{"id":15885,"date":"2019-06-03T17:51:22","date_gmt":"2019-06-03T09:51:22","guid":{"rendered":"https:\/\/swarma.org\/?p=15885"},"modified":"2019-06-20T11:49:07","modified_gmt":"2019-06-20T03:49:07","slug":"289fc751ca","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=15885","title":{"rendered":"GeniePath\uff1a\u81ea\u9002\u5e94\u611f\u53d7\u8def\u5f84\u7684\u56fe\u795e\u7ecf\u7f51\u7edc | \u56fe\u7f51\u7edc\u8bba\u6587\u5206\u4eab"},"content":{"rendered":"
\n

\"\u96c6\u667a\"<\/p>\n

\n

\n
\n

\u5bfc\u8bed<\/span><\/strong><\/span><\/p>\n

\n
\n

\u672c\u6587\u8be6\u7ec6\u89e3\u8bfb\u4e86\u8682\u8681\u91d1\u670d\u53d1\u8868\u4e8e KDD 2018 \u7684\u8bba\u6587 GeniePath\uff0c\u4e00\u79cd\u81ea\u9002\u5e94\u611f\u53d7\u8def\u5f84\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\uff08Graph Neural Networks, GNN\uff09\u65b9\u6cd5\u3002\u5176\u521b\u65b0\u70b9\u5728\u4e8e\u80fd\u591f\u6839\u636e\u5177\u4f53\u4efb\u52a1\u81ea\u52a8\u9009\u53d6\u6709\u7528\u7684\u9ad8\u9636\u90bb\u57df\u4fe1\u606f\u30026\u67083\u65e521:00-22:00\uff0c\u672c\u6587\u4f5c\u8005\u738b\u7855\u5c06\u5728\u96c6\u667a\u4ff1\u4e50\u90e8 B \u7ad9\u76f4\u64ad\u95f4\u89e3\u8bfb\u8fd9\u7bc7\u8bba\u6587\u3002\u76f4\u64ad\u95f4\u5730\u5740\u89c1\u6587\u672b\u3002<\/span><\/p>\n

\u96c6\u667a\u56fe\u7f51\u7edc\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\u6b63\u5f0f\u516c\u5f00\u62db\u52df\uff0c\u5982\u679c\u4f60\u4e5f\u5bf9\u56fe\u7f51\u7edc\u611f\u5174\u8da3\uff0c\u60f3\u5206\u4eab\u4f60\u6700\u8fd1\u8bfb\u8fc7\u7684\u8bba\u6587\uff0c\u6b22\u8fce\u586b\u8868\u62a5\u540d\u3002\uff08\u8be6\u60c5\u89c1\u6587\u672b\uff09<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

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

GeniePath \u5728\u8682\u8681\u91d1\u670d KDD 2018 \u53d1\u8868\u7684\u8bba\u6587\u4e2d\u88ab\u9996\u6b21\u63d0\u51fa\uff0c\u662f\u4e00\u79cd\u53ef\u6269\u5c55\u7684\u80fd\u591f\u5b66\u4e60\u81ea\u9002\u5e94\u611f\u53d7\u8def\u5f84\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\u6846\u67b6\u3002<\/span>\u5b9a\u4e49\u5728\u5177\u6709\u6392\u5217\u4e0d\u53d8\u6027\u7684\u56fe\u6570\u636e\u4e0a\uff08permutation invariant graph data\uff09\u3002\u5176\u81ea\u9002\u5e94\u8def\u5f84\u5c42\uff08adaptive path layer\uff09\u5305\u62ec\u4e24\u4e2a\u4e92\u8865\u7684\u529f\u80fd\u5355\u5143\uff0c\u5206\u522b\u7528\u6765\u8fdb\u884c\u5e7f\u5ea6\u4e0e\u6df1\u5ea6\u7684\u63a2\u7d22\uff0c\u524d\u8005\u7528\u6765\u5b66\u4e60\u4e00\u9636\u90bb\u57df\u8282\u70b9\u7684\u6743\u91cd\uff0c\u540e\u8005\u7528\u6765\u63d0\u53d6\u548c\u8fc7\u6ee4\u9ad8\u9636\u90bb\u57df\u5185\u6c47\u805a\u7684\u4fe1\u606f\u3002\u5728\u76f4\u63a8\uff08transductive\uff09\u548c\u5f52\u7eb3\uff08inductive\uff09\u4e24\u79cd\u5b66\u4e60\u4efb\u52a1\u7684\u5b9e\u9a8c\u4e2d\uff0c\u5747\u8fbe\u5230\u4e86 state-of-the-art \u7684\u6548\u679c\u3002<\/span><\/p>\n

\n

\n

\n
\n
<\/p>\n

\u56fe\u8868\u793a\u5b66\u4e60\u4e0e\u56fe\u795e\u7ecf\u7f51\u7edc<\/span><\/p>\n

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

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

\u56fe\uff08Graph\uff09\uff0c\u6216\u8005\u79f0\u4e3a\u7f51\u7edc<\/span>\uff08Network\uff09\uff0c\u662f\u4e00\u79cd\u7531\u5bf9\u8c61\uff08\u8282\u70b9\uff0cnode\uff09\u548c\u5173\u7cfb\uff08\u8fde\u8fb9\uff0cedge\uff09\u6784\u6210\u7684\u6570\u636e\u7ed3\u6784\u3002Graph Neural Networks: A Review of Methods and Applications [1]\u751f\u6d3b\u4e2d\u5f88\u591a\u6570\u636e\u6216\u7cfb\u7edf\u53ef\u4ee5\u88ab\u5efa\u6a21\u6210\u56fe\uff0c\u4f8b\u5982\u793e\u4ea4\u7f51\u7edc\uff0c\u86cb\u767d\u8d28-\u86cb\u767d\u8d28\u4ea4\u4e92\u7f51\u7edc\uff0c\u75be\u75c5\u4f20\u64ad\u7f51\u7edc\uff0c\u77e5\u8bc6\u56fe\u8c31\u7b49\u3002\u56fe\u5f3a\u5927\u7684\u5efa\u6a21\u548c\u8868\u8fbe\u80fd\u529b\u4e5f\u5438\u5f15\u4e86\u8d8a\u6765\u8d8a\u591a\u7684\u5b66\u8005\u7684\u5173\u6ce8\u3002\u4e0e\u6211\u4eec\u719f\u77e5\u7684\u57fa\u4e8e\u683c\u70b9\uff082D gird\uff09\u7684\u56fe\u50cf\u6570\u636e\u4e0d\u540c\uff0c\u56fe\u662f\u975e\u6b27\u6570\u636e\uff08non-Euclidean data\uff09\uff0c\u4e3a\u4e86\u80fd\u591f\u7528\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u8fd0\u7b97\uff0c\u6211\u4eec\u9700\u8981\u5148\u5f97\u5230\u8282\u70b9\u548c\u8fde\u8fb9\u7684\u4e00\u79cd\u5411\u91cf\u8868\u793a\uff0c\u8fd9\u9879\u6280\u672f\u88ab\u79f0\u4e3a\u56fe\u8868\u793a\u5b66\u4e60\uff08graph representation learning\uff09\uff0c\u6216\u8005\u53eb\u7f51\u7edc\u5d4c\u5165\uff08network embedding\uff09\uff0c\u4e0b\u6587\u4e0d\u505a\u533a\u5206\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n

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

\u53d7\u5230\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e2d word2vecEfficient Estimation of Word Representations in Vector Space[2]\u7684\u542f\u53d1\uff0c\u8bde\u751f\u4e86 DeepWalkDeepwalk: Online learning ofsocialrepresentations[3]<\/span>\u3001node2vecnode2vec: Scalable feature learning for networks[4]\u3001LINELINE: Large-scale Information Network Embedding [5]\u7b49\u7f51\u7edc\u5d4c\u5165\u7b97\u6cd5\uff0c\u5b83\u4eec\u7684\u601d\u8def\u662f\u5c06\u8282\u70b9\u7c7b\u6bd4\u4e3a\u81ea\u7136\u8bed\u8a00\u4e2d\u7684\u5355\u8bcd\uff0c\u5728\u56fe\u4e0a\u751f\u6210\u4e00\u4e9b\u7ecf\u8fc7\u968f\u673a\u6e38\u8d70\uff08Random Walk\uff09\u4ea7\u751f\u7684\u8282\u70b9\u5e8f\u5217\uff0c\u628a\u8fd9\u4e9b\u5e8f\u5217\u770b\u6210\u53e5\u5b50\u3002\u7c7b\u4f3c\u4e8e word2vec \u7684\u601d\u60f3\uff0c\u6bcf\u4e2a\u8282\u70b9\u90fd\u53ef\u4ee5\u7528\u5904\u4e8e\u5176\u4e0a\u4e0b\u6587\uff08context\uff09\u7684\u8282\u70b9\u6765\u8868\u793a\uff0c\u5f97\u5230\u4e00\u4e2a\u957f\u5ea6\u8fdc\u5c0f\u4e8e\u56fe\u4e0a\u8282\u70b9\u6570\u7684\u4f4e\u7ef4\u7a20\u5bc6\u7684\u5b9e\u503c\u5411\u91cf\u3002\u8be5\u5411\u91cf\u7f16\u7801\u4e86\uff08encode\uff09 \u56fe\u7684\u7ed3\u6784\u4fe1\u606f\uff0c\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u8be5\u5411\u91cf\u505a\u8282\u70b9\u5206\u7c7b\u3001\u94fe\u8def\u9884\u6d4b\u7b49\u4e0b\u6e38\u4efb\u52a1\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u5f88\u591a\u4eba\uff08\u5c24\u5176\u662f\u505a\u8ba1\u7b97\u673a\u89c6\u89c9\u7684\u4eba\uff09\u53ef\u80fd\u5bf9\u5d4c\u5165\u8fd9\u9879\u6280\u672f\u5e76\u4e0d\u719f\u6089\u3002\u56e0\u4e3a\u56fe\u50cf\u5929\u7136\u5c31\u662f\u7528\u4f4e\u7ef4\u7a20\u5bc6\u7684\u5b9e\u503c\u5411\u91cf\u6765\u8868\u793a\u7684\u3002\u56fe\u50cf\u4e0a\uff0c\u6bcf\u4e2a\u50cf\u7d20\u70b9\u90fd\u5bf9\u5e94\u4e00\u4e2a RGB \u503c\uff0c\u6bcf\u4e2a RGB \u901a\u9053\u90fd\u662f [0, 255] \u7684\u6570\u3002\u6b64\u5916\uff0c\u989c\u8272\u7a7a\u95f4\u6709\u591a\u79cd\uff0c\u6bd4\u5982 RGB\u3001HSV \u7b49\uff0c\u6211\u4eec\u4f1a\u5728\u4e0d\u540c\u7684\u4efb\u52a1\u4e2d\u9009\u53d6\u66f4\u9002\u5408\u7684\u989c\u8272\u8868\u793a\u65b9\u5f0f\u3002\uff08\u7528 Photoshop \u8c03\u8fc7\u7167\u7247\u7684\u670b\u53cb\u53ef\u80fd\u4f1a\u66f4\u6709\u4f53\u4f1a\u3002\uff09<\/span><\/p>\n

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

\"\u96c6\u667a\"\"\u96c6\u667a\"<\/p>\n

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

\u540c\u6837\u7684\u9053\u7406\uff0c\u5d4c\u5165\u5f97\u5230\u7684\u5411\u91cf\u5316\u7684\u8282\u70b9\u548c\u8fde\u8fb9\u7684\u8868\u793a\u4e5f\u662f\u4e3a\u4e86\u56fe\u4e0a\u7684\u4e0b\u6e38\u4efb\u52a1\u6765\u670d\u52a1\u7684\u3002\u800c\u57fa\u4e8e\u968f\u673a\u6e38\u8d70\u7684\u65b9\u6cd5\u5f97\u5230\u7684\u8868\u793a\u4fe1\u606f\u5f80\u5f80\u65e0\u6cd5\u9488\u5bf9\u5177\u4f53\u4efb\u52a1\u8fdb\u884c\u4f18\u5316\uff0c\u6b64\u5916\uff0c\u968f\u673a\u6e38\u8d70\u7b97\u6cd5\u53ea\u80fd\u5f97\u5230\u56fe\u4e0a\u7684\u7ed3\u6784\u4fe1\u606f\uff0c\u800c\u65e0\u6cd5\u878d\u5408\u8282\u70b9\u7684\u7279\u5f81\u4fe1\u606f\u3002\u56e0\u6b64\uff0c\u7814\u7a76\u4eba\u5458\u53c8\u63d0\u51fa\u4e86\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\u7684\u56fe\u8868\u793a\u5b66\u4e60\uff0c\u79f0\u4e3a\u56fe\u795e\u7ecf\u7f51\u7edc Graph Neural Network \uff08GNN\uff09\u3002\u611f\u5174\u8da3\u7684\u670b\u53cb\u53ef\u4ee5\u9605\u8bfb\u76f8\u5173\u7684\u7efc\u8ff0\u6587\u7ae0\u3002A Gentle Introduction to Graph Neural Network<\/span>[6]<\/span>\uff0cGraph Neural Networks: A Review of Methods and Applications[7]\uff0cA Comprehensive Survey on Graph Neural Networks[8\u3002\u8fd9\u4e9b\u7b97\u6cd5\u7684\u57fa\u672c\u601d\u8def\u662f\uff0c\u9488\u5bf9\u7279\u5b9a\u4e0b\u6e38\u4efb\u52a1\u7528\u7aef\u5230\u7aef (end-to-end) \u7684\u65b9\u5f0f\u8fdb\u884c\u5b66\u4e60\u8bad\u7ec3\uff0c\u5e76\u4e14\u53ef\u4ee5\u878d\u5408\u8282\u70b9\u7684\u7279\u5f81\u4fe1\u606f\uff0c\u6bd4\u5982cora [9]\u5f15\u6587\u6570\u636e\u96c6\u4e2d\uff0c\u8282\u70b9\u7684\u7ed3\u6784\u4fe1\u606f\u5305\u542b\u4e86\u6587\u7ae0\u7684\u6458\u8981\uff0c\u8fd9\u5bf9\u4e8e\u505a\u8282\u70b9\u5206\u7c7b\u548c\u94fe\u8def\u9884\u6d4b\u4f1a\u8d77\u5230\u91cd\u8981\u4f5c\u7528\u3002\u6211\u4eec\u77e5\u9053\uff0c\u5728 CNN \u4e2d\uff0c\u5377\u79ef\u6838\u6240\u5b9a\u4e49\u7684\u662f\u5bf9\u56fe\u50cf\u4e0a\u6bcf\u4e2a\u50cf\u7d20\u70b9\u4e0e\u5176\u5468\u56f4\u50cf\u7d20\u70b9\u8fdb\u884c\u4fe1\u606f\u4ea4\u4e92\u7684\u65b9\u5f0f\u3002\u540c\u6837\uff0cGNN\u4e2d\uff0c\u6bcf\u4e2a\u8282\u70b9\u4e5f\u6709\u90bb\u57df\u7684\u6982\u5ff5\uff08\u7531\u56fe\u7684\u7ed3\u6784\u4fe1\u606f\u6240\u5b9a\u4e49\uff09\uff0c\u56fe\u4e0a\u7684\u5377\u79ef\u6838\u4e5f\u662f\u8981\u8fdb\u884c\u8282\u70b9\u548c\u5176\u90bb\u5c45\u7684\u4fe1\u606f\u4ea4\u4e92\u3002\u8fd9\u91cc\uff0c\u5377\u79ef\u6838\u7684\u5927\u5c0f\u5f80\u5f80\u662f\u56fa\u5b9a\u7684\uff0c\u53d7\u9650\u4e8e\u8fd0\u7b97\u590d\u6742\u5ea6\uff0c\u6211\u4eec\u4e00\u822c\u4f7f\u7528\u4e00\u9636\u6216\u8005\u4e8c\u9636\u90bb\u57df\u3002<\/span><\/p>\n

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

\u6709\u4ee3\u8868\u6027\u7684 GNN \u7b97\u6cd5\u5305\u62ec\uff0cGCNSemi-supervised classification with graph convolutional networks[10], GATGraph Attention Networks[11]\uff0cGraphSAGEInductive Representation Learning on Large Graphs[12]\u3002\u6b64\u5916\uff0c\u5f00\u6e90\u793e\u533a\u4e5f\u51fa\u73b0\u4e86\u4e00\u4e9b GNN \u7684\u5f00\u53d1\u6846\u67b6\uff0c\u5176\u4e2d\uff0cPyTorch Geometric (PyG) [13]\u57fa\u4e8e PyTorch \u5b9e\u73b0\uff0c\u719f\u6089 PyTorch \u7684\u4eba\u53ef\u4ee5\u5f88\u8212\u670d\u5730\u4f7f\u7528 PyG \u8fdb\u884c GNN \u7b97\u6cd5\u7684\u5f00\u53d1\u3002PyG \u63d0\u4f9b\u4e86\u4e00\u4e2a\u57fa\u7840\u7ec4\u4ef6\uff0cMessagePassing\u3002\u5b83\u8ba4\u4e3a\u50cf GCN\u3001GAT\u3001GraphSAGE \u7b49\u7b97\u6cd5\u53ef\u4ee5\u62bd\u8c61\u4e3a\u5982\u56fe\u6240\u793a\u7684\u7ec4\u4ef6\u3002\u6bcf\u4e00\u5c42\u56fe\u5377\u79ef\u5c42\u8981\u505a\u7684\u5c31\u662f\u548c\u76f8\u5e94\u9886\u57df<\/span>\"\u96c6\u667a\"\u5185\u7684\u8282\u70b9\u8fdb\u884c\u4fe1\u606f\u7684\u4ea4\u4e92\u3002\u4e0d\u540c\u7684\u7b97\u6cd5\u7684\u533a\u522b\u5728\u4e8e\u5176\u6240\u5b9a\u4e49\u7684\u6d88\u606f\u7684\u6620\u5c04\u65b9\u5f0f\u548c\u9886\u57df\u7684\u9009\u53d6\u4e0d\u901a\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\n

\n
\n
<\/p>\n

\u53ef\u53d8\u5f62\u5377\u79ef\u6838<\/span><\/p>\n

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

Deformable Convolutional Networks[14] \uff0c\u8fd9\u662f\u4e00\u7bc7\u6709\u8da3\u7684\u6587\u7ae0\uff0c\u6211\u4eec\u4ee5\u5f80\u719f\u77e5\u7684\u5377\u79ef\u6838\u662f\u4e0b\u56fe\u5de6\u4e0a\u89d2\u6240\u793a\u7684\u90a3\u79cd\u89c4\u5219\u5f62\u72b6\u7684\uff0c\u800c\u8fd9\u7bc7\u6587\u7ae0\u63d0\u51fa\u4e86\u4e00\u79cd\u53ef\u53d8\u5f62\u5377\u79ef\u6838\uff0c\u5b83\u4f1a\u6839\u636e\u5177\u4f53\u7684\u4efb\u52a1\u548c\u8f93\u5165\u6570\u636e\u800c\u6539\u53d8\u5377\u79ef\u6838\u7684\u5f62\u72b6\uff0c\u4e5f\u5c31\u662f\u611f\u53d7\u91ce\u3002\u5e76\u4e14\u5728\u76ee\u6807\u68c0\u6d4b\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u6bd4\u8f83\u597d\u7684\u6548\u679c\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

GeniePath \u7684\u4f5c\u8005\u53ef\u80fd\u662f\uff08\u8bba\u6587\u4e2d\u6ca1\u6709\u63d0\u5230\uff09\u53d7\u5230\u4e86\u8fd9\u7bc7\u6587\u7ae0\u7684\u542f\u53d1\uff0c\u5c06\u8fd9\u4e2a idea \u5f15\u5165\u5230\u4e86 Graph \u4e2d\u3002\u63d0\u51fa\u4e86\u81ea\u9002\u5e94\u611f\u53d7\u91ce\u7684 GNN \u7b97\u6cd5\u3002\u4f46\u662f\u8be5\u7b97\u6cd5\u7684\u5e76\u4e0d\u662f\u901a\u8fc7\u8c03\u6574\u8282\u70b9\u7684\u9886\u57df\u6765\u5b9e\u73b0\u7684\uff0c\u800c\u5c06\u8ddd\u79bb\u591a\u8df3\uff08k-hop\uff09\u7684\u8282\u70b9\u7684\u4fe1\u606f\u5b58\u50a8\u5728 LSTM \u7684 memory \u4e2d\uff0c\u7531\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u5b66\u4e60\u81ea\u52a8\u5224\u65ad\u54ea\u4e9b\u4fe1\u606f\u5bf9\u4e8e\u81ea\u5df1\u5b8c\u6210\u4e0b\u6e38\u4efb\u52a1\u662f\u6709\u5229\u7684\uff0c\u800c\u8fdb\u884c\u63d0\u53d6\u548c\u8fc7\u6ee4\u3002<\/span><\/p>\n

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

\n

\n
\n
<\/p>\n

\u6392\u5217\u4e0d\u53d8\u6027<\/span><\/p>\n

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

\n

\u8fd9\u91cc\u63d0<\/span>\u5230\u4e86\u4e00\u4e2a\u6982\u5ff5<\/span>\uff0c\u6392\u5217\u4e0d\u53d8\u6027\uff08permutation invariant\uff09\u3002<\/span>\u8fd9\u91ccDeepSets: Modeling Permutation Invariance\u7ed9\u51fa\u4e86\u4e00\u4e2a\u901a\u4fd7\u7684\u89e3\u91ca\u3002[15]<\/span><\/p>\n

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

f(a, b, c)=f(a, c, b)=f(b, a, c)=\u2026\u540c\u6837\u7684\u53c2\u6570\u8f93\u5165\u5230\u51fd\u6570\u4e2d\uff0c\u5b83\u4eec\u7684\u6392\u5217\u987a\u5e8f\u5e76\u4e0d\u5f71\u54cd\u7ed3\u679c\u3002\u6570\u636e\u7ed3\u6784\u2014\u2014\u96c6\u5408\uff08set\uff09\u4f1a\u7528\u5230\u8fd9\u6837\u7684\u4e0d\u53d8\u6027\uff0c\u76f8\u5e94\u5730\uff0c\u90a3\u4e9b\u5728\u5efa\u6a21\u4e2d\u7528\u5230\u96c6\u5408\u6982\u5ff5\u7684\u95ee\u9898\u90fd\u8981\u9075\u5faa\u6392\u5217\u4e0d\u53d8\u6027\uff0c\u4f8b\u5982\u70b9\u4e91\uff0c\u591a\u4e3b\u4f53\u5f3a\u5316\u5b66\u4e60\uff0c\u56fe\u7247\u573a\u666f\u4e2d\u591a\u4e2a\u76ee\u6807\u7684\u96c6\u5408\u7b49\u3002\u5728 DeepMind \u6587\u7ae0Graph Matching Networks for Learning the Similarity of Graph Structured Objects [16]\u4e2d\u4e5f\u6709\u63d0\u5230\u6392\u5e8f\u4e0d\u53d8\u6027\u3002”In the past few years graph neural networks (GNNs) have emerged as an effective class of models for learning representations of structured data and for solving various supervised prediction problems on graphs.Such models are invariant to permutations of graph elements<\/strong>by design and compute graph node representations through a propagation process which iteratively aggregates local structural information. These node representations are then used directly for node classification, or pooled into a graph vector for graph classification. “<\/span><\/p>\n

\n

\u672c\u6587\u5728\u7b97\u6cd5\u90e8\u5206\uff08Proposed Approaches\uff09\u8be6\u7ec6\u4ecb\u7ecd\u4e86\u6392\u5217\u4e0d\u53d8\u6027\u3002\u51fd\u6570\u7a7a\u95f4\u8981\u6ee1\u8db3\u56fe\u6570\u636e\u6240\u9700\u8981\u7684\u6392\u5217\u4e0d\u53d8\u6027\u3002\u6211\u4eec\u8981\u5b66\u7684\u662f\u4e00\u4e2a\u51fd\u6570 f\uff0c\u5c06\u56fe G \u6620\u5c04\u5230\u5d4c\u5165 H\u3002\u6211\u4eec\u8981\u5047\u8bbe\u7684\u662f\u8be5 f \u5177\u6709\u6392\u5217\u4e0d\u53d8\u6027\uff0c\u5b66\u4e60\u4efb\u52a1\u4e0e\u90bb\u5c45\u8282\u70b9\u7684\u987a\u5e8f\u662f\u65e0\u5173\u7684\u3002\u6587\u4e2d\u7528\u4e0b\u9762\u516c\u5f0f\u6765\u5b9a\u4e49\u805a\u5408\uff08aggregator\uff09\u51fd\u6570 f \u7684\u8fd9\u4e2a\u6027\u8d28\uff0c\u5176\u4e2d sigma \u8868\u793a\u7684\u662f\u4efb\u610f\u7684\u6392\u5217\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\u7136\u540e\u63d0\u5230\u4e86\u4e00\u4e2a\u5b9a\u7406\uff0c\u5f53\u4e14\u4ec5\u5f53\u51fd\u6570 f \u53ef\u4ee5\u88ab\u5206\u89e3\u4e3a<\/span>\"\u96c6\u667a\",\u5176\u4e2d phi \u548c rho \u662f\u4e24\u4e2a\u6620\u5c04\u3002\u90a3\u4e48 f \u662f\u5177\u6709\u6392\u5217\u4e0d\u53d8\u5f62\u7684\u3002\u5e76\u4e14\u4e24\u4e2a\u5177\u6709\u6392\u5217\u4e0d\u53d8\u5f62\u7684\u51fd\u6570\u7684\u590d\u5408\u51fd\u6570g\u2218<\/span>f<\/span><\/span>\u53ef\u4ee5\u7b49\u4ef7\u4e8e g(f(x))\u3002\u4e5f\u662f\u6ee1\u8db3\u6392\u5217\u4e0d\u53d8\u6027\u7684\uff0c\u56e0\u6b64\uff0c\u6211\u4eec\u53ef\u4ee5\u5c06\u51fd\u6570\u8fdb\u884c\u5806\u53e0\u6765\u8bbe\u8ba1\u795e\u7ecf\u7f51\u7edc\u3002GraphSAGE \u63d0\u4f9b LSTM \u7684\u805a\u5408\u5668\uff0c\u8fd9\u79cd\u805a\u5408\u5668\u662f\u4e0d\u5177\u6709 Permutation In<\/span>variant \u7684\uff0c\u56e0\u4e3a LSTM \u63a5\u6536\u7684\u662f\u904d\u5386\u8282\u70b9\u7684\u9690\u53d8\u91cf\u5e8f\u5217\uff0c\u6392\u5e8f\u662f\u654f\u611f\u7684\u3002\u6b64\u5916\uff0c\u8fd9\u91cc\u7684 Permutation Invariant \u4e5f\u6307\u56fe\u662f\u9759\u6001\u7684\uff0c\u4e0d\u968f\u65f6\u95f4\u6f14\u5316\u3002 \u56e0\u4e3a\u6587\u7ae0\u4e2d\u63d0\u5230 \u201cThis is in opposition to temporal graphs.\u201d<\/span><\/p>\n

\n

\n

\n
\n
<\/p>\n

GeniePath \u7b97\u6cd5<\/span><\/p>\n

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

Adaptive Path Layer \u662f\u6587\u7ae0\u7684\u6838\u5fc3\u5185\u5bb9\uff0c\u76f8\u6bd4\u5176\u4ed6\u7b97\u6cd5\u4e2d\u4f7f\u7528\u4e8b\u5148\u5b9a\u4e49\u597d\u7684\u8def\u5f84\uff08pre-defined paths\uff09\uff0c\u4f8b\u5982\u4e00\u9636\u90bb\u5c45\u7b49\uff0cAdaptive Path Layer \u76ee\u6807\u662f\u81ea\u5b66\u4e60\u611f\u53d7\u8def\u5f84\uff0c\u4fe1\u53f7\u5728\u5b66\u5230\u7684\u8def\u5f84\u4e0a\u4f20\u64ad\u3002\u8fd9\u4e2a\u95ee\u9898\u53ef\u4ee5\u7b49\u6548\u4e8e\u5b66\u4e60\u6bcf\u4e2a\u8282\u70b9\u7684\u4e00\u4e2a\u5408\u7406\u7684\u5b50\u56fe\uff0c\u8be5\u5b50\u56fe\u5305\u62ec\u4e24\u90e8\u5206\uff0c\u5e7f\u5ea6\u65b9\u5411\u4e0a\u7ed9\u51fa\u4e00\u9636\u90bb\u5c45\u4e2d\u8282\u70b9\u7684\u6743\u91cd\uff0c\u6df1\u5ea6\u65b9\u5411\u4e0a t \u8df3\u8303\u56f4\u5185\u91cd\u8981\u7684\u90bb\u5c45\u3002\u81ea\u9002\u5e94\u5e7f\u5ea6\u51fd\u6570\u5b9a\u4e49\u4e3a<\/span>\"\u96c6\u667a\"\u5176\u4e2d theta \u4ee3\u8868\u53c2\u6570\u3002H \u4ee3\u8868\u7684\u662f\u8282\u70b9\u7684\u5d4c\u5165\uff08\u6216\u8005\u53eb\u9690\u72b6\u6001\u8868\u793a\uff09\u3002\u81ea\u9002\u5e94\u6df1\u5ea6\u51fd\u6570\u5b9a\u4e49\u4e3a\uff0c<\/span>\"\u96c6\u667a\"\u5176\u4e2d\uff0cPhi \u4ee3\u8868\u53c2\u6570\u3002<\/span><\/p>\n

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

\u6211\u4eec\u53ef\u4ee5\u5c06\u672c\u6587\u548c GAT \u6216\u8005 GraphSAGE \u7b49\u5e38\u7528\u7684\u975e\u9891\u57df\u7684 GNN \u6846\u67b6\u7684\u533a\u522b\u8ba4\u4e3a\u662f\u5728\u5b83\u4eec\u7684\u57fa\u7840\u4e0a\u52a0\u4e86 memory\uff08\u901a\u8fc7 LSTM \u5b9e\u73b0\uff09\u3002\u5b58\u50a8\u901a\u8fc7\u4e00\u9636\u90bb\u57df\u8fdb\u884c\u7684\u4fe1\u606f\u4f20\u9012\u5f97\u5230\u7684\u9ad8\u9636\u9886\u57df\u8282\u70b9\u7684\u4fe1\u606f\u3002\u6240\u8c13\u7684\u81ea\u9002\u5e94\uff0c\u81ea\u5b66\u4e60\uff0c\u5176\u5b9e\u5c31\u662f\u901a\u8fc7 LSTM \u7684\u95e8\u5bf9 LSTM \u7684 memory \u6240\u5b58\u50a8\u7684\u9ad8\u9636\u9886\u57df\u4fe1\u606f\u7684\u63d0\u53d6\u548c\u8fc7\u6ee4 \uff08\u5206\u522b\u5bf9\u5e94\u9057\u5fd8\u95e8\u548c\u8f93\u51fa\u95e8\uff09\u6240\u4f53\u73b0\u7684\u3002\u800c\u4e00\u9636\u90bb\u57df\u548c\u9ad8\u9636\u9886\u57df\u7684\u4fe1\u606f\u4ea4\u4e92\u4f53\u73b0\u5728 LSTM \u7684\u8f93\u5165\u95e8\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

\u6ce8\u610f\uff0c\u5e7f\u5ea6\u51fd\u6570\u7684\u8f93\u51fa\u7684 H \u4e0a\u6807\u662f tmp\uff0c\u53ea\u6709\u5c06 H_tmp \u518d\u8f93\u5165\u5230\u6df1\u5ea6\u51fd\u6570\u4e2d\uff0c\u624d\u8fbe\u5230\u4e00\u4e2a epoch \u7684\u8bad\u7ec3\u3002<\/span><\/p>\n

\n

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

\"\u96c6\u667a\"<\/p>\n

\n

\u672c\u8d28\u6765\u8bb2\uff0c\u8fd9\u91cc\u662f\u4e00\u4e2a\u53bb\u6389\u4e86 multi-head \u673a\u5236\u7684 GAT\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\u81ea\u9002\u5e94\u6df1\u5ea6\u51fd\u6570\u672c\u8d28\u4e0a\u5c31\u662f\u4e00\u4e2a LSTM\u3002\u5982\u56fe\u6240\u793a\uff0cLSTM \u4e00\u822c\u63a5\u6536\u7684\u662f\u4e00\u4e2a\u5e8f\u5217\uff0c\u5e8f\u5217\u6bcf\u4e2a\u5143\u7d20\u53ef\u4ee5\u5bf9\u5e94\u4e8e\u5728\u65f6\u95f4\u6b65\u4e0a\u5c55\u5f00\u540e\u7684 LSTM \u7684\u6bcf\u4e2a\u65f6\u523b\u7684\u8f93\u5165\u3002<\/span><\/p>\n

\n

\"\u96c6\u667a\"<\/p>\n

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

\u800c GeniePath \u81ea\u9002\u5e94\u6df1\u5ea6\u51fd\u6570\u63a5\u6536\u7684\u8f93\u5165\u4ec5\u662f t \u65f6\u523b\u8282\u70b9\u7684\u5185\u90e8\u8868\u793a\uff0c\u800c\u975e\u65f6\u95f4\u5e8f\u5217\u3002\u5176\u76ee\u7684\u662f\u4e3a\u4e86\u5229\u7528 LSTM \u7684\u5b58\u50a8\u673a\u5236\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/span><\/p>\n

\u5176\u4e2d\u6700\u91cd\u8981\u7684\u516c\u5f0f\u662f\u6700\u540e\u4e00\u4e2a\uff08\u5bf9\u5e94\u4e8e\u56fe\u4e0a\u65b0\u7684\u8f93\u5165\u548c memory \u7684\u201c+\u201d\u64cd\u4f5c\uff09\uff0c\u8f93\u51fa\u7684\u8282\u70b9\u8868\u793a\u7b49\u4e8e\u5b58\u50a8\u4e8e LSTM memory \u7684\u4fe1\u606f\u548c\u63d0\u53d6\u548c\u8fc7\u6ee4\u3002\u800c\u63d0\u53d6\u548c\u8fc7\u6ee4\u7684\u63a7\u5236\u662f\u7531\u5e7f\u5ea6\u51fd\u6570\u7684\u8f93\u5165\u51b3\u5b9a\u7684\uff0c\u5373 h_tmp\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

\n

\u8bba\u6587\u4ec5\u7ed9\u51fa\u4e86 Adaptive Path Layer\uff0c\u800c\u6ca1\u6709\u6a21\u578b\u7684\u7ed3\u6784\u56fe\u3002\u8fd9\u91cc\u6839\u636e\u5bf9\u6587\u7ae0\u7684\u7406\u89e3\u753b\u51fa\u795e\u7ecf\u7f51\u7edc\u6846\u67b6\u3002\u6587\u7ae0\u63d0\u51fa\u4e24\u4e2a\u6846\u67b6\uff0c\u4e00\u4e2a\u662f\u76f4\u63a5\u5c06 Adaptive Path Layer \u5faa\u73af T \u6b21\uff0c\u90a3\u4e48\u5c06 t-hop \u7684\u8282\u70b9\u4fe1\u606f\u7684\u4ea4\u4e92\u53ef\u4ee5\u878d\u5408\u8d77\u6765\u3002\u53e6\u4e00\u4e2a\u662f GeniePath-lazy\uff0c\u533a\u522b\u5728\u5728\u4e8e\u5148\u53ea\u8fc7\u5e7f\u5ea6\u51fd\u6570\uff0c\u5f97\u5230<\/span>\"\u96c6\u667a\"\uff0c\u518d\u6574\u4f53\u8f93\u5165\u5230\u6df1\u5ea6\u51fd\u6570\u4e2d\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

\"\u96c6\u667a\"<\/p>\n

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

\u4e0b\u9762\u5217\u51fa\u6587\u4e2d\u4f7f\u7528\u7684\u6570\u636e\u4fe1\u606f\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\u4e0b\u9762\u662f\u6587\u7ae0\u7ed9\u51fa\u7684\u5b9e\u9a8c\u7ed3\u679c\uff0c\u5bf9\u6bd4\u7684\u7b97\u6cd5\u5305\u62ec MLP\uff0cnode2vec\uff0c\u4ee5\u53ca\u51e0\u4e2a\u7ecf\u5178\u7684 GCN \u7b97\u6cd5\u3002\u5176\u4e2d MLP \u8f93\u5165\u7684\u4ec5\u662f\u8282\u70b9\u7684\u7279\u5f81\u4fe1\u606f\uff0c\u65e0\u6cd5\u5229\u7528\u56fe\u7ed3\u6784\u4fe1\u606f\u3002\u800c node2vec \u8f93\u5165\u7684\u4ec5\u662f\u56fe\u7ed3\u6784\u4fe1\u606f\uff0c\u65e0\u6cd5\u5229\u7528\u8282\u70b9\u7684\u7279\u5f81\u4fe1\u606f\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

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

\u6211\u4eec\u80fd\u770b\u5230 GeniePath \u5728\u86cb\u767d\u8d28\u4ea4\u4e92\u7f51\u7edc\u6570\u636e PPI \u4e0a\u7684\u8868\u73b0\u663e\u8457\u5f3a\u4e8e\u5176\u4ed6\u7b97\u6cd5\uff0c\u6211\u731c\u60f3\u5176\u4e2d\u7684\u539f\u56e0\u4e0e PPI \u7684\u6570\u636e\u7279\u70b9\u6709\u5173\u3002\u5728\u6587\u7ae0Link Prediction Based on Graph Neural Networks[17] \u6709\u8fd9\u6837\u4e00\u6bb5\u8bdd\u3002\u201cthe common neighbor heuristic assumes that two nodes are more likely to connect if they have many common neighbors. This assumption may be correct in social networks, but is shown to fail in protein-protein interaction (PPI) networks –two proteins sharing many common neighbors are actually less likely to interact\u201d \uff0c\u4e00\u822c\u60c5\u51b5\u4e0b\uff0c\u6211\u4eec\u4f1a\u5047\u8bbe\u4e24\u4e2a\u8282\u70b9\u6709\u66f4\u591a\u7684\u5171\u540c\u90bb\u5c45\uff0c\u5b83\u4eec\u4f1a\u503e\u5411\u4e8e\u66f4\u52a0\u76f8\u4f3c\uff0c\u4f46\u662f\u5728\u86cb\u767d\u8d28\u4e2d\uff0c\u5982\u679c\u4e24\u4e2a\u86cb\u767d\u8d28\u6709\u8d8a\u591a\u7684\u5171\u540c\u8282\u70b9\uff0c\u5b83\u4eec\u66f4\u503e\u5411\u4e8e\u4e0d\u53d1\u751f\u53cd\u5e94\u3002GeniePath \u901a\u8fc7 LSTM \u7ec4\u4ef6\u878d\u5408\u4e86\u56fe\u4e2d\u7684\u9ad8\u9636\u90bb\u5c45\u4fe1\u606f\uff0c\u4f7f\u5f97\u5f97\u5230\u66f4\u9ad8\u7684\u51c6\u786e\u7387\u3002<\/span><\/p>\n

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

\"\u96c6\u667a\"<\/p>\n

\"\u96c6\u667a\"<\/p>\n

GeniePath \u7684 PyTorch \u590d\u73b0\uff0c\u540e\u7eed\u4f1a\u53d1\u5230 https:\/\/shawnwang.tech\uff0c https:\/\/github.com\/shawnwang-tech\/GeniePath-pytorch\u3002 \u611f\u5174\u8da3\u7684\u670b\u53cb\u53ef\u4ee5\u5173\u6ce8\u3002<\/span><\/p>\n

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

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

\u270e<\/span>\u5f15\u7528\u6587\u7ae0\u94fe\u63a5<\/span><\/span><\/p>\n

\n

\n<\/section>\n

\n
\n
\n

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

[1]: Graph Neural Networks: A Review of Methods and Applications<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/1812.08434.pdf<\/span><\/p>\n

[2]: Efficient Estimation of Word Representations in Vector Space<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740:<\/span>https:\/\/arxiv.org\/abs\/1301.3781<\/span><\/p>\n

[3]: Deepwalk: Online learning of social representations<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/1403.6652<\/span><\/p>\n

[4]: node2vec: Scalable feature learning for networks<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1a<\/span>https:\/\/arxiv.org\/abs\/1607.00653<\/span><\/p>\n

[5]\uff1aLINE: Large-scale Information Network Embedding<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1a<\/span>https:\/\/arxiv.org\/abs\/1503.03578) <\/span><\/p>\n

[6]: A Gentle Introduction to Graph Neural Network<\/span><\/p>\n

<\/span>\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/towardsdatascience.com\/a-gentle-introduction-to-graph-neural-network-basics-deepwalk-and-graphsage-db5d540d50b3<\/span><\/p>\n

[7]: Graph Neural Networks: A Review of Methods and Applications<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1a<\/span>https:\/\/arxiv.org\/abs\/1812.08434<\/span><\/p>\n

[8]: A Comprehensive Survey on Graph Neural Networks<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/1901.00596<\/span><\/p>\n

[9]: cora<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1a<\/span>https:\/\/relational.fit.cvut.cz\/dataset\/CORA<\/span><\/p>\n

[10]: Semi-supervised classification with graph convolutional networks<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740:<\/span>https:\/\/arxiv.org\/abs\/1609.02907<\/span><\/p>\n

[11]: Graph Attention Networks<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740:<\/span>https:\/\/arxiv.org\/abs\/1710.10903<\/span><\/p>\n

[12]: Inductive Representation Learning on Large Graphs<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740:<\/span>https:\/\/arxiv.org\/abs\/1706.02216<\/span><\/p>\n

[13]: PyTorch Geometric (PyG)\uff1a<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740:<\/span><\/p>\n

<\/span>https:\/\/rusty1s.github.io\/pytorch_geometric\/build\/html\/index.html<\/span><\/p>\n

[14]:Deformable Convolutional Networks \uff1a<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/1703.06211<\/span><\/p>\n

[15]:DeepSets: Modeling Permutation Invariance\uff1a<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/www.inference.vc\/deepsets-modeling-permutation-invariance\/<\/span><\/p>\n

[16]: Graph Matching Networks for Learning the Similarity of Graph Structured Objects :<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/1904.12787<\/span><\/span><\/p>\n

[17]: Link Prediction Based on Graph Neural Networks \uff1a<\/span><\/p>\n

\u94fe\u63a5\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/1802.09691<\/span><\/p>\n

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

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

\u5217\u8868\u53ef\u4e0a\u4e0b\u6ed1\u52a8<\/span><\/strong><\/span><\/p>\n

\n

\n
\n
\n

\u76f4\u64ad\u9884\u544a\uff1a\u56fe\u7f51\u7edc\u8bba\u6587\u89e3\u8bfb<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n

6\u67083\u65e5\uff08\u5468\u4e00\uff0921:00-22:00\uff0c\u96c6\u667a\u56fe\u7f51\u7edc\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\u5c06\u8fdb\u884c\u65b0\u4e00\u671f\u7684\u8bba\u6587\u5206\u4eab\u76f4\u64ad\uff0c\u6b22\u8fce\u611f\u5174\u8da3\u7684\u670b\u53cb\u53c2\u4e0e\u3002<\/span><\/p>\n

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

\u76f4\u64ad\u4e3b\u9898\uff1aGeniePath\uff1a\u81ea\u9002\u5e94\u611f\u53d7\u8def\u5f84\u7684\u56fe\u795e\u7ecf\u7f51\u7edc<\/span><\/p>\n

\u4e3b\u8bb2\u4eba\uff1a\u738b\u7855<\/span><\/p>\n

\u76f4\u64ad\u5730\u5740\uff1a\u96c6\u667a\u4ff1\u4e50\u90e8 B \u7ad9\u76f4\u64ad\u95f4<\/span><\/p>\n

\n

\ud83d\udc40\u5173\u6ce8B\u7ad9\u4e3b\u64ad\u201c\u96c6\u667a\u4ff1\u4e50\u90e8\u201d<\/strong><\/strong><\/strong><\/p>\n

\u4e0d\u9519\u8fc7\u6bcf\u4e00\u573a\u96c6\u667a\u91cd\u78c5\u76f4\u64ad<\/strong><\/strong><\/p>\n

\n

\"\u96c6\u667a\"<\/p>\n

\u76f4\u64ad\u5730\u5740\uff1a<\/strong><\/span><\/p>\n

https:\/\/live.bilibili.com\/h5\/8091531<\/strong><\/span><\/p>\n

\u76f4\u64ad\u65f6<\/strong>\u95f4 \uff1a21:00-22:00<\/strong><\/p>\n

\n

\n
\n
\n

\u96c6\u667a\u56fe\u7f51\u7edc\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\u516c\u5f00\u62db\u52df<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n

\u56fe\u795e\u7ecf\u7f51\u7edc\u662f\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u524d\u6cbf\u70ed\u70b9\u8bae\u9898\uff0c\u5c24\u5176\u662f\u56fe\u7f51\u7edc\uff08Graph Networks\uff09\u63d0\u51fa\u4ee5\u6765\uff0c\u6df1\u5ea6\u5b66\u4e60\u6709\u4e86\u5b9e\u73b0\u56e0\u679c\u63a8\u7406\u7684\u6f5c\u529b\u3002\u4e3a\u4e86\u6301\u7eed\u8ffd\u8e2a\u76f8\u5173\u9886\u57df\u7684\u524d\u6cbf\u8fdb\u5c55\uff0c\u96c6\u667a\u4ff1\u4e50\u90e8\u8054\u5408\u5317\u5e08\u5927\u7cfb\u7edf\u79d1\u5b66\u5b66\u9662\u5f20\u6c5f\u8bfe\u9898\u7ec4\uff0c\u7ec4\u7ec7\u4e86\u4ee5\u56fe\u7f51\u7edc\u4e3a\u4e3b\u9898\u7684\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\uff0c\u7814\u8ba8\u6700\u65b0\u8bba\u6587\uff0c\u5b55\u80b2\u7814\u7a76\u601d\u8def\u3002<\/span><\/p>\n

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

\u6bcf\u4e00\u671f\u7ebf\u4e0a\u8bfb\u4e66\u4f1a\u7531\u4e00\u4f4d\u6210\u5458\u4e3b\u8bb2\uff0c\u5f62\u5f0f\u4e3a\u8bba\u6587\u5206\u4eab\uff0c\u65f6\u95f4\u4e3a\u6bcf\u5468\u4e00 21:00-22:00\u3002\u52a0\u5165\u8bfb\u4e66\u4f1a\u7fa4\u9700\u62a5\u540d\u5ba1\u6838\uff0c\u539f\u5219\u4e0a\u53c2\u4e0e\u8005\u5e94\u6709\u80fd\u529b\u72ec\u7acb\u5b8c\u6210\u4e00\u6b21\u7ebf\u4e0a\u5206\u4eab\u3002\u5982\u679c\u4f60\u4e5f\u6b63\u5728\u4ece\u4e8b\u56fe\u7f51\u7edc\u4e0e\u6df1\u5ea6\u5b66\u4e60\u65b9\u9762\u7684\u7814\u7a76\u5de5\u4f5c\u6216\u6280\u672f\u5b9e\u8df5\uff0c\u6216\u8005\u5bf9\u8be5\u9886\u57df\u6709\u5f3a\u70c8\u7684\u5b66\u4e60\u610f\u613f\uff0c\u6b22\u8fce\u586b\u5199\u62a5\u540d\u8868\uff0c\u7533\u8bf7\u52a0\u5165\u201c\u96c6\u667a\u56fe\u7f51\u7edc\u8bba\u6587\u5206\u4eab\u5c0f\u7ec4\u201d\uff01<\/span><\/p>\n

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

\u62a5\u540d\u8bf7\u70b9\u51fb\u4e0b\u65b9\u5c0f\u7a0b\u5e8f\uff0c\u586b\u5199\u62a5\u540d\u8868\u3002\u586b\u8868\u4e4b\u540e\u4f1a\u6709\u5165\u7fa4\u65b9\u5f0f\u3002<\/span><\/p>\n<\/p>\n

\n

\n

\n
\n
\n

\u4f5c\u8005\uff1a\u738b\u7855<\/span><\/p>\n

\u7f16\u8f91\uff1a\u9648\u5b89\u6797<\/p>\n<\/section>\n<\/blockquote>\n<\/section>\n

\n

\n
\n
<\/p>\n

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

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

\n

\u56fe\u7f51\u7edc\u6df1\u5ea6\u89e3\u6790\uff1a\u4e3a\u4ec0\u4e48\u8bf4\u56fe\u7f51\u7edc\u662f AI \u7684\u672a\u6765\uff1f<\/a><\/strong><\/p>\n

\u56fe\u7f51\u7edc\u91cd\u78c5\u7efc\u8ff0\uff1a\u57fa\u4e8e\u56fe\u7684\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5<\/strong><\/a><\/p>\n

\u6df1\u5ea6\u5b66\u4e60\u672a\u8fbe\u9884\u671f\uff0c\u56fe\u7f51\u7edc\u6709\u671b\u5f15\u9886\u4e0b\u4e00\u6b21AI\u9769\u547d<\/strong><\/a><\/p>\n

\u56fe\u7f51\u7edc\u2014\u2014\u6084\u7136\u5174\u8d77\u7684\u6df1\u5ea6\u5b66\u4e60\u65b0\u6d6a\u6f6e<\/a><\/strong><\/p>\n

\u56fe\u7f51\u7edc\u5165\u95e8\u7efc\u8ff0\uff1a\u4ece\u56fe\u5d4c\u5230\u56fe\u5206\u7c7b<\/span><\/a><\/strong><\/span><\/p>\n

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

\n
\n
\n
\n
\"\u96c6\u667a\"<\/div>\n
\n
\n
\n
\n
\n
\n

\u96c6\u667a\u4ff1\u4e50\u90e8QQ\u7fa4\uff5c877391004<\/span><\/strong><\/span><\/p>\n

\u5546\u52a1\u5408\u4f5c\u53ca\u6295\u7a3f\u8f6c\u8f7d\uff5cswarma@swarma.org<\/span><\/strong><\/span><\/p>\n

\n

\u25c6<\/span>\u25c6<\/span>\u25c6<\/span><\/span><\/strong><\/strong><\/span><\/h1>\n<\/section>\n

\u641c\u7d22\u516c\u4f17\u53f7\uff1a\u96c6\u667a\u4ff1\u4e50\u90e8<\/span><\/strong><\/span><\/p>\n<\/p>\n

\u52a0\u5165\u201c\u6ca1\u6709\u56f4\u5899\u7684\u7814\u7a76\u6240\u201d<\/span><\/strong><\/span><\/p>\n

\"\u96c6\u667a\"<\/span><\/strong><\/span><\/p>\n

\u8ba9\u82f9\u679c\u7838\u5f97\u66f4\u731b\u70c8\u4e9b\u5427\uff01<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n

\u539f\u6587\u59cb\u53d1\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7\uff08\u96c6\u667a\u4ff1\u4e50\u90e8\uff09\uff1a\u96c6\u667a<\/a><\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

\u5bfc\u8bed \u672c\u6587\u8be6\u7ec6\u89e3\u8bfb\u4e86\u8682\u8681\u91d1\u670d\u53d1\u8868\u4e8e KDD 2018 \u7684\u8bba\u6587 GeniePath\uff0c\u4e00\u79cd\u81ea\u9002\u5e94\u611f\u53d7\u8def\u5f84\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\uff08Graph Neural Networks, GNN\uff09\u65b9\u6cd5\u3002\u5176\u521b\u65b0\u70b9\u5728\u4e8e\u80fd\u591f\u6839\u636e\u5177\u4f53\u4efb\u52a1\u81ea\u52a8\u9009\u53d6\u6709\u7528\u7684\u9ad8\u9636\u90bb\u57df\u4fe1\u606f\u30026\u67083\u65e521:00-22:00\uff0c\u672c\u6587\u4f5c\u8005\u738b\u7855\u5c06\u5728\u96c6\u667a\u4ff1\u4e50\u90e8 B \u7ad9\u76f4\u64ad…<\/p>\n","protected":false},"author":1,"featured_media":15888,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[27],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/15885"}],"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=15885"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/15885\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/15888"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=15885"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=15885"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=15885"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=15885"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}