\u5e74\uff0c\u56fe\u673a\u5668\u5b66\u4e60\u9886\u57df\u9762\u4e34\u7740\u6d88\u606f\u4f20\u9012\u6a21\u5f0f\u7684\u57fa\u672c\u9650\u5236\u3002 \u8fd9\u4e9b\u9650\u5236\u5305\u62ec\u6240\u8c13\u7684\u201c\u74f6\u9888\u201d\u95ee\u9898\uff0c\u8fc7\u5ea6\u5e73\u6ed1\u95ee\u9898\uff0c\u4ee5\u53ca\u8868\u5f81\u80fd\u529b\u7406\u8bba\u4e0a\u7684\u9650\u5236\u3002\u5c55\u671b\u672a\u6765\uff0c\u6211\u5e0c\u671b\u57282021\u5e74\uff0c\u6211\u4eec\u80fd\u591f\u4e3a\u56fe\u673a\u5668\u5b66\u4e60\u5bfb\u627e\u4e0b\u4e00\u4e2a\u91cd\u8981\u7684\u8303\u4f8b\u3002\u6211\u4e0d\u786e\u5b9a\u4e0b\u4e00\u4ee3\u56fe\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u5230\u5e95\u4f1a\u662f\u4ec0\u4e48\u6837\u5b50\uff0c\u4f46\u6211\u76f8\u4fe1\uff0c\u8981\u60f3\u53d6\u5f97\u8fdb\u5c55\uff0c\u5c31\u9700\u8981\u6253\u78342020\u5e74\u53ca\u4e4b\u524d\u7edf\u6cbb\u8be5\u9886\u57df\u7684\u6d88\u606f\u4f20\u9012\u6a21\u5f0f\u3002
\u6211\u4e5f\u5e0c\u671b\uff0c2021\u5e74\u56fe\u673a\u5668\u5b66\u4e60\u4e5f\u80fd\u8fdb\u5165\u66f4\u6709\u5f71\u54cd\u529b\u548c\u66f4\u5177\u6311\u6218\u6027\u7684\u5e94\u7528\u9886\u57df\u3002\u6700\u8fd1\u6709\u592a\u591a\u7684\u7814\u7a76\u96c6\u4e2d\u5728\u7b80\u5355\u3001\u540c\u8d28\u7684\u8282\u70b9\u5206\u7c7b\u4efb\u52a1\u4e0a\u3002\u6211\u4e5f\u5e0c\u671b\u5728\u9700\u8981\u66f4\u590d\u6742\u7684\u7b97\u6cd5\u63a8\u7406\u4efb\u52a1\u4e0a\u770b\u5230\u65b9\u6cd5\u4e0a\u7684\u8fdb\u6b65\uff0c\u6bd4\u5982\u6d89\u53ca\u77e5\u8bc6\u56fe\u8868\u3001\u5f3a\u5316\u5b66\u4e60\u548c\u7ec4\u5408\u4f18\u5316\u7684\u4efb\u52a1\u3002<\/p>\n

2.\u7b97\u6cd5\u63a8\u7406<\/p>\n


 Petar Veli\u010dkovi\u0107\uff0cDeepMind \u9ad8\u7ea7\u7814\u7a76\u5458\uff0cGraph Attention Networks\uff09\u4f5c\u8005<\/p>\n

Petar Veli\u010dkovi\u0107\uff1a
2020\u5df2\u7ecf\u51b3\u5b9a\u6027\u5730\u3001\u4e0d\u53ef\u9006\u8f6c\u5730\u5c06\u56fe\u8868\u5f81\u5b66\u4e60\u53d8\u6210\u4e86\u673a\u5668\u5b66\u4e60\u7684\u201c\u7b2c\u4e00\u516c\u6c11\u201d\u3002
\u4eca\u5e74\u5728\u8be5\u9886\u57df\u53d6\u5f97\u7684\u5de8\u5927\u8fdb\u5c55\u6570\u4e0d\u80dc\u6570\uff0c\u65e0\u6cd5\u4e00\u4e00\u5217\u4e3e\uff0c\u4f46\u6211\u4e2a\u4eba\u6700\u4e3a\u5174\u594b\u7684\u8981\u5c5e\u795e\u7ecf\u7b97\u6cd5\u63a8\u7406\u3002\u4f20\u7edf\u4e0a\uff0c\u795e\u7ecf\u7f51\u7edc\u5728\u63d2\u503c\u673a\u5236\u4e2d\u662f\u975e\u5e38\u5f3a\u5927\u7684\uff0c\u4f46\u4f17\u6240\u5468\u77e5\uff0c\u795e\u7ecf\u7f51\u7edc\u662f\u53ef\u6015\u7684\u5916\u63a8\u5668\uff0c\u63a8\u7406\u80fd\u529b\u4e25\u91cd\u4e0d\u8db3\uff1b\u63a8\u7406\u7684\u4e3b\u8981\u7279\u5f81\u4e4b\u4e00\uff0c\u5c31\u662f\u80fd\u591f\u53d1\u6325OOD\u7684\u529f\u80fd\u3002
\u63a8\u7406\u4efb\u52a1\u5bf9\u4e8e GNN \u7684\u672a\u6765\u53d1\u5c55\u6765\u8bb2\u53ef\u80fd\u662f\u5341\u5206\u7406\u60f3\u7684\uff0c\u4e0d\u4ec5\u56e0\u4e3a\u5b83\u4eec\u4e0e\u8fd9\u7c7b\u4efb\u52a1\u975e\u5e38\u5951\u5408\uff0c\u8fd8\u56e0\u4e3a\u8bb8\u591a\u73b0\u5b9e\u4e16\u754c\u7684\u56fe\u4efb\u52a1\u8868\u73b0\u51fa\u8d8b\u540c\u6027\uff0c\u8fd9\u610f\u5473\u7740\u6700\u6709\u5f71\u54cd\u529b\u548c\u53ef\u6269\u5c55\u7684\u65b9\u6cd5\u901a\u5e38\u662f\u66f4\u7b80\u5355\u5f62\u5f0f\u7684GNN\u3002
\u57fa\u4e8e\u5148\u524d\u795e\u7ecf\u6267\u884c\u5668\u8fc7\u5f80\u7684\u6210\u529f\u6848\u4f8b\uff0c\u5982\u795e\u7ecf\u56fe\u7075\u673a\u548c\u53ef\u5fae\u5206\u795e\u7ecf\u8ba1\u7b97\u673a\uff08DNC\uff09\uff0c\u518d\u52a0\u4e0a\u73b0\u5728\u65e0\u5904\u4e0d\u5728\u7684\u56fe\u673a\u5668\u5b66\u4e60\u5de5\u5177\u7bb1\uff0c2020\u5e74\u53d1\u8868\u7684\u51e0\u9879\u76f8\u5173\u5de5\u4f5c\u63a2\u7d22\u4e86\u795e\u7ecf\u6267\u884c\u5668\u7684\u7406\u8bba\u6781\u9650\uff0c\u57fa\u4e8e GNNs \u884d\u751f\u51fa\u65b0\u9896\u800c\u5f3a\u5927\u7684\u63a8\u7406\u7ed3\u6784\uff0c\u5e76\u5728\u795e\u7ecf\u63a8\u7406\u4efb\u52a1\u4e0a\u5c55\u73b0\u51fa\u5f3a\u5927\u7684\u6cdb\u5316\u80fd\u529b\u3002 \u5c3d\u7ba1\u8fd9\u6837\u7684\u67b6\u6784\u81ea\u7136\u53ef\u4ee5\u57282021\u5e74\u5b9e\u73b0\u7ec4\u5408\u4f18\u5316\uff0c\u4f46\u6211\u4e2a\u4eba\u6700\u611f\u5230\u5174\u594b\u7684\u662f\uff0c\u9884\u8bad\u7ec3\u7684\u7b97\u6cd5\u6267\u884c\u5668\u53ef\u4ee5\u8ba9\u6211\u4eec\u5c06\u7ecf\u5178\u7b97\u6cd5\u5e94\u7528\u4e8e\u8fc7\u4e8e\u539f\u59cb\u6216\u8005\u4e0d\u9002\u5408\u8be5\u7b97\u6cd5\u7684\u8f93\u5165\u3002 \u4f8b\u5982\uff0c\u6211\u4eec\u7684 XLVIN \u4ee3\u7406\u6070\u6070\u4f7f\u7528\u4e86\u8fd9\u4e9b\u6982\u5ff5\uff0c\u4ee5\u5141\u8bb8 GNN \u5728\u5f3a\u5316\u5b66\u4e60 pipeline \u4e2d\u6267\u884c\u503c\u8fed\u4ee3\u6837\u5f0f\u7684\u7b97\u6cd5\uff0c\u5373\u4f7f\u4e0d\u77e5\u9053\u5e95\u5c42 MDP \u7684\u7ec6\u8282\u3002\u6211\u76f8\u4fe12021\u5e74\u7684\u65f6\u673a\u5df2\u7ecf\u6210\u719f\uff0cGNN \u5e94\u7528\u7a0b\u5e8f\u5c06\u666e\u904d\u5e94\u7528\u4e8e\u5f3a\u5316\u5b66\u4e60\u3002<\/p>\n

3.\u5173\u7cfb\u7ed3\u6784\u53d1\u73b0 <\/p>\n


  Thomas Kipf\uff0cGoogle Brain \u79d1\u5b66\u5bb6\uff0c\u56fe\u5377\u79ef\u7f51\u7edc\u7684\u4f5c\u8005
  Thomas Kipf\uff1a \u81ea\u4ece\u57fa\u4e8eGNN\u7684\u6a21\u578b\u88ab\u5e7f\u6cdb\u91c7\u7528\u4ee5\u6765\uff0cGraph ML\u793e\u533a\u4e2d\u4e00\u4e2a\u7279\u522b\u503c\u5f97\u6ce8\u610f\u7684\u8d8b\u52bf\u4fbf\u662f\u5c06\u8ba1\u7b97\u7ed3\u6784\u4e0e\u6570\u636e\u7ed3\u6784\u5206\u79bb\u3002 
\u5728\u6700\u8fd1\u7684\u4e00\u6b21ICML\u7814\u8ba8\u4f1a\u7684\u6f14\u8bb2\u4e2d\uff0c\u6211\u5c06\u8fd9\u79cd\u8d8b\u52bf\u79f0\u4e3a\u201c\u5173\u7cfb\u7ed3\u6784\u53d1\u73b0\u201d\u3002 GNNs \u5141\u8bb8\u5b66\u4e60\u4e00\u4e2a\u72b6\u6001\u8f6c\u6362\u56fe\uff08\u53f3\u4fa7\uff09\uff0c\u6765\u89e3\u91ca\u590d\u6742\u7684\u591a\u7c92\u5b50\u7cfb\u7edf\uff08\u5de6\u4fa7\uff09\u3002 \u901a\u5e38\u6211\u4eec\u8bbe\u8ba1\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u662f\u4e3a\u4e86\u5728\u6570\u636e\u96c6\u63d0\u4f9b\u7684\u4e00\u4e2a\u56fa\u5b9a\uff08\u6216\u968f\u65f6\u95f4\u6f14\u8fdb\uff09\u7ed3\u6784\u4e0a\u53bb\u4f20\u9012\u6d88\u606f\u3002\u4e5f\u5373\uff0c\u5c06\u6570\u636e\u96c6\u7684\u8282\u70b9\u548c\u8fb9\u7f18\u4f5c\u4e3a\u6a21\u578b\u8ba1\u7b97\u7ed3\u6784\u6216\u6d88\u606f\u4f20\u9012\u7ed3\u6784\u7684\u9ec4\u91d1\u6807\u51c6\u3002
\u4f46\u57282020\u5e74\uff0c\u6709\u8bb8\u591a\u7814\u7a76\u5f00\u59cb\u5bf9\u80fd\u591f\u9002\u5e94\u8ba1\u7b97\u7ed3\u6784\u7684\u6a21\u578b\u4ea7\u751f\u5174\u8da3\u3002\u8fd9\u4e9b\u6a21\u578b\u8d85\u8d8a\u4e86\u7b80\u5355\u7684\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u6a21\u578b\uff0c\u53ef\u4ee5\u52a8\u6001\u5730\u9009\u62e9\u4f7f\u7528\u54ea\u4e9b\u7ec4\u4ef6\u4f5c\u4e3a\u8282\u70b9\u4ee5\u53ca\u5728\u54ea\u4e9b\u8282\u70b9\u4e0a\u8fdb\u884c\u6d88\u606f\u4f20\u9012\u3002 \u6bd4\u8f83\u6709\u5f71\u54cd\u7684\u4f8b\u5b50\u5305\u62ec \u644a\u9500\u56e0\u679c\u53d1\u73b0\uff08Amortised Causal Discovery\uff09 [19\u201320]\uff0c\u8fd9\u9879\u7814\u7a76\u5229\u7528\u4e86\u795e\u7ecf\u5173\u7cfb\u63a8\u7406\u4ece\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u4e2d\u63a8\u65ad\u56e0\u679c\u56fe\uff0c\u5e76\u7528\u5b83\u6765\u505a\u63a8\u7406\u3002\u5177\u6709\u79d1\u5b66\u7cfb\u6307\u9488\u7684GNN[21,15]\uff1b\u5173\u8054\u673a\u5236[22,23]\uff1b\u5b66\u4e60\u5177\u6709\u81ea\u9002\u5e94\u8ba1\u7b97\u56fe\u7684\u57fa\u4e8e\u7f51\u683c\u7684\u7269\u7406\u6a21\u62df\u5668[24]\u5b66\u4e60\u63a8\u65ad\u62bd\u8c61\u8282\u70b9\u7684\u6a21\u578b[25,26] \u8fd9\u4e2a\u65b9\u5411\u7684\u63a2\u7d22\u7684\u610f\u4e49\u663e\u800c\u6613\u89c1\uff0c\u56e0\u4e3a\u5b83\u4f7f\u6211\u4eec\u80fd\u591f\u6709\u6548\u5229\u7528\u5176\u5b83\u9886\u57df\uff08\u4f8b\u5982\u6587\u672c\u6216\u89c6\u9891\u5904\u7406\uff09\u4e2d\u7684GNN\u67b6\u6784\u63d0\u4f9b\u7684\u5bf9\u79f0\u6027\uff08\u4f8b\u5982\u8282\u70b9\u7f6e\u6362\u65b9\u5dee\uff09\u548c\u5f52\u7eb3\u504f\u5dee\uff08\u4f8b\u5982\u6210\u5bf9\u4ea4\u6362\u51fd\u6570\u5efa\u6a21\uff09\u3002 \u5c55\u671b\u4e0b\u672a\u6765\uff0c\u6211\u6bd4\u8f83\u671f\u5f85\u7684\u8fdb\u5c55\u662f\uff1a\u5728\u4e0d\u4f9d\u8d56\u660e\u786e\u76d1\u7763\u7684\u60c5\u51b5\u4e0b\uff0c\u7ed9\u5b9a\u4e00\u4e9b\u6570\u636e\u548c\u4efb\u52a1\uff0c\u80fd\u591f\u5b66\u4e60\u51fa\u6700\u4f73\u7684\u8ba1\u7b97\u56fe\u7ed3\u6784\uff08\u4e0d\u7ba1\u662f\u8282\u70b9\u8fd8\u662f\u5173\u7cfb\uff09\u3002 \u4e00\u4e2a\u597d\u5904\u662f\uff0c\u5bf9\u5b66\u4e60\u5230\u7684\u8fd9\u79cd\u7ed3\u6784\u8fdb\u884c\u63a2\u7a76\uff0c\u6211\u4eec\u53ef\u4ee5\u5bf9\u5b66\u5230\u7684\u6a21\u578b\u5982\u4f55\u89e3\u51b3\u67d0\u4e2a\u4efb\u52a1\u505a\u53ef\u89e3\u91ca\u6027\u5206\u6790\uff0c\u4e5f\u53ef\u4ee5\u8ba9\u6211\u4eec\u5bf9\u56e0\u679c\u63a8\u7406\u505a\u8fdb\u4e00\u6b65\u7684\u7c7b\u6bd4\u3002<\/p>\n

4.\u8868\u793a\u80fd\u529b <\/p>\n


 Haggai Maron\uff0c\u82f1\u4f1f\u8fbe\u7684\u79d1\u5b66\u5bb6\uff0c\u53ef\u8868\u793a\u7684\u9ad8\u7ef4\u56fe\u795e\u7ecf\u7f51\u7edc\uff08expressive high-dimensioanl graph neural networks\uff09\u7684\u4f5c\u8005
 Haggai Maron\uff1a
\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8868\u8fbe\u80fd\u529b\u662f2020\u5e74Graph ML\u7684\u6838\u5fc3\u4e3b\u9898\u4e4b\u4e00\u3002 \u5f53\u4e0b\u6709\u8bb8\u591a\u4f18\u79c0\u7684\u8bba\u6587\uff0c\u8ba8\u8bba\u4e86\uff1a
\u5404\u79cdGNN\u67b6\u6784\u7684\u8868\u8fbe\u80fd\u529b\uff1b
\u5f53GNN\u7684\u6df1\u5ea6\u548c\u5bbd\u5ea6\u53d7\u9650\u65f6\u7684\u8868\u8fbe\u6781\u9650\uff1b
\u4f7f\u7528GNN\u54ea\u4e9b\u7c7b\u578b\u7684\u7ed3\u6784\u53ef\u4ee5\u8fdb\u884c\u68c0\u6d4b\u548c\u8ba1\u6570\uff1b
\u4f7f\u7528\u56fa\u5b9a\u6570\u91cf\u7684GNN\u5bf9\u8bb8\u591a\u56fe\u4efb\u52a1\u662f\u65e0\u6548\u7684\uff0c\u56e0\u6b64\u5efa\u8bae\u4f7f\u7528\u8fed\u4ee3GNN\u6765\u5b66\u4e60\u81ea\u9002\u5e94\u7ec8\u6b62\u6d88\u606f\u4f20\u9012\u8fc7\u7a0b\u3002 <\/p>\n

2021\u5e74\uff0c\u6211\u5e0c\u671b\u770b\u5230\u7684\u8fdb\u5c55\u5305\u62ec\uff1a\u56fe\u751f\u6210\u6a21\u578b\u7684\u539f\u7406\u65b9\u6cd5\uff1b
\u4f7f\u7528GNN\u7684\u56fe\u5339\u914d\u548cGNN\u7684\u8868\u8fbe\u80fd\u529b\u4e4b\u95f4\u7684\u8054\u7cfb\uff1b
\u5b66\u4e60\u56fe\u50cf\u3001\u8bed\u97f3\u7b49\u7ed3\u6784\u6570\u636e\u7684\u56fe\uff1b
\u5728scene graphs\u4e0a\uff0cGNN\u793e\u533a\u548cCV\u793e\u533a\u80fd\u591f\u5efa\u7acb\u66f4\u591a\u7684\u5408\u4f5c\u3002
  <\/p>\n

5.\u53ef\u6269\u5c55\u6027<\/p>\n


Matthias Fey\uff0c\u591a\u7279\u8499\u5fb7\u5de5\u4e1a\u5927\u5b66\u535a\u58eb\uff0cPyTorch\u51e0\u4f55\u56fe\u548c\u5f00\u653e\u56fe\u57fa\u51c6\u6d4b\u8bd5\u7684\u5f00\u53d1\u8005
 Matthias Fey\uff1a
2020\u5e74Graph ML\u7814\u7a76\u4e2d\u6700\u70ed\u95e8\u7684\u4e3b\u9898\u4e4b\u4e00\u662f\u89e3\u51b3GNN\u7684\u53ef\u6269\u5c55\u6027\u95ee\u9898\u3002
 
\u5728\u53ef\u6269\u5c55\u6027\u65b9\u9762\uff0c\u6709\u51e0\u79cd\u65b9\u6cd5\u91c7\u7528\u5c06\u9884\u6d4b\u4e0e\u4f20\u64ad\u89e3\u8026\uff0c\u6765\u7b80\u5316\u57fa\u7840\u8ba1\u7b97\u8fc7\u7a0b\u3002
 
\u6211\u4eec\u4e5f\u770b\u5230\u6709\u4e0d\u5c11\u8bba\u6587\u53ea\u662f\u7b80\u5355\u5730\u5c06\u4e0d\u53ef\u8bad\u7ec3\u7684\u4f20\u64ad\u7b56\u7565\u548cgraph-agnostic\uff08\u56fe\u4e0d\u53ef\u77e5\uff09\u6a21\u5757\u7ed3\u5408\u5728\u4e00\u8d77\uff0c\u7528\u4f5c\u9884\u5904\u7406[30,7]\u6216\u540e\u5904\u7406[6]\u7684\u6b65\u9aa4\u3002\u8fd9\u79cd\u65b9\u5f0f\u5f97\u5230\u4e86\u6781\u597d\u7684\u7ed3\u679c\uff0c\u5e76\u4e14\u80fd\u591f\u663e\u8457\u63d0\u9ad8\u540c\u6784\u56fe\u7684\u540c\u7b49\u6027\u80fd\u3002
 
\u968f\u7740\u6570\u636e\u96c6\u8d8a\u6765\u8d8a\u5927\uff0c\u6211\u4eec\u5e0c\u671b\u80fd\u591f\u770b\u5230\u8fd9\u65b9\u9762\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\uff0c\u4ee5\u53ca\u5982\u4f55\u4ee5\u53ef\u6269\u5c55\u7684\u65b9\u5f0f\u6765\u5229\u7528\u53ef\u8bad\u7ec3\u7684\u8868\u793a\u7684\u4f20\u64ad\u3002<\/p>\n

 <\/p>\n

6.\u52a8\u6001\u56fe <\/p>\n


 
 Emanuele Rossi\uff0cTwitter\u673a\u5668\u5b66\u4e60\u7814\u7a76\u5458\uff0c\u5e1d\u56fd\u7406\u5de5\u5b66\u9662\u535a\u58eb\uff0cTemporal Graph Networks\u4f5c\u8005\u3002
 Emanuele Rossi\uff1a
\u8bb8\u591a\u6709\u8da3\u7684Graph ML\u5e94\u7528\u672c\u8d28\u4e0a\u90fd\u662f\u52a8\u6001\u7684\uff0c\u8fd9\u79cd\u52a8\u6001\u5305\u62ec\u56fe\u62d3\u6251\u4ee5\u53ca\u5c5e\u6027\u968f\u65f6\u95f4\u7684\u6f14\u53d8\u3002
\u52a8\u6001\u56fe
\u5728\u793e\u4ea4\u7f51\u7edc\u3001\u91d1\u878d\u4ea4\u6613\u7f51\u7edc\u6216\u7528\u6237\u9879\u76ee\u4ea4\u4e92\u7f51\u7edc\u4e2d\uff0c\u56fe\u7f51\u7edc\u90fd\u662f\u52a8\u6001\u7684\u3002\u4f46\u76f4\u5230\u6700\u8fd1\uff0c\u5bf9Graph ML\u7684\u7edd\u5927\u591a\u6570\u7814\u7a76\u90fd\u96c6\u4e2d\u5728\u9759\u6001\u56fe\u4e0a\uff0c\u800c\u4e00\u4e9b\u5c1d\u8bd5\u5904\u7406\u52a8\u6001\u56fe\u7684\u5de5\u4f5c\u4e5f\u4e3b\u8981\u662f\u79bb\u6563\u65f6\u95f4\u7684\u52a8\u6001\u56fe\uff08\u5373\u56fa\u5b9a\u65f6\u95f4\u95f4\u9694\u7684\u4e00\u7cfb\u5217\u9759\u6001\u56fe\uff09\u3002 2020\u5e74\uff0c\u6211\u4eec\u770b\u5230\u6709\u4e00\u4e9b\u5173\u4e8e\u8fde\u7eed\u65f6\u95f4\u52a8\u6001\u56fe\u7684\u5de5\u4f5c\uff0c\u8fd9\u91cc\u5c06\u52a8\u6001\u56fe\u89c6\u4e3a\u4e8b\u4ef6\u7684\u5f02\u6b65\u6d41\u3002\u4e5f\u6709\u4e00\u4e9b\u5c06\u52a8\u6001\u56fe\u6a21\u578b\u6210\u529f\u5e94\u7528\u7684\u6848\u4f8b\uff0c\u4f8b\u5982\u5047\u8d26\u6237\u68c0\u6d4b\uff0c\u6b3a\u8bc8\u68c0\u6d4b\uff0c\u6d41\u884c\u75c5\u4f20\u64ad\u63a7\u5236\u3002 \u6211\u4e2a\u4eba\u8ba4\u4e3a\uff0c\u8fd9\u4e2a\u9886\u57df\u4ecd\u7136\u5904\u4e8e\u63a2\u7d22\u9636\u6bb5\uff0c\u4ecd\u7136\u6709\u8bb8\u591a\u6709\u8da3\u7684\u95ee\u9898\u5c1a\u672a\u5f97\u5230\u89e3\u7b54\u3002\u6bd4\u8f83\u91cd\u8981\u7684\u95ee\u9898\u5305\u62ec \u52a8\u6001\u56fe\u7684\u53ef\u6269\u5c55\u6027\uff1b
\u5bf9\u52a8\u6001\u6a21\u578b\u7684\u7406\u8bba\u89e3\u91ca\uff1b
\u5355\u4e2a\u6846\u67b6\u4e2d\u5982\u4f55\u66f4\u597d\u5730\u7ed3\u5408\u4fe1\u606f\u7684\u65f6\u7a7a\u6269\u6563\uff1b
\u5982\u4f55\u6784\u5efa\u66f4\u53ef\u9760\u3001\u66f4\u5177\u6311\u6218\u6027\u7684\u57fa\u51c6\uff0c\u4ece\u800c\u786e\u4fdd\u53ef\u4ee5\u66f4\u597d\u5730\u8bc4\u4f30\u548c\u8ddf\u8e2a\u7814\u7a76\u8fdb\u5c55\uff1b
\u5982\u4f55\u5c06\u52a8\u6001\u56fe\u795e\u7ecf\u67b6\u6784\u5e94\u7528\u5230\u66f4\u591a\u7684\u884c\u4e1a\u573a\u666f\u4e2d\u7b49\u3002<\/p>\n

 <\/p>\n

7.\u65b0\u7684\u786c\u4ef6<\/p>\n

  
Mark Saroufim\uff0cGraphcore\u7684\u673a\u5668\u5b66\u4e60\u5de5\u7a0b\u5e08
 
Mark Saroufim\uff1a
\u5728\u4e0e\u6211\u5408\u4f5c\u8fc7\u7684\u5ba2\u6237\u4e2d\uff0c\u51e0\u4e4e\u6240\u6709\u4eba\u90fd\u5728\u751f\u4ea7\u8fc7\u7a0b\u4e2d\u90e8\u7f72\u8fc7\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u6216\u8005\u81f3\u5c11\u662f\u5728\u8ba1\u5212\u8fd9\u6837\u505a\u3002 \u5728NLP\u3001\u86cb\u767d\u8d28\u8bbe\u8ba1\u548c\u5206\u5b50\u7279\u6027\u9884\u6d4b\u7b49\u9886\u57df\u4e2d\uff0c\u4eba\u4eec\u5f80\u5f80\u4f1a\u628a\u8bed\u8a00\u3001\u86cb\u767d\u8d28\u6216\u8005\u5206\u5b50\u76f4\u63a5\u5f53\u505a\u5e8f\u5217\u8fdb\u884c\u5904\u7406\uff0c\u4ee5\u9002\u5e94\u5f53\u4e0b\u6280\u672f\u6bd4\u8f83\u6210\u719f\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\uff0c\u4f8b\u5982Transformers\u3002\u4f46Transformers \u672c\u8d28\u4e0a\u5c31\u662f\u6709\u7740\u591a\u5934\u6ce8\u610f\u529b\u7684 GNN\u3002
 \u534a\u5bfc\u4f53\u516c\u53f8Graphcore\uff0c\u6b63\u5728\u5f00\u53d1\u7528\u4e8e\u56fe\u6570\u636e\u7684\u65b0\u786c\u4ef6\u3002
\u5728\u8ba1\u7b97\u673a\u9886\u57df\uff0c\u67d0\u4e9b\u7b97\u6cd5\u4e4b\u6240\u4ee5\u83b7\u80dc\uff0c\u5e76\u4e0d\u662f\u56e0\u4e3a\u5b83\u4eec\u975e\u5e38\u9002\u5408\u89e3\u51b3\u67d0\u4e2a\u95ee\u9898\uff0c\u800c\u662f\u56e0\u4e3a\u5b83\u4eec\u5728\u73b0\u6709\u786c\u4ef6\u4e0a\u8fd0\u884c\u826f\u597d\uff0c\u8fd9\u79cd\u73b0\u8c61\u79f0\u4e4b\u4e3a\u201c\u786c\u4ef6\u5f69\u7968\u201d(Hardware Lottery)\u2014\u2014 Transformers \u80fd\u8fd0\u884c\u5728 GPU \u4e0a\u5c31\u662f\u8fd9\u79cd\u60c5\u51b5\u3002 \u5728 Graphcore\u4e0a\uff0c\u6211\u4eec\u6784\u5efa\u4e86\u4e00\u4e2a\u67091472\u4e2a\u6838\u7684 MIMD \u67b6\u6784\uff0c\u53ef\u4ee5\u5e76\u884c\u8fd0\u884c8832\u4e2a\u7a0b\u5e8f\uff0c\u6211\u4eec\u79f0\u4e4b\u4e3a\u667a\u80fd\u5904\u7406\u5355\u5143(IPU)\u3002\u8fd9\u79cd\u67b6\u6784\u975e\u5e38\u9002\u5408 GNN \u52a0\u901f\u3002Graphcore \u7684\u8f6f\u4ef6\u6808\u5229\u7528\u7a00\u758f\u6027\u5c06\u8ba1\u7b97\u56fe\u7684\u4e0d\u540c\u8282\u70b9\u5206\u914d\u7ed9\u4e0d\u540c\u7684\u6838\u3002 \u5bf9\u4e8e\u53ef\u4ee5\u7eb3\u5165 IPU 900MB \u82af\u7247\u5b58\u50a8\u7684\u6a21\u578b\uff0c\u6211\u4eec\u7684\u67b6\u6784\u901a\u8fc7 GPU \u5b9e\u73b0\u4e86\u541e\u5410\u91cf\u7684\u5b9e\u8d28\u6027\u6539\u8fdb\uff1b\u6b64\u5916\uff0c\u53ea\u9700\u51e0\u884c\u4ee3\u7801\uff0c\u5c31\u53ef\u4ee5\u5c06\u6a21\u578b\u5206\u53d1\u5230\u6570\u5343\u4e2a GPU \u4e0a\u3002 \u6211\u5f88\u9ad8\u5174\u770b\u5230\u6211\u4eec\u7684\u7528\u6237\u5229\u7528\u6211\u4eec\u7684\u67b6\u6784\u505a\u4e86\u5927\u91cf\u7684\u7814\u7a76\uff0c\u5305\u62ec\u5404\u7c7b\u5e94\u7528\uff0c\u4f8b\u5982 SLAM \u4e2d\u7684\u5149\u675f\u6cd5\u5e73\u5dee\uff0c\u4f7f\u7528\u672c\u5730\u66f4\u65b0\u8bad\u7ec3\u6df1\u5c42\u7f51\u7edc\uff0c\u6216\u8005\u89e3\u51b3\u7c92\u5b50\u7269\u7406\u5b66\u4e2d\u7684\u5404\u79cd\u95ee\u9898\u3002\u6211\u5e0c\u671b\u57282021\u5e74\u770b\u5230\u66f4\u591a\u7814\u7a76\u4eba\u5458\u5229\u7528\u6211\u4eec\u5148\u8fdb\u7684\u673a\u5668\u5b66\u4e60\u786c\u4ef6\u3002
 <\/p>\n

8.\u5de5\u4e1a\u5e94\u7528<\/p>\n

 Sergey Ivanov\uff0cCriteo\u79d1\u5b66\u5bb6 Sergey Ivanov\uff1a \u5bf9\u4e8eGraph ML\u7814\u7a76\u6765\u8bf4\uff0c\u8fd9\u662f\u4ee4\u4eba\u9707\u60ca\u7684\u4e00\u5e74\u3002\u5728\u6240\u6709\u4e3b\u8981\u7684ML\u4f1a\u8bae\u4e0a\uff0c\u6709\u5173\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u8bba\u6587\u4e2d\u7ea6\u5360\u5168\u90e8\u7684 10\uff05\u81f320\uff05\u3002\u5728\u8fd9\u79cd\u89c4\u6a21\u4e0b\uff0c\u6bcf\u4e2a\u4eba\u90fd\u53ef\u4ee5\u627e\u5230\u81ea\u5df1\u611f\u5174\u8da3\u4e14\u6709\u8da3\u7684\u7814\u7a76\u8fdb\u5c55\u3002 \u5728NeurIPS 2020\u4e0a\uff0c\u8c37\u6b4c\u7684\u56fe\u6316\u6398\u56e2\u961f\u5168\u65b9\u4f4d\u4ecb\u7ecd\u4e86\u4ed6\u4eec\u5728\u4ea7\u54c1\u4e2d\u5982\u4f55\u4f7f\u7528Graph ML\uff0c\u5728312\u9875\u7684slide\u91cc\uff0c\u4ed6\u4eec\u4ecb\u7ecd\u4e86\u4ed6\u4eec\u5229\u7528\u65f6\u7a7aGNN\u5efa\u6a21COVID-19\u3001\u6b3a\u8bc8\u68c0\u6d4b\u3001\u9690\u79c1\u4fdd\u62a4\u7b49\u65b9\u9762\u7684\u5e94\u7528\u5de5\u4f5c\u3002
\u53e6\u5916\uff0cDeepMind\u5229\u7528GNN\u5728\u8c37\u6b4c\u5730\u56fe\u4e2d\u505a\u4e86\u5168\u7403\u65c5\u884c\u65f6\u95f4\u9884\u6d4b\uff0c\u8fd9\u4e2a\u5de5\u4f5c\u4e00\u4e2a\u6709\u8da3\u7684\u7ec6\u8282\u662f\uff0c\u5c06\u5f3a\u5316\u5b66\u4e60\u6a21\u578b\u96c6\u6210\u5230\u4e00\u4e2abatch\u4e2d\uff0c\u9009\u62e9\u76f8\u4f3c\u7684\u91c7\u6837\u5b50\u56fe\u6765\u8bad\u7ec3GNN\u7684\u53c2\u6570\uff0c\u8fd9\u79cd\u8d85\u53c2\u8c03\u6574\u80fd\u4f7f\u5b9e\u65f6\u5230\u8fbe\u65f6\u95f4\u4f30\u7b97\u7684\u7cbe\u5ea6\u63d0\u9ad850%\u4ee5\u4e0a\u3002
SuperGlue\uff0c\u4f7f\u7528GNN\u89e3\u51b3\u4e86\u8ba1\u7b97\u673a\u89c6\u89c9\u4e2d\u7684\u7279\u5f81\u5339\u914d\u95ee\u9898\u3002 
\u53e6\u4e00\u4e2a\u6bd4\u8f83\u6709\u8da3\u7684\u5e94\u7528\u6765\u81ea Magic Leap\uff0c\u8fd9\u5bb6\u516c\u53f8\u4e13\u95e8\u7814\u7a763D\u56fe\u5f62\u751f\u6210\u3002\u4ed6\u4eec\u5c06SuperGlue\u67b6\u6784[39]\u5e94\u7528\u5230\u4e86\u56fe\u50cf\u7279\u5f81\u5339\u914d\uff08\u8fd9\u662f3D\u91cd\u5efa\u3001\u4f4d\u7f6e\u8bc6\u522b\u3001\u5b9a\u4f4d\u548c\u6620\u5c04\u7684\u91cd\u8981\u8bfe\u9898\uff09\u4e0a\u3002\u8fd9\u79cd\u7aef\u5230\u7aef\u7684\u7279\u5f81\u8868\u793a\uff0c\u4e0e\u6700\u4f73\u8fd0\u8f93\u4f18\u5316\u76f8\u7ed3\u5408\uff0c\u5728\u5b9e\u65f6\u7684\u5ba4\u5185\u548c\u5ba4\u5916\u59ff\u6001\u4f30\u8ba1\u4e0a\u8868\u73b0\u6781\u4f73\u3002
\u8fd9\u4e9b\u5f53\u7136\u53ea\u662fGNN\u57282020\u5e74\u5de5\u4e1a\u5e94\u7528\u7684\u4e00\u4e2a\u7f29\u5f71\u3002\u57282021\u5e74\uff0c\u6211\u4eec\u4f1a\u5728\u5de5\u4e1a\u4e2d\u770b\u5230\u66f4\u591aGraph ML\u7684\u8eab\u5f71\uff0c\u4f8b\u5982\u751f\u4ea7\u6d41\u7a0b\u548c\u6846\u67b6\u3001\u65b0\u7684\u5f00\u6e90\u56fe\u6570\u636e\u96c6\u3001\u7535\u5b50\u5546\u52a1\u3001\u5de5\u7a0b\u8bbe\u8ba1\u3001\u5236\u836f\u7b49\uff0c\u90fd\u4f1a\u6709GNN\u7684\u5927\u89c4\u6a21\u90e8\u7f72\u3002
 <\/p>\n

9.\u5728\u7269\u7406\u5b66\u4e2d\u7684\u5e94\u7528<\/p>\n

 Kyle Cranmer\uff0c\u7ebd\u7ea6\u5927\u5b66\u7269\u7406\u5b66\u6559\u6388\uff0c\u5e0c\u683c\u65af\u73bb\u8272\u5b50\u7684\u53d1\u73b0\u8005\u4e4b\u4e00<\/p>\n

Kyle Cranmer\uff1a
\u770b\u5230\u56fe\u673a\u5668\u5b66\u4e60\u5728\u8fc7\u53bb\u4e24\u5e74\u4e2d\u5728\u7269\u7406\u9886\u57df\u53d8\u5f97\u975e\u5e38\u6d41\u884c\uff0c\u771f\u662f\u592a\u795e\u5947\u4e86\u3002 \u65e9\u671f\u5728\u7c92\u5b50\u7269\u7406\u4e2d\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60\uff0c\u901a\u5e38\u7684\u505a\u6cd5\u662f\uff0c\u628a\u5404\u79cd\u6570\u636e\u8f6c\u6362\u6210\u56fe\u50cf\u7684\u8868\u793a\u5f62\u5f0f\uff0c\u7136\u540e\u8f93\u5165\u5230CNN\u4e2d\uff0c\u8fd9\u6574\u4e2a\u8fc7\u7a0b\u662f\u6781\u4e3a\u4e0d\u81ea\u7136\u7684\uff0c\u56e0\u4e3a\u6211\u4eec\u7684\u6570\u636e\u672c\u8eab\u5e76\u975e\u7f51\u683c\u72b6\u7684\uff0c\u8868\u793a\u6210\u56fe\u50cf\u4f1a\u975e\u5e38\u7a00\u758f\u3002\u56fe\uff0c\u5bf9\u6211\u4eec\u7684\u6570\u636e\u6765\u8bf4\uff0c\u5219\u662f\u4e00\u4e2a\u975e\u5e38\u81ea\u7136\u7684\u8868\u793a\u5f62\u5f0f\u3002
\u7c92\u5b50\u55b7\u5c04\u53ef\u4ee5\u8868\u793a\u4e3a\u4e00\u4e2a\u56fe\u3002\u7814\u7a76\u4eba\u5458\u6b63\u5728\u63a2\u7d22\u4f7f\u7528GNN\u6765\u68c0\u6d4b\u7c92\u5b50\u7269\u7406\u5b66\u4e2d\u7684\u53d1\u73b0\u3002\u56fe\u7247\u6765\u6e90\uff1aLHC
 \u5927\u578b\u5f3a\u5b50\u5bf9\u649e\u673a\u7684\u7814\u7a76\u4eba\u5458\u73b0\u5728\u6b63\u5728\u52aa\u529b\u5c06\u56fe\u673a\u5668\u5b66\u4e60\u96c6\u6210\u5230\u6bcf\u79d2\u5904\u7406\u6570\u5341\u4ebf\u6b21\u78b0\u649e\u7684\u5b9e\u65f6\u6570\u636e\u5904\u7406\u7cfb\u7edf\u5f53\u4e2d\u3002\u7814\u7a76\u4eba\u5458\u6b63\u5728\u52aa\u529b\u901a\u8fc7\u90e8\u7f72\u63a8\u7406\u670d\u52a1\u5668\u5c06\u56fe\u673a\u5668\u5b66\u4e60\u4e0e\u5b9e\u65f6\u6570\u636e\u91c7\u96c6\u7cfb\u7edf\u96c6\u6210\u5728\u4e00\u8d77\uff0c\u5e76\u8bd5\u56fe\u5728FPGA\u548c\u5176\u4ed6\u7279\u6b8a\u786c\u4ef6\u4e0a\u5b9e\u73b0\u8fd9\u4e9b\u7b97\u6cd5\u3002 Graph ML \u57282020\u5e74\u7684\u53e6\u4e00\u4e2a\u4eae\u70b9\u662f\u8bc1\u660e\u5176\u5f52\u7eb3\u504f\u7f6e\u53ef\u4ee5\u4e0e\u7b26\u53f7\u65b9\u6cd5\u914d\u5bf9\u3002 \u4f8b\u5982\uff0c\u6211\u4eec\u4f7f\u7528GNN\u6765\u5b66\u4e60\u9884\u6d4b\u5404\u79cd\u52a8\u529b\u5b66\u7cfb\u7edf\uff0c\u7136\u540e\u5bf9\u6cbf\u7740\u8fb9\u53d1\u9001\u7684\u6d88\u606f\u8fdb\u884c\u7b26\u53f7\u56de\u5f52\u3002\u6211\u4eec\u4e0d\u4ec5\u80fd\u591f\u590d\u73b0\u90a3\u4e9b\u5df2\u77e5\u52a8\u529b\u7cfb\u7edf\u7684\u529b\u5b66\u89c4\u5f8b\uff0c\u800c\u4e14\u80fd\u591f\u83b7\u77e5\u90a3\u4e9b\u5b8c\u5168\u4e0d\u77e5\u9053\u89c4\u5f8b\u7684\u7cfb\u7edf\u7684\u529b\u5b66\u65b9\u7a0b\u3002 \u6b64\u5916\uff0c\u83b7\u5f97\u7684\u8fd9\u4e9b\u65b9\u7a0b\u8fd8\u53ef\u4ee5\u653e\u5230GNN\u4e2d\uff0c\u66f4\u65b0\u539f\u59cb\u7684\u5b66\u4e60\u7ec4\u5efa\uff0c\u4ece\u800c\u6765\u66f4\u597d\u5730\u5f52\u7eb3\u5206\u5e03\u6570\u636e\u3002
 <\/p>\n

10.\u533b\u5b66\u4e2d\u7684\u5e94\u7528 <\/p>\n


 
Anees Kazi\uff0c\u6155\u5c3c\u9ed1\u5de5\u4e1a\u5927\u5b66\u7684\u535a\u58eb\u751f \uff0c\u66fe\u64b0\u5199\u4e86\u591a\u7bc7\u6709\u5173 Graph ML \u5728\u533b\u5b66\u6210\u50cf\u4e2d\u5e94\u7528\u7684\u8bba\u6587\u3002<\/p>\n

 Anees Kazi\uff1a
\u5728\u533b\u5b66\u9886\u57df\uff0cGraph ML\u6539\u53d8\u4e86\u5206\u6790\u591a\u6a21\u6001\u6570\u636e\u7684\u65b9\u5f0f\uff0c\u8fd9\u79cd\u65b9\u5f0f\u4e0e\u4e13\u5bb6\u5982\u4f55\u901a\u8fc7\u4e34\u5e8a\u5e38\u89c4\u4e2d\u7684\u6240\u6709\u53ef\u7528\u7ef4\u5ea6\u89c2\u5bdf\u60a3\u8005\u7684\u72b6\u51b5\u975e\u5e38\u76f8\u4f3c\u3002 \u6700\u8fd1\uff0c\u5728\u533b\u5b66\u6210\u50cf\u548c\u533b\u7597\u4fdd\u5065\u5e94\u7528\u4e2d\uff0c\u4e0eGraph ML\u76f8\u5173\u7684\u7814\u7a76\u51fa\u73b0\u4e86\u5927\u7206\u53d1\uff0c\u5305\u62ec\u8111\u533a\u56fe\u50cf\u5206\u5272\uff0c\u4ee5\u53ca\u4f7f\u7528\u9488\u5bf9\u75be\u75c5\u9884\u6d4b\u7684MRI \/ fMRI\u6570\u636e\u8fdb\u884c\u8111\u7ed3\u6784\u5206\u6790\u548c\u836f\u7269\u4f5c\u7528\u5206\u6790\u3002
GNN \u53ef\u4ee5\u5229\u7528\u4eba\u53e3\u56fe\u8fdb\u884c\u75be\u75c5\u5206\u7c7b
 \u8bf4\u8d772020\u5e74\uff0cGraph ML\u5728\u533b\u5b66\u9886\u57df\u4e2d\u7684\u5e94\u7528\uff0c\u6709\u4ee5\u4e0b\u65b9\u9762\u8131\u9896\u800c\u51fa\u3002 \u9996\u5148\uff0c\u6f5c\u56fe\u5b66\u4e60\uff0c\u901a\u8fc7\u7ed9\u5b9a\u6570\u636e\u6765\u7ecf\u9a8c\u6027\u5b9a\u4e49\u4e00\u4e2a\u56fe\u7684\u6f5c\u56fe\u5b66\u4e60\u4e4b\u524d\u662f\u4f18\u5316\u7ed3\u679c\u7684\u74f6\u9888\uff0c\u4f46\u73b0\u5728\u5df2\u7ecf\u88ab\u81ea\u52a8\u5b66\u4e60\u6f5c\u56fe\u7ed3\u6784\u7684\u65b9\u6cd5\u89e3\u51b3\u4e86\u3002\u5176\u6b21\uff0c\u6570\u636e\u91cd\u5efa\uff0c\u56e0\u4e3a\u6570\u636e\u4e22\u5931\u662f\u533b\u5b66\u9886\u57df\u8bb8\u591a\u6570\u636e\u96c6\u7684\u4e00\u4e2a\u5e38\u89c1\u95ee\u9898\uff0c\u56e0\u6b64\u57fa\u4e8e\u56fe\u7684\u65b9\u6cd5\u53ef\u4ee5\u6839\u636e\u90bb\u8fd1\u56fe\u7684\u5173\u7cfb\u5b8c\u6210\u6570\u636e\u91cd\u5efa\u3002\u7b2c\u4e09\uff0cGraph ML\u6a21\u578b\u7684\u53ef\u89e3\u91ca\u6027\uff0c\u5bf9\u4e8e\u4e34\u5e8a\u548c\u6280\u672f\u4e13\u5bb6\u6765\u8bf4\uff0c\u91cd\u70b9\u5728\u4e8e\u63a8\u7406\u51faGraph ML\u6a21\u578b\u7684\u7ed3\u679c\uff0c\u5e76\u5c06\u5176\u53ef\u9760\u5730\u5408\u5e76\u5230CAD\u7cfb\u7edf\uff08\u8ba1\u7b97\u673a\u8f85\u52a9\u68c0\u6d4b\uff09\u4e2d\u30022020\u5e74\u533b\u5b66\u9886\u57df\u7684\u53e6\u4e00\u4e2a\u91cd\u8981\u4eae\u70b9\u5f53\u7136\u662f\u65b0\u51a0\uff0c\u5e76\u4e14\u4f7f\u7528Graph ML\u65b9\u6cd5\u68c0\u6d4bCovid-19\u3002 \u52302021\u5e74\uff0cGraph ML\u53ef\u4ee5\u7528\u4e8e\u8fdb\u4e00\u6b65\u63d0\u5347ML\u6a21\u578b\u7684\u53ef\u89e3\u91ca\u6027\uff0c\u4ece\u800c\u505a\u51fa\u66f4\u597d\u7684\u51b3\u7b56\u3002\u5176\u6b21\uff0c\u7814\u7a76\u4eba\u5458\u89c2\u5bdf\u5230 Graph ML\u65b9\u6cd5\u4ecd\u7136\u5bf9\u56fe\u7ed3\u6784\u654f\u611f\uff0c\u56e0\u6b64\uff0c\u9488\u5bf9\u56fe\u6270\u52a8\u548c\u5bf9\u6297\u653b\u51fb\u7684\u9c81\u68d2\u6027\u662f\u91cd\u8981\u7684\u8bae\u9898\u3002\u6700\u540e\uff0c\u5c06\u81ea\u76d1\u7763\u5b66\u4e60\u4e0eGraph ML\u7684\u96c6\u6210\u5e94\u7528\u4e8e\u533b\u7597\u9886\u57df\u5c06\u4f1a\u662f\u5f88\u6709\u8da3\u7684\u3002
 <\/p>\n

11.\u86cb\u767d\u8d28\u7ed3\u6784\u9884\u6d4b <\/p>\n

Bruno Correia\uff0c\u745e\u58eb\u6d1b\u6851\u8054\u90a6\u7406\u5de5\u5b66\u9662\u52a9\u7406\u6559\u6388\u3001\u86cb\u767d\u8d28\u8bbe\u8ba1\u548c\u514d\u75ab\u5de5\u7a0b\u5b9e\u9a8c\u5ba4\u8d1f\u8d23\u4eba\uff0cMaSIF\u7684\u5f00\u53d1\u5546\u4e4b\u4e00\u3002<\/p>\n

 2020\u5e74\uff0c\u86cb\u767d\u8d28\u7ed3\u6784\u9884\u6d4b\u53d6\u5f97\u4e86\u4ee4\u4eba\u5174\u594b\u7684\u8fdb\u5c55\uff0c\u8fd9\u662f\u751f\u7269\u4fe1\u606f\u5b66\u4e2d\u7684\u5173\u952e\u95ee\u9898\u3002 \u5bf9\u4e8e\u86cb\u767d\u8d28\u6765\u8bf4\uff0c\u5176\u5206\u5b50\u8868\u9762\u7684\u5316\u5b66\u548c\u51e0\u4f55\u6a21\u5f0f\u76f4\u63a5\u5f71\u54cd\u5176\u529f\u80fd\u3002  \u4f7f\u7528\u51e0\u4f55\u673a\u5668\u5b66\u4e60\u67b6\u6784MaSIF\u8bbe\u8ba1\u7684\u4e0d\u540c\u86cb\u767d\u7ed3\u5408\u5242\uff0c\u7528\u4e8e\u80bf\u7624\u9776\u5411\u6cbb\u7597\u3002<\/p>\n

\u57fa\u4e8e\u5206\u5b50\u8868\u9762\u8868\u793a\u7684\u7814\u7a76\uff0c\u5df2\u7ecf\u6709\u6570\u5341\u5e74\u4e86\uff0c\u4e5f\u540c\u65f6\u7ed9\u673a\u5668\u5b66\u4e60\u5e26\u6765\u4e86\u6781\u5927\u7684\u6311\u6218\u3002 \u5176\u4e2d\u8ba9\u4eba\u6bd4\u8f83\u5370\u8c61\u6df1\u523b\u7684\u7814\u7a76\u662f\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\u7684\u65b9\u6cd5\uff0c\u8fd9\u79cd\u65b9\u6cd5\u80fd\u591f\u5904\u7406\u4e0d\u89c4\u5219\u6570\u636e\uff0c\u7279\u522b\u9002\u5408\u86cb\u767d\u8d28\u7684\u8868\u793a\u3002\u5728MASIF\u4e2d\uff0c\u5728\u57fa\u4e8e\u7f51\u683c\u7684\u5206\u5b50\u8868\u9762\u8868\u793a\u4e0a\uff0c\u4f7f\u7528\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\u8fdb\u884c\u8bad\u7ec3\uff0c\u53ef\u4ee5\u5b66\u4e60\u5230\u53ef\u9884\u6d4b\u86cb\u767d\u8d28\u548c\u5176\u4ed6\u5206\u5b50\u76f8\u4e92\u4f5c\u7528\u7684\u6a21\u5f0f\uff0c\u800c\u4e14\u53ef\u4ee5\u5c06\u8ba1\u7b97\u901f\u5ea6\u63d0\u5347\u51e0\u4e2a\u6570\u91cf\u7ea7\u3002\u53cd\u8fc7\u6765\uff0c\u8fd9\u4e5f\u53ef\u4ee5\u4fc3\u8fdb\u66f4\u5927\u8303\u56f4\u7684\u86cb\u767d\u8d28-\u86cb\u767d\u8d28\u76f8\u4e92\u4f5c\u7528\u7f51\u7edc\u7684\u9884\u6d4b\u3002 \u5728MASIF\u6846\u67b6\u7684\u57fa\u7840\u4e0a\uff0c\u6211\u4eec\u505a\u5230\u4e86\u52a8\u6001\u751f\u6210\u5316\u5b66\u548c\u8868\u9762\u6a21\u5f0f\uff0c\u4ece\u800c\u907f\u514d\u4e86\u6240\u6709\u9884\u8ba1\u7b97\u7684\u9636\u6bb5\u3002\u6211\u4eec\u9884\u8ba1\uff0c\u8fd9\u4e9b\u8fdb\u5c55\u5c06\u4e3a\u86cb\u767d\u8d28\u548c\u5c0f\u5206\u5b50\u8bbe\u8ba1\u5e26\u6765\u53d8\u9769\uff0c\u957f\u8fdc\u6765\u770b\uff0c\u5c06\u6709\u52a9\u4e8e\u66f4\u5feb\u5730\u5f00\u53d1\u751f\u7269\u836f\u7269\u3002
 <\/p>\n

12.\u751f\u547d\u79d1\u5b66 <\/p>\n

Marinka Zitnik\uff0c\u54c8\u4f5b\u533b\u5b66\u9662\u751f\u7269\u533b\u5b66\u4fe1\u606f\u5b66\u52a9\u7406\u6559\u6388\uff0cDecagon\u7cfb\u7edf\u7684\u4f5c\u8005\u3002<\/p>\n

Marinka Zitnik\uff1a \u5f88\u9ad8\u5174\u770b\u5230Graph ML\u57282020\u5e74\u8fdb\u5165\u5230\u751f\u547d\u79d1\u5b66\u9886\u57df\u3002 \u6211\u4eec\u5df2\u7ecf\u770b\u5230\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\u5728\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u57fa\u51c6\u6570\u636e\u96c6\u4e0a\u53ef\u4ee5\u80dc\u8fc7\u4f20\u7edf\u65b9\u6cd5\uff0c\u53e6\u4e00\u65b9\u9762\uff0c\u5728\u65b0\u836f\u5f00\u53d1\u4e0a\uff0c\u4e5f\u4ece\u6839\u672c\u4e0a\u5f00\u8f9f\u4e86\u4e00\u6761\u65b0\u7684\u9014\u5f84\u3002\u6bd4\u8f83\u663e\u8457\u7684\u8fdb\u5c55\u5305\u62ec\u5355\u7ec6\u80de\u751f\u7269\u5b66\u3001\u86cb\u767d\u8d28\u548c\u7ed3\u6784\u751f\u7269\u5b66\uff0c\u836f\u7269\u53d1\u73b0\u548c\u91cd\u5b9a\u4f4d\u7b49\u3002 \u5728Decagon\u4e2d\u4f7f\u7528GNN\uff0c\u6765\u9884\u6d4b\u836f\u7269\u7684\u526f\u4f5c\u7528\u3002
  \u51e0\u4e2a\u4e16\u7eaa\u4ee5\u6765\uff0c\u79d1\u5b66\u5bb6\u6240\u4f7f\u7528\u7684\u79d1\u5b66\u65b9\u6cd5\u4e00\u76f4\u4fdd\u6301\u4e0d\u53d8\u3002\u6211\u4eec\u5e0c\u671b\uff0c\u57282021\u5e74\uff0c\u6211\u4eec\u80fd\u591f\u5229\u7528Graph ML \u5bf9\u8fd9\u4e00\u73b0\u8c61\u505a\u51fa\u4e00\u4e9b\u5b9e\u8d28\u6027\u7684\u8fdb\u5c55\u3002 \u56e0\u6b64\uff0c\u6211\u8ba4\u4e3a\uff0c\u6211\u4eec\u9700\u8981\u8bbe\u8ba1\u51fa\u4e00\u79cd\u65b9\u6cd5\uff0c\u80fd\u591f\u4f18\u5316\u548c\u64cd\u7eb5\u7f51\u7edc\u7cfb\u7edf\uff0c\u7136\u540e\u9884\u6d4b\u5176\u7ed3\u679c\uff1b\u57fa\u4e8e\u8fd9\u79cd\u65b9\u6cd5\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u5fae\u6270\u548c\u5e72\u9884\u6570\u636e\u6765\u63a7\u5236\u5b9e\u9a8c\uff0c\u800c\u6570\u636e\u672c\u8eab\u4e0d\u9700\u8981\u6211\u4eec\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u8fdb\u884c\u89c2\u6d4b\uff09 \u53e6\u5916\uff0c\u6211\u5e0c\u671b\u6211\u4eec\u53ef\u4ee5\u5f00\u53d1\u51fa\u66f4\u591a\u7684\u65b9\u6cd5\u6765\u5b66\u4e60\u53ef\u64cd\u4f5c\u7684\u8868\u793a\u5f62\u5f0f\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u53ef\u4ee5\u66f4\u52a0\u5bb9\u6613\u5730\u5e94\u7528\u5230\u79d1\u5b66\u4e2d\u7684\u53ef\u64cd\u4f5c\u7684\u5047\u8bbe\u4e2d\uff0c\u4ece\u800c\u907f\u514d\u5728\u9ad8\u98ce\u9669\u73af\u5883\uff08\u4f8b\u5982\u5316\u5b66\u6d4b\u8bd5\u3001\u7c92\u5b50\u7269\u7406\u3001\u4e34\u5e8a\u8bd5\u9a8c\u7b49\uff09\u4e2d\u8fdb\u884c\u51b3\u7b56\uff0c\u8fd8\u80fd\u83b7\u5f97\u66f4\u6709\u610f\u4e49\u7684\u7cbe\u786e\u89e3\u91ca\u4ee5\u53ca\u53ef\u9760\u7684\u9884\u6d4b\u3002<\/p>\n

\u539f\u6587\u5730\u5740\uff1a
https:\/\/towardsdatascience.com\/predictions-and-hopes-for-graph-ml-in-2021-6af2121c3e3d<\/p>\n

\u53c2\u8003\u8d44\u6599
[1] U. Alon and E. Yahav, On the bottleneck of graph neural networks and its practical implications (2020) arXiv:2006.05205.
[2] Q. Li, Z. Han, X.-M. Wu, Deeper insights into graph convolutional networks for semi-supervised learning (2019) Proc. AAAI.
[3] K. Xu et al. How powerful are graph neural networks? (2019) Proc. ICLR.
[4] C. Morris et al. Weisfeiler and Leman go neural: Higher-order graph neural networks (2019) Proc. AAAI.
[5] K. Xu et al. What can neural networks reason about? (2019) arXiv:1905.13211.
[6] Q. Huang et al. Combining label propagation and simple models out-performs graph neural networks (2020) arXiv:2010.13993.
[7] F. Frasca et al. SIGN: Scalable Inception Graph Neural Networks (2020) arXiv:2004.11198.
[8] A. Graves, G. Wayne, and I. Danihelka, Neural Turing Machines (2014) arXiv:1410.5401.
[9] A. Graves et al. Hybrid computing using a neural network with dynamic external memory (2016). Nature 538:471\u2013476.
[10] G. Yehuda, M. Gabel, and A. Schuster. It\u2019s not what machines can learn, it\u2019s what we cannot teach (2020) arXiv:2002.09398.
[11] K. Xu et al. How neural networks extrapolate: From feedforward to graph neural networks (2020) arXiv:2009.11848.
[12] P. Veli\u010dkovi\u0107 et al., Neural execution of graph algorithms (2019) arXiv:1910.10593.
[13] O. Richter and R. Wattenhofer, Normalized attention without probability cage (2020) arXiv:2005.09561.
[14] H. Tang et al., Towards scale-invariant graph-related problem solving by iterative homogeneous graph neural networks (2020) arXiv:2010.13547.
[15] P. Veli\u010dkovi\u0107 et al. Pointer Graph Networks (2020) Proc. NeurIPS.
[16] Y. Yan et al. Neural execution engines: Learning to execute subroutines (2020) Proc. ICLR.
[17] C. K. Joshi et al. Learning TSP requires rethinking generalization (2020) arXiv:2006.07054.
[18] A. Deac et al. XLVIN: eXecuted Latent Value Iteration Nets (2020) arXiv:2010.13146.
[19] S. L\u00f6we et al., Amortized Causal Discovery: Learning to infer causal graphs from time-series data (2020) arXiv:2006.10833.
[20] Y. Li et al., Causal discovery in physical systems from videos (2020) Proc. NeurIPS.
[21] D. Bieber et al., Learning to execute programs with instruction pointer attention graph neural networks (2020) Proc. NeurIPS.
[22] A. Kazi et al., Differentiable Graph Module (DGM) for graph convolutional networks (2020) arXiv:2002.04999
[23] D. D. Johnson, H. Larochelle, and D. Tarlow., Learning graph structure with a finite-state automaton layer (2020). arXiv:2007.04929.
[24] T. Pfaff et al., Learning mesh-based simulation with graph networks (2020) arXiv:2010.03409.
[25] T. Kipf et al., Contrastive learning of structured world models (2020) Proc. ICLR
[26] F. Locatello et al., Object-centric learning with slot attention (2020) Proc. NeurIPS.
[27] W. Azizian and M. Lelarge, Characterizing the expressive power of invariant and equivariant graph neural networks (2020) arXiv:2006.15646.
[28] A. Loukas, What graph neural networks cannot learn: depth vs width (2020) Proc. ICLR.
[29] Z. Chen et al., Can graph neural networks count substructures? (2020) Proc. NeurIPS.
[30] A. Bojchevski et al., Scaling graph neural networks with approximate PageRank (2020) Proc. KDD.
[31] E. Rossi et al., Temporal Graph Networks for deep learning on dynamic graphs (2020) arXiv:2006.10637.
[32] S. Kumar, X. Zhang, and J. Leskovec, Predicting dynamic embedding trajectory in temporal interaction networks (2019) Proc. KDD.
[33] R. Trivedi et al., DyRep: Learning representations over dynamic graphs (2019) Proc. ICLR.
[34] D. Xu et al., Inductive representation learning on temporal graphs (2019) Proc. ICLR.
[35] M. Noorshams, S. Verma, and A. Hofleitner, TIES: Temporal Interaction Embeddings for enhancing social media integrity at Facebook (2020) arXiv:2002.07917.
[36] X. Wang et al., APAN: Asynchronous Propagation Attention Network for real-time temporal graph embedding (2020) arXiv:2011.11545.
[37] E. A. Meirom et al., How to stop epidemics: Controlling graph dynamics with reinforcement learning and graph neural networks (2020) arXiv:2010.05313.
[38] S. Hooker, Hardware lottery (2020), arXiv:2009.06489.
[39] P. E. Sarlin et al., SuperGlue: Learning feature matching with graph neural networks (2020). Proc. CVPR.
[40] S. Ruhk et al., Learning representations of irregular particle-detector geometry with distance-weighted graph networks (2019) arXiv:1902.07987.
[41] J. Shlomi, P. Battaglia, J.-R. Vlimant, Graph Neural Networks in particle physics (2020) arXiv:2007.13681.
[42] J. Krupa et al., GPU coprocessors as a service for deep learning inference in high energy physics (2020) arXiv:2007.10359.
[43] A. Heintz et al., Accelerated charged particle tracking with graph neural networks on FPGAs (2020) arXiv:2012.01563.
[44] M. Cranmer et al., Discovering symbolic models from deep learning with inductive biases (2020) arXiv:2006.11287. Miles Cranmer is unrelated to Kyle Cranmer, though both are co-authors of the paper. See also the video presentation of the paper.
[45] Q. Cai et al., A survey on multimodal data-driven smart healthcare systems: Approaches and applications (2020) IEEE Access 7:133583\u2013133599
[46] K. Gopinath, C. Desrosiers, and H. Lombaert, Graph domain adaptation for alignment-invariant brain surface segmentation (2020) arXiv:2004.00074
[47] J. Liu et al., Identification of early mild cognitive impairment using multi-modal data and graph convolutional networks (2020) BMC Bioinformatics 21(6):1\u201312
[48] H. E. Manoochehri and M. Nourani, Drug-target interaction prediction using semi-bipartite graph model and deep learning (2020). BMC Bioinformatics 21(4):1\u201316
[49] Y. Huang and A. C. Chung, Edge-variational graph convolutional networks for uncertainty-aware disease prediction (2020) Proc. MICCAI
[50] L. Cosmo et al., Latent-graph learning for disease prediction (2020) Proc. MICCAI
[51] G. Vivar et al., Simultaneous imputation and disease classification in incomplete medical datasets using Multigraph Geometric Matrix Completion (2020) arXiv:2005.06935.
[52] X. Li and J. Duncan, BrainGNN: Interpretable brain graph neural network for fMRI analysis (2020) bioRxiv:2020.05.16.100057
[53] X. Yu et al., ResGNet-C: A graph convolutional neural network for detection of COVID-19 (2020) Neurocomputing.
[54] P. Gainza et al., Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning (2020) Nature Methods 17(2):184\u2013192.
[55] F. Sverrisson et al., Fast end-to-end learning on protein surfaces (2020) bioRxiv:2020.12.28.424589.
[56] A. Klimovskaia et al., Poincar\u00e9 maps for analyzing complex hierarchies in single-cell data (2020) Nature Communications 11.
[57] J. Jumper et al., High accuracy protein structure prediction using deep learning (2020) a.k.a. AlphaFold 2.0 (paper not yet available).
[58] J. M. Stokes et al., A deep learning approach to antibiotic discovery (2020) Cell 180(4):688\u2013702.
[59] D. Morselli Gysi et al., Network medicine framework for identifying drug repurposing opportunities for COVID-19 (2020) arXiv:2004.07229.<\/p>\n

\uff08\u53ef\u4e0a\u4e0b\u6ed1\u52a8\u6d4f\u89c8\uff09<\/p>\n


\u590d\u6742\u79d1\u5b66\u6700\u65b0\u8bba\u6587<\/p>\n

\u96c6\u667a\u6591\u56fe\u9876\u520a\u8bba\u6587\u901f\u9012\u680f\u76ee\u4e0a\u7ebf\u4ee5\u6765\uff0c\u6301\u7eed\u6536\u5f55\u6765\u81eaNature\u3001Science\u7b49\u9876\u520a\u7684\u6700\u65b0\u8bba\u6587\uff0c\u8ffd\u8e2a\u590d\u6742\u7cfb\u7edf\u3001\u7f51\u7edc\u79d1\u5b66\u3001\u8ba1\u7b97\u793e\u4f1a\u79d1\u5b66\u7b49\u9886\u57df\u7684\u524d\u6cbf\u8fdb\u5c55\u3002\u73b0\u5728\u6b63\u5f0f\u63a8\u51fa\u8ba2\u9605\u529f\u80fd\uff0c\u6bcf\u5468\u901a\u8fc7\u5fae\u4fe1\u670d\u52a1\u53f7\u300c\u96c6\u667a\u6591\u56fe\u300d\u63a8\u9001\u8bba\u6587\u4fe1\u606f\u3002\u626b\u63cf\u4e0b\u65b9\u4e8c\u7ef4\u7801\u5373\u53ef\u4e00\u952e\u8ba2\u9605\uff1a
<\/p>\n

\u63a8\u8350\u9605\u8bfb
\u56fe\u795e\u7ecf\u7f51\u7edc\u524d\u6cbf\u7efc\u8ff0\uff1a\u52a8\u6001\u56fe\u7f51\u7edc<\/a><\/p>\n

DeepMind\u518d\u53d1Nature\u89e3\u51b3\u7269\u7406\u96be\u9898\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\u5c55\u793a\u5f3a\u5927\u80fd\u529b<\/a>
PM2.5-GNN \u56fe\u7f51\u7edc\u96fe\u973e\u9884\u62a5\u6a21\u578b\uff1a\u590d\u6742\u7cfb\u7edf\u81ea\u52a8\u5efa\u6a21\u5728\u5927\u6c14\u6c61\u67d3\u4e2d\u7684\u5e94\u7528<\/a><\/p>\n

\u52a0\u5165\u96c6\u667a\uff0c\u4e00\u8d77\u590d\u6742\uff01<\/a><\/p>\n

\u70b9\u51fb\u201c\u9605\u8bfb\u539f\u6587\u201d\uff0c\u8ffd\u8e2a\u590d\u6742\u79d1\u5b66\u9876\u520a\u8bba\u6587 <\/div>\n","protected":false},"excerpt":{"rendered":"

\u5bfc\u8bed \u8fd1\u6bb5\u65f6\u95f4\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\uff08GNN\uff09\u4f5c\u4e3a\u673a\u5668\u5b66\u4e60\u6700\u70ed\u95e8\u7684\u9886\u57df\u4e4b\u4e00\uff0c\u5531\u76db\u6709\u4e4b\uff0c\u5531\u8870\u4e5f\u4e0d\u5c11\u3002\u5c31\u5728\u524d\u4e24\u5929\uff0c\u4e00\u4f4d\u6570\u636e\u79d1\u5b66\u5bb6\u53d1\u8868\u4e86\u4e00\u7bc7\u9898\u4e3a\u201c\u4e3a\u4ec0\u4e48\u6211\u5bf9GNN\u4e0d\u611f\u5192\u201d\u7684\u6587\u7ae0\uff0c\u5386\u6570 GNN \u7684\u591a\u9879\u5f0a\u7aef\uff0c\u5728reddit\u4e0a\u5f15\u53d1\u5e7f\u6cdb\u8ba8\u8bba\u3002\u4f46\u5c3d\u7ba1\u5982\u6b64\uff0cGNN\u4f5c\u4e3a\u4e00\u9879\u65b9\u5174\u672a\u827e\u7684\u7814\u7a76\u9886\u57df\uff0c\u5728 2020\u5e74\u4f9d\u7136\u53d6\u5f97\u4e86\u8bf8\u591a\u8fdb\u5c55\uff0c\u5728…<\/p>\n","protected":false},"author":1,"featured_media":23247,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/23269"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=23269"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/23269\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/23247"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=23269"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=23269"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=23269"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=23269"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}