{"id":477375,"date":"2023-08-09T09:11:34","date_gmt":"2023-08-09T09:11:34","guid":{"rendered":""},"modified":"2023-09-05T11:14:34","modified_gmt":"2023-09-05T11:14:34","slug":"graph-neural-networks","status":"publish","type":"wiki","link":"https:\/\/oneproxy.pro\/cn\/wiki\/graph-neural-networks\/","title":{"rendered":"\u56fe\u795e\u7ecf\u7f51\u7edc"},"content":{"rendered":"<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u6982\u8ff0<\/h2>\n<p>\u56fe\u795e\u7ecf\u7f51\u7edc (GNN) \u4ee3\u8868\u4e86\u673a\u5668\u5b66\u4e60\u548c\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u91cd\u5927\u53d1\u5c55\uff0c\u65e8\u5728\u6355\u83b7\u548c\u5904\u7406\u56fe\u7ed3\u6784\u6570\u636e\u3002\u672c\u8d28\u4e0a\uff0cGNN \u662f\u4e00\u79cd\u4e13\u95e8\u8bbe\u8ba1\u7528\u4e8e\u5904\u7406\u56fe\u7ed3\u6784\u6570\u636e\u7684\u795e\u7ecf\u7f51\u7edc\uff0c\u53ef\u5e2e\u52a9\u5b83\u4eec\u89e3\u51b3\u4f20\u7edf\u795e\u7ecf\u7f51\u7edc\u96be\u4ee5\u89e3\u51b3\u7684\u5404\u79cd\u95ee\u9898\u3002\u8fd9\u5305\u62ec\u4f46\u4e0d\u9650\u4e8e\u793e\u4ea4\u7f51\u7edc\u8868\u793a\u3001\u63a8\u8350\u7cfb\u7edf\u3001\u751f\u7269\u6570\u636e\u89e3\u91ca\u548c\u7f51\u7edc\u6d41\u91cf\u5206\u6790\u3002<\/p>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u5386\u53f2\u548c\u51fa\u73b0<\/h2>\n<p>GNN \u7684\u6982\u5ff5\u6700\u65e9\u51fa\u73b0\u5728 21 \u4e16\u7eaa\u521d\uff0c\u7531 Franco Scarselli\u3001Marco Gori \u7b49\u4eba\u63d0\u51fa\u3002\u4ed6\u4eec\u5f00\u53d1\u4e86\u539f\u59cb\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u4ee5\u8fed\u4ee3\u65b9\u5f0f\u5206\u6790\u8282\u70b9\u7684\u5c40\u90e8\u90bb\u57df\u3002\u7136\u800c\uff0c\u8fd9\u4e2a\u539f\u59cb\u6a21\u578b\u9762\u4e34\u7740\u8ba1\u7b97\u6548\u7387\u548c\u53ef\u6269\u5c55\u6027\u7684\u6311\u6218\u3002<\/p>\n<p>\u76f4\u5230\u56fe\u5377\u79ef\u795e\u7ecf\u7f51\u7edc (CNN)\uff08\u901a\u5e38\u79f0\u4e3a\u56fe\u5377\u79ef\u7f51\u7edc (GCN)\uff09\u7684\u51fa\u73b0\uff0cGNN \u624d\u5f00\u59cb\u53d7\u5230\u66f4\u591a\u5173\u6ce8\u3002Thomas N. Kipf \u548c Max Welling \u5728 2016 \u5e74\u7684\u5de5\u4f5c\u6781\u5927\u5730\u666e\u53ca\u4e86\u8fd9\u4e00\u6982\u5ff5\uff0c\u4e3a GNN \u9886\u57df\u5960\u5b9a\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002<\/p>\n<h2>\u6269\u5c55\u4e3b\u9898\uff1a\u56fe\u795e\u7ecf\u7f51\u7edc<\/h2>\n<p>\u56fe\u795e\u7ecf\u7f51\u7edc (GNN) \u5229\u7528\u6570\u636e\u7684\u56fe\u7ed3\u6784\u6765\u9884\u6d4b\u8282\u70b9\u3001\u8fb9\u6216\u6574\u4e2a\u56fe\u3002\u672c\u8d28\u4e0a\uff0cGNN \u5c06\u6bcf\u4e2a\u8282\u70b9\u7684\u7279\u5f81\u53ca\u5176\u90bb\u5c45\u7684\u7279\u5f81\u89c6\u4e3a\u8f93\u5165\uff0c\u901a\u8fc7\u6d88\u606f\u4f20\u9012\u548c\u805a\u5408\u6765\u66f4\u65b0\u8282\u70b9\u7684\u7279\u5f81\u3002\u6b64\u8fc7\u7a0b\u901a\u5e38\u91cd\u590d\u51e0\u6b21\u8fed\u4ee3\uff0c\u79f0\u4e3a GNN \u7684\u201c\u5c42\u201d\uff0c\u5141\u8bb8\u4fe1\u606f\u5728\u7f51\u7edc\u4e2d\u4f20\u64ad\u3002<\/p>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u5185\u90e8\u7ed3\u6784<\/h2>\n<p>GNN \u67b6\u6784\u7531\u51e0\u4e2a\u6838\u5fc3\u7ec4\u4ef6\u7ec4\u6210\uff1a<\/p>\n<ol>\n<li>\u8282\u70b9\u7279\u5f81\uff1a\u56fe\u4e2d\u7684\u6bcf\u4e2a\u8282\u70b9\u90fd\u5305\u542b\u521d\u59cb\u7279\u5f81\uff0c\u8fd9\u4e9b\u7279\u5f81\u53ef\u4ee5\u57fa\u4e8e\u771f\u5b9e\u4e16\u754c\u7684\u6570\u636e\u6216\u4efb\u610f\u8f93\u5165\u3002<\/li>\n<li>\u8fb9\u7279\u5f81\uff1a\u8bb8\u591a GNN \u8fd8\u4f7f\u7528\u8fb9\u7684\u7279\u5f81\uff0c\u8868\u793a\u8282\u70b9\u4e4b\u95f4\u7684\u5173\u7cfb\u3002<\/li>\n<li>\u6d88\u606f\u4f20\u9012\uff1a\u8282\u70b9\u805a\u5408\u6765\u81ea\u90bb\u5c45\u7684\u4fe1\u606f\u6765\u66f4\u65b0\u5176\u7279\u5f81\uff0c\u4ece\u800c\u6709\u6548\u5730\u5728\u56fe\u4e2d\u4f20\u9012\u201c\u6d88\u606f\u201d\u3002<\/li>\n<li>\u8bfb\u51fa\u51fd\u6570\uff1a\u7ecf\u8fc7\u51e0\u5c42\u4fe1\u606f\u4f20\u64ad\u540e\uff0c\u53ef\u4ee5\u5e94\u7528\u8bfb\u51fa\u51fd\u6570\u6765\u751f\u6210\u56fe\u5f62\u7ea7\u8f93\u51fa\u3002<\/li>\n<\/ol>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u4e3b\u8981\u7279\u5f81<\/h2>\n<ul>\n<li><strong>\u5904\u7406\u4e0d\u89c4\u5219\u6570\u636e\u7684\u80fd\u529b\uff1a<\/strong> GNN \u64c5\u957f\u5904\u7406\u4e0d\u89c4\u5219\u6570\u636e\uff0c\u5176\u4e2d\u5b9e\u4f53\u4e4b\u95f4\u7684\u5173\u7cfb\u5f88\u91cd\u8981\uff0c\u5e76\u4e14\u4e0d\u6613\u88ab\u4f20\u7edf\u795e\u7ecf\u7f51\u7edc\u6355\u6349\u3002<\/li>\n<li><strong>\u666e\u904d\u6027\uff1a<\/strong> GNN \u53ef\u4ee5\u5e94\u7528\u4e8e\u4efb\u4f55\u53ef\u4ee5\u8868\u793a\u4e3a\u56fe\u7684\u95ee\u9898\uff0c\u56e0\u6b64\u5176\u7528\u9014\u6781\u4e3a\u5e7f\u6cdb\u3002<\/li>\n<li><strong>\u8f93\u5165\u987a\u5e8f\u7684\u4e0d\u53d8\u6027\uff1a<\/strong> GNN \u63d0\u4f9b\u4e0d\u53d8\u7684\u8f93\u51fa\uff0c\u4e0e\u56fe\u4e2d\u8282\u70b9\u7684\u987a\u5e8f\u65e0\u5173\uff0c\u4ece\u800c\u786e\u4fdd\u4e00\u81f4\u7684\u6027\u80fd\u3002<\/li>\n<li><strong>\u6355\u6349\u5c40\u90e8\u548c\u5168\u5c40\u6a21\u5f0f\u7684\u80fd\u529b\uff1a<\/strong> GNN \u51ed\u501f\u5176\u72ec\u7279\u7684\u67b6\u6784\uff0c\u53ef\u4ee5\u63d0\u53d6\u6570\u636e\u4e2d\u7684\u5c40\u90e8\u548c\u5168\u5c40\u6a21\u5f0f\u3002<\/li>\n<\/ul>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u7c7b\u578b<\/h2>\n<table>\n<thead>\n<tr>\n<th>GNN \u7c7b\u578b<\/th>\n<th>\u63cf\u8ff0<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u56fe\u5377\u79ef\u7f51\u7edc\uff08GCN\uff09<\/td>\n<td>\u4f7f\u7528\u5377\u79ef\u8fd0\u7b97\u6765\u805a\u5408\u90bb\u57df\u4fe1\u606f\u3002<\/td>\n<\/tr>\n<tr>\n<td>\u56fe\u6ce8\u610f\u529b\u7f51\u7edc\uff08GAT\uff09<\/td>\n<td>\u5e94\u7528\u6ce8\u610f\u529b\u673a\u5236\u6765\u52a0\u6743\u90bb\u8fd1\u8282\u70b9\u7684\u5f71\u54cd\u3002<\/td>\n<\/tr>\n<tr>\n<td>\u56fe\u540c\u6784\u7f51\u7edc\uff08GIN\uff09<\/td>\n<td>\u65e8\u5728\u901a\u8fc7\u533a\u5206\u4e0d\u540c\u7684\u56fe\u7ed3\u6784\u6765\u6355\u83b7\u4e0d\u540c\u7684\u62d3\u6251\u4fe1\u606f\u3002<\/td>\n<\/tr>\n<tr>\n<td>\u56fe\u5f62\u5206\u6790\u8f6f\u4ef6<\/td>\n<td>\u5b66\u4e60\u5f52\u7eb3\u8282\u70b9\u5d4c\u5165\uff0c\u53ef\u4ee5\u9884\u6d4b\u672a\u77e5\u6570\u636e\u3002<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u5e94\u7528\u4e0e\u6311\u6218<\/h2>\n<p>GNN \u7684\u5e94\u7528\u975e\u5e38\u5e7f\u6cdb\uff0c\u4ece\u793e\u4ea4\u7f51\u7edc\u5206\u6790\u548c\u751f\u7269\u4fe1\u606f\u5b66\u5230\u4ea4\u901a\u9884\u6d4b\u548c\u7a0b\u5e8f\u9a8c\u8bc1\u3002\u7136\u800c\uff0c\u5b83\u4eec\u4e5f\u9762\u4e34\u6311\u6218\u3002\u4f8b\u5982\uff0cGNN \u5f88\u96be\u6269\u5c55\u5230\u5927\u578b\u56fe\uff0c\u800c\u8bbe\u8ba1\u5408\u9002\u7684\u56fe\u8868\u793a\u53ef\u80fd\u5f88\u590d\u6742\u3002<\/p>\n<p>\u89e3\u51b3\u8fd9\u4e9b\u6311\u6218\u901a\u5e38\u9700\u8981\u5728\u51c6\u786e\u6027\u548c\u8ba1\u7b97\u6548\u7387\u4e4b\u95f4\u8fdb\u884c\u6743\u8861\uff0c\u9700\u8981\u7cbe\u5fc3\u8bbe\u8ba1\u548c\u5b9e\u9a8c\u3002PyTorch Geometric\u3001DGL \u548c Spektral \u7b49\u5404\u79cd\u5e93\u53ef\u4ee5\u7b80\u5316\u5b9e\u65bd\u548c\u5b9e\u9a8c\u8fc7\u7a0b\u3002<\/p>\n<h2>\u4e0e\u5176\u4ed6\u795e\u7ecf\u7f51\u7edc\u7684\u6bd4\u8f83<\/h2>\n<table>\n<thead>\n<tr>\n<th>\u65b9\u9762<\/th>\n<th>\u5730\u78c1\u795e\u7ecf\u7f51\u7edc<\/th>\n<th>CNN<\/th>\n<th>\u5faa\u73af\u795e\u7ecf\u7f51\u7edc<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\u6570\u636e\u7ed3\u6784<\/td>\n<td>\u56fe\u8868<\/td>\n<td>\u7f51\u683c\uff08\u4f8b\u5982\u56fe\u50cf\uff09<\/td>\n<td>\u5e8f\u5217\uff08\u4f8b\u5982\u6587\u672c\uff09<\/td>\n<\/tr>\n<tr>\n<td>\u5173\u952e\u7279\u70b9<\/td>\n<td>\u5229\u7528\u56fe\u7ed3\u6784<\/td>\n<td>\u5229\u7528\u7a7a\u95f4\u5c40\u90e8\u6027<\/td>\n<td>\u5229\u7528\u65f6\u95f4\u52a8\u6001<\/td>\n<\/tr>\n<tr>\n<td>\u5e94\u7528\u9886\u57df<\/td>\n<td>\u793e\u4f1a\u7f51\u7edc\u5206\u6790\u3001\u5206\u5b50\u7ed3\u6784\u5206\u6790<\/td>\n<td>\u56fe\u50cf\u8bc6\u522b\u3001\u89c6\u9891\u5206\u6790<\/td>\n<td>\u8bed\u8a00\u5efa\u6a21\u3001\u65f6\u95f4\u5e8f\u5217\u5206\u6790<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u7684\u672a\u6765\u524d\u666f\u548c\u6280\u672f<\/h2>\n<p>GNN \u4ee3\u8868\u7740\u4e00\u4e2a\u4e0d\u65ad\u53d1\u5c55\u7684\u9886\u57df\uff0c\u5177\u6709\u5de8\u5927\u7684\u8fdb\u4e00\u6b65\u63a2\u7d22\u548c\u6539\u8fdb\u6f5c\u529b\u3002\u672a\u6765\u7684\u53d1\u5c55\u53ef\u80fd\u5305\u62ec\u5904\u7406\u52a8\u6001\u56fe\u3001\u63a2\u7d22 3D \u56fe\u4ee5\u53ca\u5f00\u53d1\u66f4\u9ad8\u6548\u7684\u8bad\u7ec3\u65b9\u6cd5\u3002GNN \u4e0e\u5f3a\u5316\u5b66\u4e60\u548c\u8fc1\u79fb\u5b66\u4e60\u7684\u7ed3\u5408\u4e5f\u4e3a\u7814\u7a76\u63d0\u4f9b\u4e86\u6709\u5e0c\u671b\u7684\u9014\u5f84\u3002<\/p>\n<h2>\u56fe\u795e\u7ecf\u7f51\u7edc\u548c\u4ee3\u7406\u670d\u52a1\u5668<\/h2>\n<p>\u4f7f\u7528\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u95f4\u63a5\u652f\u6301 GNN \u7684\u8fd0\u884c\u3002\u4f8b\u5982\uff0c\u5728\u6d89\u53ca\u4ece\u5404\u79cd\u5728\u7ebf\u6765\u6e90\u6536\u96c6\u6570\u636e\u7684\u5b9e\u9645\u5e94\u7528\u4e2d\uff08\u4f8b\u5982\uff0c\u7528\u4e8e\u793e\u4ea4\u7f51\u7edc\u5206\u6790\u7684\u7f51\u9875\u6293\u53d6\uff09\uff0c\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u534f\u52a9\u9ad8\u6548\u4e14\u533f\u540d\u7684\u6570\u636e\u6536\u96c6\uff0c\u4ece\u800c\u53ef\u80fd\u6709\u52a9\u4e8e\u56fe\u5f62\u6570\u636e\u96c6\u7684\u6784\u5efa\u548c\u66f4\u65b0\u3002<\/p>\n<h2>\u76f8\u5173\u94fe\u63a5<\/h2>\n<ol>\n<li><a href=\"https:\/\/ieeexplore.ieee.org\/document\/9046288\" target=\"_new\" rel=\"noopener nofollow\">\u56fe\u795e\u7ecf\u7f51\u7edc\u7efc\u5408\u7efc\u8ff0<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/1812.08434\" target=\"_new\" rel=\"noopener nofollow\">\u56fe\u795e\u7ecf\u7f51\u7edc\uff1a\u65b9\u6cd5\u4e0e\u5e94\u7528\u56de\u987e<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/1812.04202\" target=\"_new\" rel=\"noopener nofollow\">\u56fe\u6df1\u5ea6\u5b66\u4e60\uff1a\u7efc\u8ff0<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/rusty1s\/pytorch_geometric\" target=\"_new\" rel=\"noopener nofollow\">PyTorch \u51e0\u4f55\u5e93<\/a><\/li>\n<\/ol>","protected":false},"featured_media":468487,"menu_order":0,"template":"","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"class_list":["post-477375","wiki","type-wiki","status-publish","has-post-thumbnail","hentry"],"acf":{"faq_title":"Frequently Asked Questions about <mark>Graph Neural Networks: Harnessing Power from Graph-Structured Data<\/mark>","faq_items":[{"question":"What are Graph Neural Networks (GNNs)?","answer":"<p>Graph Neural Networks (GNNs) are a type of neural network designed to process and make predictions about data structured as a graph. They are particularly useful in problems where the relationships between entities are complex and cannot be efficiently captured by traditional neural networks.<\/p>"},{"question":"When was the concept of GNNs first introduced?","answer":"<p>The concept of Graph Neural Networks first emerged in the early 2000s with the work of Franco Scarselli, Marco Gori, and others. They laid the groundwork for future development of GNNs.<\/p>"},{"question":"How do GNNs work?","answer":"<p>GNNs operate by treating each node's features and its neighbors' features as inputs to update the node's feature through a process called message passing and aggregation. This process is often repeated for several iterations or \"layers\", which allows information to propagate through the network.<\/p>"},{"question":"What are some key features of GNNs?","answer":"<p>Key features of GNNs include their capability to handle irregular data, generalizability to any problem that can be represented as a graph, invariance to input order, and their ability to capture both local and global patterns in the data.<\/p>"},{"question":"What types of Graph Neural Networks exist?","answer":"<p>Several types of Graph Neural Networks exist, including Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), Graph Isomorphism Networks (GINs), and GraphSAGE.<\/p>"},{"question":"What are some applications of GNNs and what challenges do they face?","answer":"<p>Applications of GNNs are diverse and include social network analysis, bioinformatics, traffic prediction, and program verification. However, they do face challenges like scalability to large graphs and complexity in designing the appropriate graph representation.<\/p>"},{"question":"How do GNNs compare with other neural networks?","answer":"<p>Unlike Convolutional Neural Networks (CNNs) that exploit spatial locality in grid-like data (like images), and Recurrent Neural Networks (RNNs) that exploit temporal dynamics in sequential data (like text), GNNs exploit the graph structure in the data.<\/p>"},{"question":"What is the future of GNNs?","answer":"<p>The field of GNNs is rapidly growing, with potential for further exploration and improvement. Future developments may include handling dynamic graphs, exploring 3D graphs, and developing more efficient training methods.<\/p>"},{"question":"How can proxy servers be used with Graph Neural Networks?","answer":"<p>Proxy servers can indirectly support the operation of GNNs. In real-world applications like data collection from various online sources, proxy servers can assist in efficient and anonymous data collection, thereby aiding in the construction and updating of graph datasets.<\/p>"}]},"_links":{"self":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki\/477375","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki"}],"about":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/types\/wiki"}],"version-history":[{"count":0,"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki\/477375\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/media\/468487"}],"wp:attachment":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/media?parent=477375"}],"curies":[{"name":"\u53ef\u6e7f\u6027\u7c89\u5242","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}