{"id":43750,"date":"2023-07-13T19:20:09","date_gmt":"2023-07-13T11:20:09","guid":{"rendered":"https:\/\/swarma.org\/?p=43750"},"modified":"2023-07-13T19:20:09","modified_gmt":"2023-07-13T11:20:09","slug":"%e5%87%a0%e4%bd%95%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%ef%bc%9a%e8%ae%a9%e7%89%a9%e7%90%86%e4%b8%96%e7%95%8c%e6%8b%a5%e6%9c%89ai-%e9%bb%84%e6%96%87%e7%82%b3%e5%88%86%e4%ba%ab%e6%95%b4%e7%90%86","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=43750","title":{"rendered":"\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\uff1a\u8ba9\u7269\u7406\u4e16\u754c\u62e5\u6709AI | \u9ec4\u6587\u70b3\u5206\u4eab\u6574\u7406"},"content":{"rendered":"
\n
\n
\n
\n

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

\n
\n
\n
\n
\n
\n

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

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

\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\u4ece\u5fae\u89c2\u5206\u5b50\u7a7a\u95f4\u4e2d\u7684\u86cb\u767d\u8d28\u3001\u6297\u4f53\u3001\u6676\u4f53\u7b49\u7269\u8d28\uff0c\u5230\u5b8f\u89c2\u7269\u7406\u4e16\u754c\u4e2d\u4e0d\u540c\u5f62\u72b6\u7684\u7269\u4f53\uff0c\u5747\u5177\u6709\u7279\u5b9a\u7684\u7a7a\u95f4\u51e0\u4f55\u7ed3\u6784\uff0c\u6784\u6210\u4e00\u7c7b\u91cd\u8981\u7684\u6570\u636e\u5f62\u6001\u2014\u2014\u51e0\u4f55\u56fe\u3002\u4e0e\u793e\u4ea4\u7f51\u7edc\u4e2d\u7684\u62d3\u6251\u56fe\u4e0d\u540c\uff0c\u51e0\u4f55\u56fe\u4e2d\u7684\u8282\u70b9\u5360\u636e\u4e00\u5b9a\u7684\u7a7a\u95f4\u4f4d\u7f6e\uff0c\u9700\u8981\u6ee1\u8db3\u5bf9\u79f0\u6027\u7b49\u7269\u7406\u6027\u8d28\uff0c\u4f20\u7edf\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\u96be\u4ee5\u5904\u7406\uff0c\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u7531\u6b64\u4ea7\u751f\u3002<\/span><\/strong><\/section>\n

<\/span><\/strong><\/section>\n
\u5728\u96c6\u667a\u4ff1\u4e50\u90e8\u300c\u56fe\u795e\u7ecf\u7f51\u7edc\u4e0e\u7ec4\u5408\u4f18\u5316<\/a>\u300d\u8bfb\u4e66\u4f1a\u4e2d\uff0c\u4e2d\u56fd\u4eba\u6c11\u5927\u5b66\u9ad8\u74f4\u4eba\u5de5\u667a\u80fd\u5b66\u9662\u52a9\u7406\u6559\u6388\u9ec4\u6587\u70b3\u4ecb\u7ecd\u4e86\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u53ca\u5176\u5728\u5fae\u89c2\u5206\u5b50\u7a7a\u95f4\u548c\u5b8f\u89c2\u5177\u8eab\u7a7a\u95f4\u7684\u5e7f\u6cdb\u5e94\u7528\u573a\u666f\u3002\u4eca\u5929\u7684\u6587\u7ae0\u662f\u6b64\u6b21\u8bfb\u4e66\u4f1a\u7684\u6587\u5b57\u6574\u7406\u7a3f\u3002<\/span><\/strong><\/section>\n

<\/span><\/strong><\/span><\/strong><\/section>\n
\u6b22\u8fce\u626b\u7801\u89c2\u770b\u8bfb\u4e66\u4f1a\u89c6\u9891\u56de\u653e\uff1a<\/span><\/strong><\/span><\/strong><\/section>\n

<\/p>\n


<\/p>\n

\u7814\u7a76\u9886\u57df\uff1a\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\uff0c\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u4e0d\u53d8\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u7b49\u53d8\u56fe\u795e\u7ecf\u7f51\u7edc\uff0c\u5206\u5b50\u8868\u793a<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/section>\n
<\/span><\/strong><\/span><\/strong><\/section>\n<\/section>\n
\n
<\/span><\/strong><\/section>\n<\/section>\n
\n
<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
<\/strong><\/span><\/strong><\/section>\n<\/section>\n
\n

\u9ec4\u6587\u70b3<\/span><\/span> <\/span>| \u8bb2\u8005<\/strong><\/span><\/strong><\/p>\n

\u674e\u53ef <\/span>| <\/strong><\/span>\u6574\u7406<\/strong><\/span><\/p>\n

\u6881\u91d1<\/span> <\/span>| <\/strong><\/span>\u7f16\u8f91<\/strong><\/strong><\/span><\/p>\n


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


<\/p>\n

\n
\n
\n
\n
\n
\n
\n

\u76ee\u5f55<\/strong><\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n

    \n
  • \n
    \u80cc\u666f\uff1a\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60<\/span><\/strong><\/section>\n<\/li>\n
  • \n
    \u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc<\/span><\/strong><\/section>\n<\/li>\n
  • \n
    \u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u5728\u5fae\u89c2\u5206\u5b50\u7a7a\u95f4\u7684\u5e94\u7528<\/span><\/strong><\/section>\n<\/li>\n<\/ul>\n
      \n
    • \n
      \u6297\u4f53\u8bbe\u8ba1<\/span><\/strong><\/section>\n<\/li>\n
    • \n
      \u6676\u4f53\u751f\u6210<\/span><\/strong><\/section>\n<\/li>\n
    • \n
      \u50ac\u5316\u5242\u8bbe\u8ba1<\/span><\/strong><\/section>\n<\/li>\n<\/ul>\n
        \n
      • \n
        \u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u5728\u5b8f\u89c2\u5177\u8eab\u7a7a\u95f4\u7684\u5e94\u7528<\/span><\/strong><\/section>\n<\/li>\n<\/ul>\n
          \n
        • \n
          \u7269\u7406\u573a\u666f\u6a21\u62df<\/span><\/strong><\/section>\n<\/li>\n
        • \n
          \u667a\u80fd\u4f53\u8fd0\u52a8\u63a7\u5236<\/span><\/strong><\/section>\n<\/li>\n<\/ul>\n

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

          \n
          \n
          \n
          \n
          \n

          <\/section>\n

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

          \u80cc\u666f\uff1a\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60<\/span><\/strong><\/strong><\/p>\n<\/section>\n

          \n

          <\/section>\n

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

          <\/section>\n
          \u7269\u7406\u7a7a\u95f4\u7684\u6570\u636e\u548c\u4efb\u52a1\u975e\u5e38\u590d\u6742\uff0c\u4ece\u5fae\u89c2\u7684\u5206\u5b50\u7a7a\u95f4<\/strong>\u5230\u5b8f\u89c2\u7684\u5177\u8eab\u7a7a\u95f4<\/strong>\uff0c\u4e0d\u4ec5\u7814\u7a76\u5c3a\u5ea6\u53d1\u751f\u5de8\u5927\u53d8\u5316\uff0c\u6570\u636e\u7ed3\u6784\u4e5f\u6ca1\u6709\u7edf\u4e00\u7684\u5f62\u6001\u3001\u552f\u4e00\u7684\u5f62\u5f0f\uff0c\u5e76\u4e14\u53d8\u91cf\u95f4\u7684\u5173\u8054\u65b9\u5f0f\u4e5f\u662f\u591a\u6837\u7684\u3002\u5982\u4f55\u5904\u7406\u5982\u6b64\u590d\u6742\u591a\u6837\u7684\u6570\u636e\u7ed3\u6784\uff0c\u662f\u4e00\u4e2a\u5341\u5206\u5177\u6709\u6311\u6218\u7684\u95ee\u9898\u3002<\/span><\/section>\n

          <\/section>\n\n\n\n
          \n
          <\/section>\n<\/td>\n
          		\n
          <\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          \u56fe1. \u4ece\u5fae\u89c2\u5206\u5b50\u7a7a\u95f4\u5230\u5b8f\u89c2\u5177\u8eab\u7a7a\u95f4\uff0c\u6570\u636e\u7ed3\u6784\u590d\u6742<\/span><\/section>\n

          <\/section>\n
          \u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60<\/strong>\u65e8\u5728\u5bfb\u627e\u4e00\u79cd\u901a\u7528\u5316\u7684\u6570\u636e\u8868\u793a\uff0c\u5e76\u4ee5\u6b64\u4e3a\u57fa\u7840\u8bbe\u8ba1\u6a21\u578b\uff0c\u5176\u6838\u5fc3\u662f\u4ece\u5bf9\u79f0\u6027\u6216\u8005\u4e0d\u53d8\u6027\u51fa\u53d1\uff0c\u5c06\u51e0\u4f55\u6027\u8d28\u5d4c\u5165\u5230\u673a\u5668\u5b66\u4e60\u6d41\u7a0b\u5f53\u4e2d\uff0c\u66f4\u597d\u5730\u523b\u753b\u6570\u636e\u7684\u5185\u5728\u7ed3\u6784\u548c\u53d8\u5316\u89c4\u5f8b\uff0c\u63d0\u9ad8\u7b97\u6cd5\u7684\u8868\u793a\u3001\u6cdb\u5316\u3001\u901a\u7528\u80fd\u529b\u3002<\/strong><\/span><\/section>\n
          \n

          <\/section>\n<\/section>\n
          \n
          \n
          \n
          \n
          \n
          \n
          Michael Bronstein \u662f\u725b\u6d25\u5927\u5b66\u8ba1\u7b97\u673a\u79d1\u5b66\u6559\u6388\uff0c\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\uff08geometric deep learning\uff09\u6982\u5ff5\u7684\u63d0\u51fa\u8005\u3002<\/span><\/section>\n

          <\/span><\/section>\n
          \u5ef6\u4f38\u9605\u8bfb\uff1a<\/span><\/section>\n
          Michael M. Bronstein, Joan Bruna, Taco Cohen, Petar Veli\u010dkovi\u0107. Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges, https:\/\/arxiv.org\/abs\/2104.13478<\/span><\/section>\n

          <\/span><\/section>\n
          Michael Bronstein et al, Geometric foundations of Deep Learning. https:\/\/towardsdatascience.com\/geometric-foundations-of-deep-learning-94cdd45b451d<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n


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

          \u4e3e\u4f8b\u6765\u8bf4\uff0c\u4e0b\u56fe\u4e2d\u5de6\u8fb9\u7684\u4e09\u89d2\u5f62\u7ecf\u8fc7\u65cb\u8f6c\u53d8\u6362\uff0c\u4e0e\u53f3\u8fb9\u7684\u4e09\u89d2\u5f62\u662f\u5168\u7b49\u7684\uff0c\u4e24\u8005\u5bf9\u5e94\u7684\u8fb9\u957f\u548c\u5939\u89d2\u662f\u76f8\u7b49\u7684\uff0c\u8fd9\u4fbf\u662f\u6b27\u5f0f\u7a7a\u95f4\u4e2d\u7684\u5bf9\u79f0\u6027\uff0c\u6216\u8005\u8bf4\u4e0d\u53d8\u6027\u3002<\/span><\/section>\n

          <\/span><\/section>\n
          <\/section>\n
          \u56fe2\uff1a\u65cb\u8f6c\u4e0d\u53d8\u6027\u793a\u610f\u56fe<\/span><\/section>\n

          <\/section>\n
          \u4ece\u5bf9\u79f0\u6027\u51fa\u53d1\uff0c\u6211\u4eec\u53ef\u4ee5\u5bf9\u4e0d\u540c\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u505a\u7edf\u4e00\u7684\u89e3\u91ca\u548c\u5206\u7c7b\u3002<\/strong>\u6bd4\u5982\u8bf4\uff0cCNN\u5177\u6709\u5e73\u79fb\u4e0d\u53d8\u6027\uff0c\u7f51\u7edc\u53ef\u4ee5\u5bf9\u51fa\u73b0\u5728\u56fe\u7247\u4e2d\u4e0d\u540c\u4f4d\u7f6e\u7684\u540c\u4e00\u7269\u4f53\u505a\u540c\u6837\u7684\u5206\u7c7b\uff1bRNN \u5177\u6709\u65f6\u79fb\u4e0d\u53d8\u6027\uff0c\u8f93\u5165\u5e8f\u5217\u6574\u4f53\u524d\u79fb\u6216\u540e\u79fb\u82e5\u5e72 token \u5e76\u4e0d\u4f1a\u5f71\u54cd\u7f51\u7edc\u7684\u8f93\u51fa\uff1bGNN \u5219\u5177\u6709\u987a\u5e8f\u4e0d\u53d8\u6027\uff0c\u4e0d\u7ba1\u4e00\u4e2a\u8282\u70b9\u7684\u90bb\u5c45\u4ee5\u600e\u6837\u7684\u987a\u5e8f\u8f93\u5165\u7f51\u7edc\uff0c\u7ecf\u8fc7\u805a\u5408\u51fd\u6570\u5f97\u5230\u7684\u7ed3\u679c\u90fd\u662f\u4e00\u6837\u7684\u3002<\/span><\/section>\n

          <\/span><\/section>\n
          <\/section>\n
          \u56fe3. \u4ece\u5bf9\u79f0\u6027\u89d2\u5ea6\u53ef\u4ee5\u7edf\u4e00\u4e0d\u540c\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b<\/span><\/section>\n

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

          \n
          \n
          \n
          \n
          \n

          <\/section>\n

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

          \u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc<\/span><\/strong><\/strong><\/p>\n<\/section>\n

          \n

          <\/section>\n

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

          <\/section>\n
          \u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\u4e2d\u7684\u4e00\u4e2a\u91cd\u8981\u6a21\u578b\u662f\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc<\/span>\uff08Geometric GNN\uff09<\/span>\u3002<\/span><\/section>\n

          <\/section>\n
          \u79d1\u5b66\u9886\u57df\u4e2d\u7684\u5206\u5b50\u3001\u86cb\u767d\u8d28\u3001\u6297\u4f53\u3001\u6676\u4f53\u7b49\u7269\u8d28\uff0c\u4ee5\u53ca\u5b8f\u89c2\u7269\u7406\u4e16\u754c\u4e2d\u4e0d\u540c\u5f62\u72b6\u7684\u7269\u4f53\uff0c\u5747\u5177\u6709\u7279\u5b9a\u7684\u7a7a\u95f4\u51e0\u4f55\u7ed3\u6784\uff0c\u6784\u6210\u4e86\u4e00\u7c7b\u91cd\u8981\u7684\u6570\u636e\u5f62\u6001\u2014\u2014\u51e0\u4f55\u56fe<\/strong>\u3002\u4e0e\u793e\u4ea4\u7f51\u7edc\u4e2d\u7684\u62d3\u6251\u56fe\u4e0d\u540c\uff0c\u51e0\u4f55\u56fe\u4e2d\u7684\u8282\u70b9\u5360\u636e\u4e00\u5b9a\u7684\u7a7a\u95f4\u4f4d\u7f6e<\/strong>\u3002<\/span><\/section>\n

          <\/section>\n
          \u8fd1\u5e74\u6765\uff0c\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u7531\u4e8e\u6709\u6548\u5d4c\u5165\u4e86\u5bf9\u79f0\u6027\uff0c\u5177\u6709\u826f\u597d\u7684\u89e3\u91ca\u6027\u3001\u6cdb\u5316\u6027\u548c\u901a\u7528\u6027\uff0c\u5728\u5206\u5b50\u7cfb\u7edf\u8868\u793a\u4e0a\u5f97\u5230\u4e86\u5e7f\u6cdb\u5e94\u7528\u3002\u4ee5\u5c0f\u5206\u5b50\u7684\u8868\u793a<\/strong>\u4e3a\u4f8b\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u5230\uff0c\u6a21\u578b\u4e0d\u65ad\u53d1\u5c55\uff0c\u53ef\u4ee5\u8868\u5f81\u6108\u52a0\u9ad8\u7ef4\u7684\u51e0\u4f55\u6027\u8d28\uff1a<\/span><\/section>\n

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

          \u56fe4\uff1a\u5c0f\u5206\u5b50\u8868\u5f81\u6a21\u578b\u7684\u53d1\u5c55\uff1a\u591a\u5c42\u611f\u77e5\u673a\uff08MLP\uff09\uff0c\u9012\u5f52\u795e\u7ecf\u7f51\u7edc\uff08RNN\uff09\uff0c\u56fe\u795e\u7ecf\u7f51\u7edc\uff08GNN\uff09\uff0c\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\uff08Geometric GNN\uff09<\/span><\/p>\n


          <\/section>\n
          \u6700\u521d\u7684 MLP \u4f9d\u8d56\u7684\u662f\u4eba\u4e3a\u63d0\u53d6\u7684\u7279\u5f81\uff0c\u5373\u4e0d\u540c\u7684\u5b98\u80fd\u56e2\u662f\u5426\u5b58\u5728\u4e8e\u5206\u5b50\u4e2d<\/span>\uff080\u7ef4\uff09<\/span>\uff1bRNN \u4ee5\u5206\u5b50\u5e8f\u5217\u4e3a\u8f93\u5165\uff0c\u5728\u4e00\u7ef4\u7a7a\u95f4\u4e2d\u8868\u5f81\u5206\u5b50\u5e8f\u5217\u7ed3\u6784\uff1bGNN \u5219\u4ee5\u5e73\u9762\u5316\u7684\u5206\u5b50\u56fe\u4e3a\u8f93\u5165\uff0c\u5728\u4e8c\u7ef4\u7a7a\u95f4\u4e2d\u8868\u5f81\u5206\u5b50\u7684\u62d3\u6251\u7ed3\u6784\u3002\u4ee5\u4e0a\u51e0\u79cd\u8868\u5f81\u5206\u5b50\u7684\u65b9\u6cd5\u90fd\u8fdb\u884c\u4e86\u4e0d\u540c\u7a0b\u5ea6\u7684\u7b80\u5316\u3002\u800c\u6700\u65b0\u7684\u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u4ee5\u5206\u5b50\u7684\u51e0\u4f55\u56fe\u4e3a\u8f93\u5165\uff0c\u5728\u4e09\u7ef4\u7a7a\u95f4\u4e2d\u8868\u5f81\u5206\u5b50\u7684\u7a7a\u95f4\u7ed3\u6784<\/strong>\u3002<\/span><\/section>\n

          <\/section>\n
          \u4e0b\u56fe\u5c55\u793a\u4e86\u4e8c\u7ef4\u7684\u62d3\u6251\u56fe<\/strong>\u548c\u4e09\u7ef4\u7684\u51e0\u4f55\u56fe<\/strong>\u7684\u533a\u522b\u3002\u9664\u4e86\u539f\u5b50\u7c7b\u578b\u548c\u539f\u5b50\u952e\u7c7b\u578b\u4e4b\u5916\uff0c\u51e0\u4f55\u56fe\u8fd8\u5305\u62ec\u4e86\u539f\u5b50\u7684\u4e09\u7ef4\u5750\u6807\uff0c\u56e0\u6b64\u4e5f\u5305\u542b\u4e86\u952e\u957f\u952e\u89d2\u7b49\u4fe1\u606f\u3002<\/span><\/section>\n

          <\/span><\/section>\n
          <\/section>\n
          \u56fe5\uff1a\u4ece\u62d3\u6251\u56fe\u5230\u51e0\u4f55\u56fe<\/span><\/section>\n

          <\/section>\n
            \n
          • \n
            \u4e0d\u53d8\u6027\u4e0e\u7b49\u53d8\u6027<\/strong><\/span><\/section>\n<\/li>\n<\/ul>\n

            <\/section>\n
            \u4ee5\u4e0a\u662f\u4ece\u8f93\u5165\u6570\u636e\u7684\u7279\u5f81\u5bf9\u6a21\u578b\u505a\u51fa\u7684\u533a\u5206\u3002\u6b64\u5916\uff0c\u6211\u4eec\u8fd8\u53ef\u4ee5\u4ece\u5bf9\u79f0\u6027\u51fa\u53d1\u7406\u89e3\u4e0d\u540c\u7684\u6a21\u578b\u3002\u4e3a\u6b64\u6211\u4eec\u9996\u5148\u9700\u8981\u4e86\u89e3\u4e0e\u5bf9\u79f0\u6027\u6709\u5173\u7684\u4e24\u4e2a\u6982\u5ff5\uff1a\u4e0d\u53d8\u6027\uff08Invariance\uff09<\/strong>\u4e0e\u7b49\u53d8\u6027\uff08Equivariance\uff09<\/strong>\u3002\u4e0b\u56fe\u4ee5\u4e09\u7ef4\u7a7a\u95f4\u91cc\u7684\u836f\u7269\u5206\u5b50\u4e3a\u4f8b\u8fdb\u884c\u8bf4\u660e\u3002<\/span><\/section>\n

            <\/span><\/section>\n
            \u4e00\u79cd\u836f\u7269\u7684\u6548\u679c\u901a\u5e38\u4e0e\u5b83\u548c\u9776\u70b9\u7ed3\u5408\u7684\u7262\u56fa\u4e0e\u5426\u6709\u5173\uff0c\u5e76\u4e14\u53ef\u4ee5\u901a\u8fc7\u81ea\u7531\u80fd\u8fdb\u884c\u5224\u65ad\uff1a\u81ea\u7531\u80fd\u8d8a\u4f4e\uff0c\u7ed3\u5408\u5f3a\u5ea6\u8d8a\u5927\u3002\u81ea\u7531\u80fd\u5c31\u662f\u4e00\u4e2a\u5177\u6709\u65cb\u8f6c\u4e0d\u53d8\u6027\u7684\u91cf<\/strong>\uff1a\u65e0\u8bba\u6211\u4eec\u4ece\u4f55\u79cd\u89d2\u5ea6\u89c2\u5bdf\u8fd9\u4e2a\u7269\u7406\u4e8b\u4ef6\uff0c\u6216\u8005\u5982\u4f55\u65cb\u8f6c\u8fd9\u4e2a\u5206\u5b50\u4ee5\u53ca\u4e0e\u5b83\u7ed3\u5408\u7684\u86cb\u767d\u8d28\uff0c\u4e24\u8005\u7ed3\u5408\u7684\u81ea\u7531\u80fd\u90fd\u4e0d\u4f1a\u6539\u53d8\u3002\u7528\u6570\u5b66\u8bed\u8a00\u6765\u8bf4\uff0c<\/sub>\u4ee3\u8868\u5173\u6ce8\u7684\u7269\u7406\u5bf9\u8c61\u7684\u5750\u6807\u96c6\u5408\uff0c\u65cb\u8f6c\u53ef\u4ee5\u8868\u793a\u4e3a\u65cb\u8f6c\u77e9\u9635R\uff0c\u65cb\u8f6c\u540e\u7684\u5750\u6807\u4e3a<\/sub>\uff0c<\u6c42\u81ea\u7531\u80fd> \u8868\u793a\u4e3a\u51fd\u6570Energy()\uff0c\u90a3\u4e48<\/sub>\u3002\u6211\u4eec\u53ef\u80fd\u4e5f\u4f1a\u5173\u5fc3\u836f\u7269\u5206\u5b50\u7684\u8fd0\u52a8\u8f68\u8ff9<\/strong>\uff1a\u7ed9\u5b9at\u65f6\u523b\u7684\u5750\u6807<\/sub>\uff0c\u6211\u4eec\u60f3\u7528\u4e00\u4e2a\u51fd\u6570<\/sub>\u9884\u6d4b\u4e0b\u4e00\u65f6\u523b\u7684\u5750\u6807<\/sub>\uff0c\u90a3\u4e48<\/sub>\u5c31\u5e94\u8be5\u662f\u5177\u6709\u65cb\u8f6c\u7b49\u53d8\u6027<\/strong>\u7684\uff1a\u7ecf\u8fc7\u65cb\u8f6cR\uff0c\u4e0b\u4e00\u65f6\u523b\u7684\u5750\u6807<\/sub>\u53d8\u6210\u4e86<\/sub>\uff0c<\/sub>\u3002<\/span><\/section>\n

            <\/span><\/section>\n\n\n\n
            \n
            <\/section>\n<\/td>\n
            		\n
            <\/section>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

            \u56fe6\uff1a\uff08\u5de6\uff09\u4e09\u7ef4\u7a7a\u95f4\u4e2d\u7684\u65cb\u8f6c\u4e0d\u53d8\u6027\u793a\u610f\u56fe\uff0c\uff08\u53f3\uff09\u4e09\u7ef4\u7a7a\u95f4\u4e2d\u7684\u65cb\u8f6c\u7b49\u53d8\u6027\u793a\u610f\u56fe<\/span><\/p>\n


            <\/section>\n
            \u501f\u52a9\u7b49\u53d8\u6027\u548c\u4e0d\u53d8\u6027\u7684\u6982\u5ff5\uff0c\u6211\u4eec\u6765\u770b\u770b\u51e0\u79cd\u4e0d\u540c GNN \u7684\u7279\u70b9:<\/span><\/section>\n

            <\/section>\n
            \uff081\uff09One-layer GNN<\/strong><\/span><\/section>\n
            <\/section>\n
            \u56fe7\uff1aOne-layer GNN\uff0c\u4e0d\u6ee1\u8db3\u65cb\u8f6c\u7b49\u53d8\u6027<\/span><\/section>\n

            <\/section>\n
            \u5355\u5c42\u7684 GNN \u901a\u8fc7 message computation<\/span>\uff08\u4e5f\u79f0message passing\uff09<\/span>\u548c feature update \u4e24\u6b65\u64cd\u4f5c\u5f97\u5230\u4e00\u4e2a\u8f93\u51fa\u3002\u5176\u4e2d message computation \u4ee5\u5f53\u524d\u8282\u70b9\u548c\u4efb\u4e00\u90bb\u5c45\u8282\u70b9\u7684\u7279\u5f81\u4f5c\u4e3a\u8f93\u5165\uff0c\u7ecf\u8fc7\u4e00\u4e2a MLP\uff0c\u5f97\u5230\u8be5\u90bb\u5c45\u7684message\uff1bfeature update \u5219\u662f\u4ee5\u8282\u70b9\u5f53\u524d\u7279\u5f81\u548c\u4e0d\u540c\u90bb\u5c45 message \u7684\u805a\u5408\u4e3a\u8f93\u5165\uff0c\u7ecf\u8fc7\u53e6\u4e00\u4e2a MLP\uff0c\u5f97\u5230\u8282\u70b9\u66f4\u65b0\u540e\u7684\u7279\u5f81\u3002\u53ef\u4ee5\u770b\u5230\uff0c\u7531\u4e8e\u8981\u7ecf\u8fc7\u4e24\u4e2a\u4e0d\u540c\u7684 MLP\uff0c\u5176\u4e2d\u5305\u542b\u4e86\u82e5\u5e72\u975e\u7ebf\u6027\u53d8\u5316\uff0c\u6700\u7ec8\u7f51\u7edc\u8f93\u51fa\u7684\u8282\u70b9\u7279\u5f81\u66f4\u65b0\u662f\u4e0d\u6ee1\u8db3\u65cb\u8f6c\u4e0d\u53d8\u6027\u6216\u7b49\u53d8\u6027<\/strong>\u7684\u3002<\/span><\/section>\n

            <\/section>\n
            \uff082\uff09\u4e0d\u53d8GNN<\/strong><\/span><\/section>\n

            <\/section>\n
            \u524d\u9762\u6211\u4eec\u63d0\u5230\uff0c\u5728\u6b27\u5f0f\u7a7a\u95f4\u4e2d\uff0c\u957f\u5ea6\u548c\u5939\u89d2\u662f\u4e0d\u4f1a\u56e0\u4e3a\u65cb\u8f6c\u800c\u53d1\u751f\u53d8\u5316\u7684\u3002\u4e0d\u53d8GNN\u5c31\u5229\u7528\u4e86\u8fd9\u4e2a\u51e0\u4f55\u6027\u8d28\uff0c\u628a\u539f\u5b50\u7684\u5750\u6807\u66ff\u6362\u4e3a\u4e86\u539f\u5b50\u952e\u957f\u5ea6\u548c\u5939\u89d2<\/strong>\uff0c\u56e0\u6b64\u65e0\u8bba\u600e\u6837\u53d8\u6362\u5750\u6807\u7cfb\uff0c\u8f93\u5165\u662f\u4e0d\u53d8\u7684\uff0c\u8f93\u51fa\u4e5f\u5c31\u662f\u4e0d\u53d8\u7684\uff0c\u56e0\u6b64\u53ef\u4ee5\u7528\u666e\u901a\u7684\u56fe\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u5904\u7406\u3002<\/span><\/section>\n

            <\/span><\/section>\n
            <\/section>\n
            \u56fe8\uff1a\u4e0d\u53d8GNN\uff0c\u8f93\u5165\u5177\u6709\u65cb\u8f6c\u4e0d\u53d8\u6027<\/span><\/section>\n

            <\/section>\n
            \u4f46\u8fd9\u6837\u7684\u65b9\u6cd5\u4e5f\u5b58\u5728\u4e00\u4e9b\u7f3a\u9677\uff0c\u5b83\u53ea\u80fd\u63cf\u8ff0\u5c40\u90e8\u7684\u51e0\u4f55\u5173\u7cfb\uff0c\u4e22\u5931\u4e86\u5168\u5c40\u5b50\u56fe\u95f4\u7684\u76f8\u5bf9\u4f4d\u7f6e\u5173\u7cfb\u3002\u5c31\u50cf\u6bd5\u52a0\u7d22\u7684\u753b\u4f5c\uff0c\u5404\u4e94\u5b98\u88ab\u6253\u4e71\u53c8\u91cd\u65b0\u7ec4\u5408\u5728\u8138\u4e0a\uff0c\u770b\u8d77\u6765\u5b8c\u5168\u5931\u771f\u4e86\u3002\u56e0\u6b64\u8be5\u7f3a\u9677\u4e5f\u88ab\u79f0\u4e3a\u6bd5\u52a0\u7d22\u95ee\u9898<\/strong><\/span>\uff08Picasso Problem\uff09<\/span>\u3002<\/strong> \u57fa\u4e8e WL-test \u7684\u56fe\u540c\u6784\u68c0\u6d4b\u7b97\u6cd5\u65e0\u6cd5\u68c0\u6d4b\u51fa\u6b64\u7c7b\u533a\u522b\uff0c\u56e0\u6b64\u8fa8\u522b\u80fd\u529b\u4e0e WL-test \u76f8\u5f53\u7684GNN\u7ed3\u6784\u90fd\u4f1a\u9762\u4e34\u540c\u6837\u7684\u95ee\u9898\u3002<\/span><\/section>\n

            <\/span><\/section>\n
            <\/section>\n
            \u56fe9\uff1a\u6bd5\u52a0\u7d22\u95ee\u9898\uff08Picasso Problem\uff09\uff0c\u53d8\u5f62\u7684\u4eba\u8138<\/span><\/section>\n

            <\/span><\/section>\n
            \n
            \n
            \n
            \n
            Johannes Gasteiger, Janek Gro\u00df, Stephan G\u00fcnnemann. Directional Message Passing for Molecular Graphs. ICLR 2020.<\/span><\/section>\n
            \u7b80\u79f0DimNet\uff0c\u5728GNN\u5f15\u5165\u4e86\u5e26\u65b9\u5411\u7684\u6d88\u606f\u4f20\u64ad\u3002<\/span><\/section>\n

            Sch\u00fctt et al. Equivariant message passing for the prediction of tensorial properties and molecular spectra, ICML 2021<\/span><\/section>\n

            <\/section>\n
            Johannes Gasteiger, Florian Becker, Stephan G\u00fcnnemann. GemNet: Universal Directional Graph Neural Networks for Molecules. NeurIPS 2021.<\/span><\/section>\n
            \u7b80\u79f0GemNet\uff0c\u4e00\u79cd\u4e0d\u53d8GNN\uff0c\u8003\u8651\u4e86\u7403\u9762\u5750\u6807\uff0c\u6548\u679c\u503c\u5f97\u4fe1\u8d56\u3002\u7531DimNet\u7684\u4e3b\u8981\u4f5c\u8005\u6253\u9020<\/span>\u3002<\/em><\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n


            <\/p>\n

            \uff083\uff09\u7b49\u53d8 GNN<\/strong><\/span><\/section>\n

            <\/section>\n
            <\/section>\n
            \u56fe10\uff1a\u7b49\u53d8GNN<\/span><\/section>\n

            <\/section>\n
            \u7b49\u53d8 GNN \u5219\u5728\u6bcf\u4e00\u5c42\u7f51\u7edc\u4e2d\u90fd\u4fdd\u7559\u4e86\u8282\u70b9\u7684\u5750\u6807\u4fe1\u606f\uff0c\u80fd\u5f88\u597d\u5730\u89e3\u51b3\u4ee5\u4e0a\u95ee\u9898\u3002\u8ba1\u7b97\u6bcf\u4e00\u5c42\u7684\u8f93\u51fa\u65f6\uff0c\u7b49\u53d8 GNN \u4e5f\u8fdb\u884c\u4e86\u7279\u522b\u8bbe\u8ba1\u3002\u9996\u5148\uff0c\u7528\u76f8\u5bf9\u5750\u6807\u4ee3\u66ff\u7edd\u5bf9\u5750\u6807<\/strong>\uff0c\u6ee1\u8db3\u4e86\u5e73\u79fb\u4e0d\u53d8\u6027<\/strong>\uff1b\u5176\u6b21\uff0c\u5728\u8fdb\u884c message computation \u65f6\uff0c\u4ee5\u5185\u79ef\u4e3a\u8f93\u5165<\/strong>\uff0c\u6ee1\u8db3\u4e86\u65cb\u8f6c\u4e0d\u53d8\u6027<\/strong>\uff0c\u4e8e\u662f\u6700\u7ec8\u7684\u8f93\u51fa\u6ee1\u8db3\u4e86\u65cb\u8f6c\u7b49\u53d8\u6027\u3002<\/span><\/section>\n

            <\/span><\/section>\n
            <\/section>\n
            \u56fe11\uff1a\u7b49\u53d8GNN\u5355\u5c42\u8ba1\u7b97\u8fc7\u7a0b<\/span><\/section>\n

            <\/span><\/section>\n
            \n
            \n
            \n
            \n
            Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, Patrick Riley. Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds. 2018.<\/span><\/section>\n
            \u8fd9\u7bc7\u8bba\u6587\u7b80\u79f0\u5f20\u91cf\u573a\u7f51\u7edcTFN\uff0c\u662f\u6700\u65e9\u540c\u65f6\u6ee1\u8db3\u65cb\u8f6c\u3001\u5e73\u79fb\u7b49\u53d8\u7684GNN\uff0c\u5728\u5206\u5b50\u52a8\u529b\u6a21\u62df\u4e0a\u8fdb\u884c\u9a8c\u8bc1\uff0c\u9886\u57df\u5fc5\u8bfb\u6587\u7ae0\u3002<\/span><\/section>\n

            <\/section>\n
            Victor Garcia Satorras, Emiel Hoogeboom, Max Welling. E(n) Equivariant Graph Neural Networks. ICML 2021.<\/span><\/section>\n
            \u77a9\u76ee\u7684 EGNN\uff0c\u76ee\u524d\u88ab\u5e7f\u6cdb\u4f7f\u7528\u7684\u7b49\u53d8\u56fe\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\uff0c\u9886\u57df\u5fc5\u8bfb\u6587\u7ae0\u3002<\/span><\/section>\n

            <\/section>\n
            Johannes Brandstetter, Rob Hesselink, Elise van der Pol, Erik J Bekkers, Max Welling. Geometric and Physical Quantities Improve E(3) Equivariant Message Passing. ICLR 2022.<\/span><\/section>\n
            \u7b80\u79f0SEGNN\uff0c\u5728EGNN\u57fa\u7840\u4e0a\u5f15\u5165\u4e86 higer-degree \u4e0d\u53ef\u7ea6\u8868\u793a\uff0c\u57fa\u4e8e\u8fd9\u7bc7\u6587\u7ae0\u53ef\u4ee5\u6982\u62ec\u6027\u5b66\u4e60E3\u4e0d\u53d8\u8868\u793a\u76f8\u5173\u77e5\u8bc6\u3002<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n

            <\/section>\n
            \u4ee5\u4e0a\u601d\u60f3\u53ef\u62d3\u5c55\u5230\u591a\u901a\u9053\u8f93\u5165\u7684\u51e0\u4f55\u7279\u5f81\uff0c\u5f97\u5230\u591a\u901a\u9053\u7b49\u53d8 GNN<\/span>\uff08GMN\uff09<\/span>\u3002\u6b64\u5916\uff0c\u8fd8\u6709\u57fa\u4e8e\u7403\u8c10\u51fd\u6570\u3001\u57fa\u4e8e\u7fa4\u8bba\u7b49\u65b9\u5f0f\u8bbe\u8ba1\u7684\u7b49\u53d8GNN\uff0c\u611f\u5174\u8da3\u7684\u8bfb\u8005\u53ef\u4ee5\u81ea\u884c\u67e5\u9605\u3002<\/span><\/section>\n

            <\/section>\n
            \n
            \n
            \n
            Equivariant Graph Mechanics Networks with Constraints, ICLR 2022  <\/span><\/section>\n
            Geometrically Equivariant Graph Neural Networks: A Survey. arXiv:2202.07230.<\/span><\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n


            <\/p>\n

             <\/span><\/section>\n

            \n
            \n
            \n
            \n
            \n

            <\/section>\n

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

            \u51e0\u4f55\u56fe\u795e\u7ecf\u7f51\u7edc\u5728\u5fae\u89c2\u5206\u5b50\u7a7a\u95f4\u7684\u5e94\u7528<\/span><\/strong><\/strong><\/p>\n<\/section>\n

            \n

            <\/section>\n

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

            <\/section>\n