{"id":3605,"date":"2023-09-11T09:37:04","date_gmt":"2023-09-11T01:37:04","guid":{"rendered":""},"modified":"2023-09-11T09:37:04","modified_gmt":"2023-09-11T01:37:04","slug":"","status":"publish","type":"post","link":"http:\/\/www.idea2003.com\/dongtai\/3605.html","title":{"rendered":"\u5b57\u8282\u8df3\u52a8\u674e\u822a\uff1aAI for Science\u7684\u4e00\u4e9b\u63a2\u7d22\u548c\u8fdb\u5c55"},"content":{"rendered":"
\n

\u58f0\u660e:\u672c\u6587\u6765\u81ea\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7 \u673a\u5668\u4e4b\u5fc3\uff08ID:almosthuman2014\uff09\uff0c\u4f5c\u8005:\u673a\u5668\u4e4b\u5fc3\uff0c\u6388\u6743\u5fae\u65b0\u521b\u60f3\u8f6c\u8f7d\u53d1\u5e03\u3002<\/p>\n<\/blockquote>\n

\u8fd1\u5e74\uff0c\u4eba\u5de5\u667a\u80fd\u7684\u5404\u4e2a\u9886\u57df\uff0c\u5305\u62ec\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u8ba1\u7b97\u673a\u89c6\u89c9\u3001\u8bed\u97f3\u5904\u7406\uff0c\u501f\u52a9\u6df1\u5ea6\u5b66\u4e60\u7684\u5f3a\u5927\u5a01\u529b\uff0c\u90fd\u53d6\u5f97\u4e86\u4ee4\u4eba\u53f9\u4e3a\u89c2\u6b62\u7684\u5de8\u5927\u8fdb\u6b65\u3002\u5c06\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\u5e94\u7528\u4e8e\u4f20\u7edf\u7684\u79d1\u5b66\u9886\u57df\uff0c\u5982\u7269\u7406\u3001\u5316\u5b66\u3001\u751f\u7269\u3001\u533b\u5b66\uff0c\u5373\u6240\u8c13\u7684 AI for Science\uff08\u79d1\u5b66\u667a\u80fd\uff09\uff0c\u4f5c\u4e3a\u4e00\u4e2a\u65b0\u7684\u4ea4\u53c9\u5b66\u79d1\uff0c\u4e5f\u9010\u6e10\u5174\u8d77\uff0c\u5b55\u80b2\u7740\u5de8\u5927\u7684\u6f5c\u529b\uff0c\u53d7\u5230\u5e7f\u6cdb\u7684\u5173\u6ce8\u3002<\/p>\n

ByteDance Research \u4e5f\u5728\u8fdb\u884c AI for Science \u7684\u7814\u7a76\uff0c\u5305\u62ec\u673a\u5668\u5b66\u4e60\u4e0e\u91cf\u5b50\u5316\u5b66\u3001\u5927\u89c4\u6a21\u91cf\u5b50\u5316\u5b66\u8ba1\u7b97\u3001AI \u5236\u836f\u7b49\u9886\u57df\u4e00\u4e9b\u95ee\u9898\u7684\u7814\u7a76\uff0c\u5e0c\u671b\u8ddf\u4e1a\u754c\u4e00\u8d77\u63a8\u52a8\u9886\u57df\u7684\u53d1\u5c55\u3002\u672c\u6587\u7b80\u8981\u4ecb\u7ecd\u6211\u4eec\u8fd9\u4e24\u5e74\u6765\u53d6\u5f97\u7684\u4e00\u4e9b\u8fdb\u5c55\u3002\u4e5f\u629b\u7816\u5f15\u7389\uff0c\u5e0c\u671b\u4e0e\u4e1a\u754c\u8fdb\u884c\u66f4\u591a\u7684\u4ea4\u6d41\u548c\u5408\u4f5c\u3002<\/p>\n

\u5728\u673a\u5668\u5b66\u4e60\u548c\u91cf\u5b50\u5316\u5b66\u65b9\u5411\uff0c\u6211\u4eec\u63d0\u51fa\u7684 LapNet \u7b97\u6cd5\uff0c\u6bd4\u6709\u4ee3\u8868\u6027\u7684 FermiNet \u6a21\u578b\u8bad\u7ec3\u901f\u5ea6\u63d0\u9ad8\u4e8610\u500d\uff0c\u80fd\u8ba1\u7b97\u7684\u5316\u5b66\u4f53\u7cfb\u7684\u89c4\u6a21\u548c\u7cbe\u5ea6\u76ee\u524d\u662f\u9886\u57df\u6700\u5927<\/span>\u7684\u3002<\/p>\n

\u5728\u5927\u89c4\u6a21\u91cf\u5b50\u5316\u5b66\u8ba1\u7b97\u65b9\u5411\uff0c\u6211\u4eec\u5f00\u53d1\u4e86 Periodic DMET \u7b97\u6cd5\uff0c\u4f7f\u7528\u7ecf\u5178\u548c\u91cf\u5b50\u6df7\u5408\u8ba1\u7b97\u673a\uff08\u5b9e\u9645\u662f\u5728\u7ecf\u5178\u8ba1\u7b97\u673a\u4e0a\u7684\u6a21\u62df\uff09\uff0c\u7528\u4e8e\u5468\u671f\u6027\u4f53\u7cfb\u7684\u8ba1\u7b97\uff0c\u53ea\u752820\u4e2a\u91cf\u5b50\u6bd4\u7279\u5c31\u8fbe\u5230\u4e86\u4e4b\u524d\u65b9\u6cd5\u7528\u8fd1\u4e07\u4e2a\u91cf\u5b50\u6bd4\u7279\u624d\u80fd\u8fbe\u5230\u7684\u7cbe\u5ea6\u3002<\/p>\n

\u5728 AI \u5236\u836f\u65b9\u5411\uff0c\u6211\u4eec\u5f00\u53d1\u7684 LM-Design \u6a21\u578b\uff0c\u5229\u7528\u5927\u91cf\u86cb\u767d\u8d28\u5e8f\u5217\u6570\u636e\uff0c\u4ee5\u53ca\u4e00\u5b9a\u6570\u91cf\u7684\u86cb\u767d\u8d28\u7ed3\u6784\u548c\u5e8f\u5217\u5bf9\u5e94\u6570\u636e\uff0c\u5b66\u4e60\u4ece\u86cb\u767d\u8d28\u7ed3\u6784\u5230\u5e8f\u5217\u8f6c\u6362\u7684\u6a21\u578b\uff0c\u8fbe\u5230\u4e86\u76ee\u524d\u86cb\u767d\u8d28\u5e8f\u5217\u8bbe\u8ba1\u7684\u6700\u9ad8<\/span>\u7cbe\u5ea6\u3002<\/p>\n

\u673a\u5668\u5b66\u4e60\u4e0e\u91cf\u5b50\u5316\u5b66<\/strong><\/p>\n

\u7269\u7406\u5b66\u5bb6\u72c4\u62c9\u514b\u66fe\u8bf4:\u5bf9\u5927\u90e8\u5206\u7269\u7406\u5b66\u548c\u6574\u4e2a\u5316\u5b66\uff0c\u8fdb\u884c\u6570\u5b66\u5efa\u6a21\u6240\u9700\u8981\u7684\u57fa\u672c\u5b9a\u5f8b\u5df2\u5b8c\u5168\u6e05\u695a\uff0c\u56f0\u96be\u53ea\u5728\u4e8e\u8fd9\u4e9b\u5b9a\u5f8b\u7684\u5e94\u7528\uff0c\u5f97\u5230\u7684\u65b9\u7a0b\u4e00\u822c\u90fd\u592a\u590d\u6742\u800c\u65e0\u6cd5\u6c42\u89e3\u3002<\/p>\n

\u91cf\u5b50\u5316\u5b66\u662f\u6839\u636e\u91cf\u5b50\u529b\u5b66\u7684\u539f\u7406\u7814\u7a76\u5316\u5b66\u73b0\u8c61\u7684\u5b66\u79d1\u3002\u5176\u91cd\u8981\u7684\u95ee\u9898\u662f\u7528\u8ba1\u7b97\u7684\u65b9\u6cd5\u6c42\u89e3\u5206\u5b50\u6216\u5468\u671f\u6027\u4f53\u7cfb\uff08\u5982\u56fa\u4f53\uff09\u7684\u7535\u5b50\u859b\u5b9a\u8c14\u65b9\u7a0b\uff0c\u4ece\u800c\u63a8\u7b97\u51fa\u5206\u5b50\u6216\u5468\u671f\u6027\u4f53\u7cfb\u7684\u57fa\u6001\u80fd\u91cf\u3001\u7535\u6781\u6027\u7b49\u7279\u6027\u3002\u662f\u6240\u8c13\u7684\u4ece\u5934\u8ba1\u7b97(ab initio)\u95ee\u9898\u3002\u4f20\u7edf\u7684\u65b9\u6cd5\u6709\u5bc6\u5ea6\u6cdb\u51fd\u7406\u8bba DFT\u3001\u8026\u5408\u7c07 CCSD \u7b49\u3002\u8981\u4e48\u8ba1\u7b97\u7684\u7cbe\u5ea6\u4e0d\u591f\u9ad8\uff0c\u8981\u4e48\u80fd\u8ba1\u7b97\u7684\u89c4\u6a21\u4e0d\u591f\u5927\u3002<\/p>\n

\u8fd1\u5e74\uff0c\u7528\u673a\u5668\u5b66\u4e60\u7684\u65b9\u6cd5\u89e3\u51b3\u4ece\u5934\u8ba1\u7b97\u95ee\u9898\u6210\u4e3a\u500d\u53d7\u5173\u6ce8\u7684\u65b0\u65b9\u5411\u3002\u5176\u57fa\u672c\u60f3\u6cd5\u662f\u501f\u52a9\u6df1\u5ea6\u5b66\u4e60\u5f3a\u5927\u7684\u8868\u793a\u548c\u5b66\u4e60\u80fd\u529b\uff0c\u5927\u5e45\u63d0\u5347\u4ece\u5934\u8ba1\u7b97\u7684\u7cbe\u5ea6\u548c\u89c4\u6a21\u3002\u5176\u4e2d\u7684\u4e00\u4e2a\u8def\u5f84\u662f NN-VMC\uff08Neural Network based Variational Monte Carlo\uff09 \u3002\u7528\u795e\u7ecf\u7f51\u7edc\u8fd1\u4f3c\u859b\u5b9a\u8c14\u65b9\u7a0b\u7684\u6ce2\u51fd\u6570\uff0c\u901a\u8fc7\u968f\u673a\u91c7\u6837\u7684\u65b9\u5f0f\u83b7\u5f97\u4f53\u7cfb\u4e2d\u7535\u5b50\u5728\u7a7a\u95f4\u4e2d\u7684\u6837\u672c\uff0c\u8fd9\u6837\u53ef\u4ee5\u8ba1\u7b97\u57fa\u4e8e\u859b\u5b9a\u8c14\u65b9\u7a0b\u7684\u6574\u4e2a\u4f53\u7cfb\u7684\u80fd\u91cf\u3002\u901a\u8fc7\u6700\u5c0f\u5316\u80fd\u91cf\u7684\u4e0a\u754c\uff0c\u4f18\u5316\u795e\u7ecf\u7f51\u7edc\u7684\u53c2\u6570\uff0c\u4e0d\u65ad\u8fed\u4ee3\uff0c\u6700\u540e<\/span>\u5f97\u5230\u8fd1\u4f3c\u6700\u4f18<\/span>\u7684\u795e\u7ecf\u7f51\u7edc(\u6ce2\u51fd\u6570)\uff0c\u4ee5\u53ca\u4f53\u7cfb\u7684\u8fd1\u4f3c\u57fa\u6001\u80fd\u91cf(\u6700\u5c0f\u80fd\u91cf)\u3002(\u6ce8:\u6ce2\u51fd\u6570\u7684\u5e73\u65b9\u662f\u7535\u5b50\u5728\u7a7a\u95f4\u51fa\u73b0\u7684\u6982\u7387\u5bc6\u5ea6\u51fd\u6570\uff0c\u6709\u4e86\u6ce2\u51fd\u6570\uff0c\u5c31\u53ef\u4ee5\u8fdb\u884c\u7535\u5b50\u5728\u7a7a\u95f4\u4e2d\u7684\u968f\u673a\u91c7\u6837\u3002)\u56fe1\u663e\u793a NN-VMC \u7684\u57fa\u672c\u539f\u7406\u3002\u5176\u6838\u5fc3\u95ee\u9898\u662f\u5982\u4f55\u8bbe\u8ba1\u795e\u7ecf\u7f51\u7edc\u548c\u5b66\u4e60\u7b97\u6cd5\u3002<\/p>\n

\n

\u56fe1. NN-VMC \u65b9\u6cd5\u7684\u57fa\u672c\u539f\u7406<\/p>\n

NN-VMC \u4e2d\u6709\u4ee3\u8868\u6027\u7684\u65b9\u6cd5\u662f DeepMind \u548c ICL \u4e8e2019\u5e74\u63d0\u51fa\u7684 FermiNet\u3002\u4e4b\u540e\u4e00\u4e9b\u7814\u7a76\u673a\u6784\u53c8\u63d0\u51fa\u4e86\u4e00\u4e9b\u65b0\u7684\u65b9\u6cd5\u3002ByteDance Research \u4ece2021\u5e74\u8d77\uff0c\u4e0e\u5317\u4eac\u5927\u5b66\u5408\u4f5c\uff0c\u8fdb\u884c\u4e86\u4e00\u7cfb\u5217\u76f8\u5173\u7814\u7a76\uff0c\u63d0\u51fa\u4e86\u51e0\u4e2a\u65b0\u7684\u65b9\u6cd5\u3002\u4e0b\u9762\u5bf9\u8fd9\u4e9b\u65b9\u6cd5\u505a\u4e00\u7b80\u5355\u4ecb\u7ecd\u3002<\/p>\n

NN-VMC+ECP\uff0c\u662f\u6211\u4eec\u5f00\u53d1\u7684\u7ed3\u5408 NN-VMC \u548c\u8d5d\u52bf ECP\uff08Effective Core Potential\uff09\u7684\u65b9\u6cd5 [1]\uff0c\u53ef\u4ee5\u8fdb\u4e00\u6b65\u63d0\u9ad8\u8ba1\u7b97\u7684\u6548\u7387\u548c\u4f53\u7cfb\u7684\u89c4\u6a21\u3002\u8ba1\u7b97\u5316\u5b66\u4f53\u7cfb\u7684\u7279\u6027\u65f6\uff0c\u5f80\u5f80\u53ea\u9700\u8981\u5173\u6ce8\u539f\u5b50\u4e2d\u5916\u4fa7\u8f68\u9053\u7684\u7535\u5b50\u3002\u5c06\u539f\u5b50\u4e2d\u5185\u4fa7\u8f68\u9053\u7684\u7535\u5b50\u7684\u52bf\u80fd\u7528\u5b9a\u91cf\u8868\u793a\uff0c\u5c31\u53ef\u4ee5\u5927\u5e45\u51cf\u5c11\u6240\u9700\u8981\u7684\u8ba1\u7b97\u91cf\u3002\u6211\u4eec\u5c06 ECP \u6280\u5de7\u5e94\u7528\u4e8e NN-VMC\uff0c\u5f97\u5230\u4e86\u8fd9\u4e2a\u65b0\u65b9\u6cd5\uff0c\u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6548\u679c\u3002<\/p>\n

NN-DMC\uff0c\u662f\u6211\u4eec\u63d0\u51fa\u7684\u5c06\u795e\u7ecf\u7f51\u7edc\u548c\u6269\u6563\u8499\u7279\u5361\u6d1b\u6cd5 DMC\uff08Diffusoon Monte Carlo\uff09\u7ed3\u5408\u7684\u53e6\u4e00\u4e2a\u65b9\u6cd5 [2]\u3002DMC \u4e0e VMC \u4e0d\u540c\uff0c\u4e0d\u662f\u8ba1\u7b97\u4f53\u7cfb\u57fa\u6001\u80fd\u91cf\u7684\u4e0a\u754c\uff0c\u800c\u662f\u4f7f\u7528\u865a\u65f6\u6f14\u5316\u6765\u8ba1\u7b97\u4f53\u7cfb\u7684\u57fa\u6001\u80fd\u91cf\u3002\u8fd9\u4e2a\u65b9\u6cd5\uff0c\u76f8\u6bd4 FermiNet \u7b49\u5df2\u6709\u65b9\u6cd5\u4e5f\u80fd\u5927\u5e45\u63d0\u9ad8\u8ba1\u7b97\u7684\u7cbe\u5ea6\u548c\u89c4\u6a21\u3002<\/p>\n

\u6700\u8fd1\u5f00\u53d1\u7684 LapNet \u4e5f\u662f\u4e00\u79cd NN-VMC \u65b9\u6cd5 [3]\uff0c\u7279\u70b9\u662f\u5728\u795e\u7ecf\u7f51\u7edc\u5b66\u4e60\u65f6\u4f7f\u7528\u524d\u5411\u62c9\u666e\u62c9\u65af\u7b97\u5b50\uff08 Forward Laplacian\uff09\u3002\u57fa\u4e8e\u859b\u5b9a\u8c14\u65b9\u7a0b\u8ba1\u7b97\u4f53\u7cfb\u7684\u80fd\u91cf\u4e0a\u754c\u7684\u8fc7\u7a0b\u4e2d\uff0c\u9700\u8981\u8ba1\u7b97\u54c8\u5bc6\u987f\u7b97\u5b50\uff0c\u5305\u62ec\u5176\u4e2d\u7684\u52a8\u80fd\u90e8\u5206\u3002\u4e4b\u524d\u7684\u65b9\u6cd5\u90fd\u662f\u901a\u8fc7\u8ba1\u7b97\u76f8\u5173\u7684\u9ed1\u585e\u77e9\u9635\u7684\u65b9\u5f0f\u8ba1\u7b97\u52a8\u80fd\uff0c\u5176\u7b97\u6cd5\u590d\u6742\u5ea6\u9ad8\uff0c\u6210\u4e3a\u5b66\u4e60\u7684\u4e00\u4e2a\u74f6\u9888\u3002LapNet \u5728\u5b66\u4e60\u7684\u524d\u5411\u4f20\u64ad\u4e2d\uff0c\u901a\u8fc7\u62c9\u666e\u62c9\u65af\u7b97\u5b50\u7684\u8ba1\u7b97\uff0c\u76f4\u63a5\u8ba1\u7b97\u52a8\u80fd\uff0c\u4ee5\u53ca\u54c8\u5bc6\u987f\u7b97\u5b50\uff0c\u4ece\u800c\u7701\u53bb\u4e86\u9ed1\u585e\u77e9\u9635\u7684\u8ba1\u7b97\u3002\u8fd9\u6837\u53ef\u4ee5\u5927\u5e45\u63d0\u9ad8\u5b66\u4e60\u7684\u8ba1\u7b97\u6548\u7387\u3002\u76f8\u6bd4 FermiNet\uff0cLapNet \u6709\u5e73\u574710\u500d\u5de6\u53f3\u7684\u52a0\u901f\u3002<\/p>\n

ECP\u3001DMC \u548c Forward Laplace \u5c5e\u4e8e\u4e09\u79cd\u4e0d\u540c\u7684\u6280\u672f\u6539\u8fdb\uff08\u7b80\u5316\u52bf\u80fd\u8ba1\u7b97\u3001\u4f18\u5316\u91c7\u6837\uff0c\u63d0\u9ad8\u8ba1\u7b97\u6548\u7387\uff09\uff0c\u4e09\u4e2a\u6280\u672f\u7ed3\u5408\u8d77\u6765\u539f\u7406\u4e0a\u53ef\u4ee5\u66f4\u5927\u7a0b\u5ea6\u4e0a\u63d0\u9ad8\u8ba1\u7b97\u89c4\u6a21\uff0c\u4e5f\u662f\u6211\u4eec\u6b63\u5728\u5c1d\u8bd5\u7684\u65b9\u6cd5\u3002\u53e6\u5916\uff0c\u6211\u4eec\u8fd8\u5c06 NN-VMC \u65b9\u6cd5\u5e94\u7528\u4e8e\u56fa\u4f53\u7684\u859b\u5b9a\u8c14\u65b9\u7a0b\u6c42\u89e3 [4]\uff0c\u5206\u5b50\u4f53\u7cfb\u7684\u529b\u573a [5]\u3001\u7535\u6781\u5316\u8ba1\u7b97 [6] \u7b49\u95ee\u9898\uff0c\u8bc1\u660e\u4e86 NN-VMC \u65b9\u6cd5\u7684\u5b9e\u7528\u6027\u3002<\/p>\n

\u56fe2\u663e\u793a\u76ee\u524d NN-VMC \u65b9\u6cd5\u4e2d\u4ee3\u8868\u6027\u5de5\u4f5c\u7684\u7cbe\u5ea6\u548c\u89c4\u6a21\uff0c\u7eb5\u8f74\u8868\u793a\u7cbe\u5ea6\uff0c\u5706\u7684\u5927\u5c0f\u8868\u793a\u89c4\u6a21\u3002\u6211\u4eec\u63d0\u51fa\u7684 LapNet \u65b9\u6cd5\u80fd\u591f\u4ee5\u66f4\u9ad8\u7684\u7cbe\u5ea6\u8ba1\u7b97\u66f4\u5927\u7684\u4f53\u7cfb\u3002\u6700\u5927<\/span>\u7684\u4f53\u7cfb\u6709116\u4e2a\u7535\u5b50\u3002<\/p>\n

\n

\u56fe2. NN-VMC \u65b9\u6cd5\u7684\u89c4\u6a21\u548c\u7cbe\u5ea6<\/p>\n

\u5927\u89c4\u6a21\u91cf\u5b50\u5316\u5b66\u8ba1\u7b97<\/strong><\/p>\n

\u901a\u8fc7\u76f4\u63a5\u6c42\u89e3\u859b\u5b9a\u8c14\u65b9\u7a0b\u8ba1\u7b97\u5316\u5b66\u4f53\u7cfb\u7279\u6027\uff08\u6bd4\u5982\u57fa\u6001\u80fd\u91cf\uff09\u7684\u65b9\u6cd5\u80fd\u5904\u7406\u7684\u89c4\u6a21\u4ecd\u7136\u6709\u9650\u3002\u91cf\u5b50\u5d4c\u5165\u65b9\u6cd5(Quantum Embedding Method)\u88ab\u8ba4\u4e3a\u662f\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u7684\u4e00\u6761\u6709\u6548\u8def\u5f84\u3002\u57fa\u672c\u60f3\u6cd5\u662f\u901a\u8fc7\u5206\u800c\u6cbb\u4e4b\u548c\u591a\u7cbe\u5ea6\u8ba1\u7b97\u5b9e\u73b0\u5927\u89c4\u6a21\u5316\u3002\u4ee3\u8868\u6027\u7684\u65b9\u6cd5\u6709\u5bc6\u5ea6\u77e9\u9635\u5d4c\u5165\u7406\u8bba DMET(Density Matrix Embedding Theory)\u3002\u5c06\u4f53\u7cfb\u5212\u5206\u4e3a\u82e5\u5e72\u90e8\u5206(Fragment)\uff0c\u5bf9\u5176\u4e2d\u7684\u6bcf\u4e2a Fragment \u53ca\u5176\u5bf9\u5e94\u7684\u73af\u5883(Bath)\u8fdb\u884c\u9ad8\u7cbe\u5ea6\u8ba1\u7b97\uff0c\u5bf9\u5176\u4ed6\u90e8\u5206\u8fdb\u884c\u4f4e\u7cbe\u5ea6\u8ba1\u7b97\u3002\u800c\u4e14\u6839\u636e\u9700\u8981\u5bf9\u6bcf\u4e2a Fragment \u8fdb\u884c\u5e76\u884c\u5904\u7406\u3002\u6700\u540e<\/span>\u518d\u628a\u9ad8\u7cbe\u5ea6\u8ba1\u7b97\u7684\u7ed3\u679c\u5408\u5e76\u8d77\u6765\uff0c\u4e0d\u65ad\u8fed\u4ee3\u903c\u8fd1\u539f\u59cb\u4f53\u7cfb\u3002\u8fd9\u6837\u53ef\u4ee5\u5927\u5e45\u63d0\u9ad8\u53ef\u8ba1\u7b97\u7684\u4f53\u7cfb\u7684\u89c4\u6a21\u3002<\/p>\n

\n

\u56fe3DMET \u65b9\u6cd5\u7684\u76f4\u89c2\u89e3\u91ca<\/p>\n

\u56fe3\u793a\u610f DMET \u65b9\u6cd5\u7684\u8fc7\u7a0b\u3002\u9996\u5148\u5bf9\u539f\u59cb\u4f53\u7cfb\u8fdb\u884c\u5212\u5206\uff0c\u5f97\u5230\u4e00\u7ec4 Fragment\u3002\u5047\u8bbe\u56fe\u4e2d\u9ec4\u8272\u90e8\u5206\u662f\u6211\u4eec\u5173\u6ce8\u7684 Fragment\uff0c\u4f8b\u5982\u662f\u4e24\u4e2a\u539f\u5b50\u3002\u56fe\u4e2d\u84dd\u8272\u7684\u90e8\u5206\u5305\u542b\u73af\u5883\u548c\u5176\u4ed6\u90e8\u5206\u3002\u5bf9\u5173\u6ce8\u7684 Fragment \u53ca\u5176\u73af\u5883\u7528\u9ad8\u7cbe\u5ea6\u7684\u65b9\u6cd5\u8ba1\u7b97\uff0c\u4f8b\u5982 CCSD\uff0c\u5bf9\u5176\u4ed6\u90e8\u5206\u7528\u4f4e\u7cbe\u5ea6\u7684\u65b9\u6cd5\u8ba1\u7b97\uff0c\u4f8b\u5982\uff0cHartree\u2013Fock \u6cd5\u3002\u5bf9\u6240\u6709\u7684 Fragment \u505a\u540c\u6837\u7684\u5e76\u884c\u5904\u7406\u3002<\/p>\n

\u5177\u4f53\u7b97\u6cd5\u5982\u4e0b\u3002\u9996\u5148\uff0c\u901a\u8fc7\u4f4e\u7cbe\u5ea6\u6c42\u89e3\uff0c\u5f97\u5230\u6574\u4f53\uff08\u5173\u6ce8\u7684 Fragment\u3001\u73af\u5883\u548c\u5176\u4ed6\u90e8\u5206\uff09\u7684\u7ea6\u5316\u5bc6\u5ea6\u77e9\u9635\uff0c\u8fd9\u4e2a\u4f4e\u7cbe\u5ea6\u89e3\u5305\u542b\u53c2\u6570\u3002\u5176\u6b21\uff0c\u5bf9\u8fd9\u4e2a\u77e9\u9635\u7684 Fragment \u53ca\u5176\u73af\u5883\u8fdb\u884c\u5947\u5f02\u503c\u5206\u89e3\uff0c\u6784\u5efa\u6295\u5f71\u7b97\u7b26 P(\u8be5\u6295\u5f71\u7b97\u7b26\u4ec5\u5173\u6ce8 Fragment \u53ca\u5176\u73af\u5883)\uff0c\u5229\u7528\u6295\u5f71\u7b97\u7b26\u6784\u5efa\u4f4e\u7ef4\u5ea6\u7684\u4f53\u7cfb(\u56fe\u7247)\uff0c\u5e76\u5bf9\u5176\u8fdb\u884c\u9ad8\u7cbe\u5ea6\u6c42\u89e3\u3002\u4e4b\u540e\uff0c\u5c06\u6240\u6709 Fragment \u7684\u8ba1\u7b97\u7ed3\u679c\u5408\u5e76\uff0c\u4f5c\u4e3a\u4f53\u7cfb\u6574\u4f53\u7684\u8fd1\u4f3c\u3002\u6700\u540e<\/span>\uff0c\u901a\u8fc7\u8fed\u4ee3\uff0c\u8c03\u8282\u53c2\u6570\uff0c\u4f7f\u5f97\u539f\u59cb\u7684\u4f4e\u7cbe\u5ea6\u89e3\u9010\u6e10\u903c\u8fd1\u5408\u5e76\u7684\u9ad8\u7cbe\u5ea6\u89e3(\u5728 L2norm \u7684\u610f\u4e49\u4e0b)\uff0c\u76f4\u5230\u5f97\u5230\u6700\u7ec8\u7ed3\u679c\u3002<\/p>\n

\u6211\u4eec\u57fa\u4e8e\u4e24\u79cd\u5b8c\u5168\u4e0d\u540c\u7684\u8ba1\u7b97\u8303\u5f0f\uff0c\u5b9e\u73b0 DMET \u53ca\u5176\u53d8\u79cd\u7684 SIE\uff0c\u8fdb\u884c\u5927\u89c4\u6a21\u91cf\u5b50\u5316\u5b66\u4f53\u7cfb\u7684\u8ba1\u7b97\u3002\u4e00\u662f\u4f7f\u7528\u7ecf\u5178\u8ba1\u7b97\u673a\uff0c\u53e6\u4e00\u4e2a\u662f\u4f7f\u7528\u91cf\u5b50\u8ba1\u7b97\u673a\u3002\u672c\u6587\u4e3b\u8981\u4ecb\u7ecd\u540e\u8005\u7684\u76f8\u5173\u5de5\u4f5c\uff08\u524d\u8005\u7684\u5de5\u4f5c\u8ba1\u5212\u4eca\u540e\u6709\u66f4\u5927\u8fdb\u5c55\u65f6\u4ecb\u7ecd\uff09\uff0c\u4e5f\u79f0\u4e3a\u91cf\u5b50\u8ba1\u7b97\u5316\u5b66\u3002\u6211\u4eec\u8003\u8651\u5728\u91cf\u5b50\u8ba1\u7b97\u673a\u4e0a\u7684\u5b9e\u73b0\uff0c\u4f46\u53ea\u5728\u7ecf\u5178\u8ba1\u7b97\u673a\u4e0a\u8fdb\u884c\u6a21\u62df\uff0c\u5e0c\u671b\u4e3a\u91cf\u5b50\u5316\u5b66\u7684\u53d1\u5c55\u505a\u51fa\u8d21\u732e\u3002\u65e0\u8bba\u662f\u54ea\u79cd\u8ba1\u7b97\u8303\u5f0f\uff0cDMET \u65b9\u6cd5\u4f7f\u5f97\u5927\u89c4\u6a21\u8ba1\u7b97\u6210\u4e3a\u53ef\u80fd\uff0c\u6211\u4eec\u6b63\u5728\u5c1d\u8bd5\u52aa\u529b\u5b9e\u73b0 DMET\uff0c\u5c06\u53ef\u8ba1\u7b97\u7684\u4f53\u7cfb\u63d0\u5347\u51e0\u4e2a\u6570\u91cf\u7ea7\u3002<\/p>\n

\u7269\u7406\u5b66\u5bb6\u8d39\u66fc\u66fe\u8bf4:\u81ea\u7136\u4e0d\u662f\u53e4\u5178\u529b\u5b66\uff0c\u5982\u679c\u4f60\u60f3\u6a21\u62df\u81ea\u7136\uff0c\u4f60\u6700\u597d<\/span>\u7528\u91cf\u5b50\u529b\u5b66\u3002\u5f00\u53d1\u91cf\u5b50\u8ba1\u7b97\u6280\u672f\u7684\u9a71\u52a8\u529b\u5c31\u662f\u7528\u91cf\u5b50\u5c42\u9762\u7684\u8ba1\u7b97\u8bbe\u5907\u6a21\u62df\u91cf\u5b50\u73b0\u8c61\u3002\u6362\u8a00\u4e4b\uff0c\u91cf\u5b50\u5316\u5b66\u662f\u91cf\u5b50\u8ba1\u7b97\u6700\u5408\u9002\u7684\u5e94\u7528\u9886\u57df\u4e4b\u4e00\u3002<\/p>\n

\n

\u56fe4. Periodic DMET \u65b9\u6cd5\u7684\u793a\u610f<\/p>\n

\u6211\u4eec\u5148\u540e\u5f00\u53d1\u4e86\u4e24\u4e2a\u91cf\u5b50\u8ba1\u7b97\u5316\u5b66\u65b9\u6cd5\uff0c\u7ed3\u5408\u91cf\u5b50\u548c\u7ecf\u5178\u6df7\u5408\u8ba1\u7b97\u673a\u548c DMET \u7684\u7279\u70b9\uff0c\u5927\u5e45\u63d0\u9ad8\u4e86\u8ba1\u7b97\u4f53\u7cfb\u7684\u7cbe\u5ea6\u548c\u89c4\u6a21\u3002\u57fa\u672c\u60f3\u6cd5\u662f\u7528\u91cf\u5b50\u8ba1\u7b97\u673a\u5b9e\u73b0 DMET \u9ad8\u7cbe\u5ea6\u8ba1\u7b97\u7684\u90e8\u5206\uff0c\u7528\u7ecf\u5178\u8ba1\u7b97\u673a\u5b9e\u73b0 DMET \u4f4e\u7cbe\u5ea6\u8ba1\u7b97\u7684\u90e8\u5206\u3002DMET-ESVQE \u8ba1\u7b97\u5206\u5b50\u4f53\u7cfb [7]\uff0c Periodic DMET \u8ba1\u7b97\u5468\u671f\u6027\u4f53\u7cfb [8]\u3002\u524d\u8005\u53ea\u752816\u4e2a\u91cf\u5b50\u6bd4\u7279\uff0c\u5c31\u80fd\u5b9e\u73b0\u4e4b\u524d\u7684\u65b9\u6cd5\u7528144\u4e2a\u91cf\u5b50\u6bd4\u7279\u7684\u8ba1\u7b97\u3002\u540e\u8005\u53ea\u752820\u4e2a\u91cf\u5b50\u6bd4\u7279\u5c31\u80fd\u5b9e\u73b0\u4e4b\u524d\u7684\u65b9\u6cd5\u7528\u5c06\u8fd11\u4e07\u4e2a\u91cf\u5b50\u6bd4\u7279\u7684\u8ba1\u7b97\u3002<\/p>\n

\u56fe4\u663e\u793a\u5728\u6df7\u5408\u8ba1\u7b97\u673a\u4e0a\u5b9e\u73b0\u7684 DMET Periodic \u65b9\u6cd5\u3002\u8f93\u5165\u662f\u6676\u4f53\uff0c\u8f93\u51fa\u662f\u4f53\u7cfb\u7684\u80fd\u91cf\u3002\u9996\u5148\u5bf9\u4f53\u7cfb\u8fdb\u884c\u5212\u5206\uff0c\u7136\u540e\u5e76\u884c\u5730\u8ba1\u7b97\u6bcf\u4e00\u4e2a Fragment\u3002\u5bf9\u5173\u6ce8 Fragment \u53ca\u5176\u73af\u5883\u5728\u91cf\u5b50\u8ba1\u7b97\u673a\u4e0a\u7528 U-CCSD \u6c42\u89e3\u3002\u5bf9\u5176\u4ed6\u90e8\u5206\u5728\u7ecf\u5178\u8ba1\u7b97\u673a\u4e0a\u7528 Hartree\u2013Fock \u6cd5\u6c42\u89e3\u3002<\/p>\n

\u91cf\u5b50\u8499\u7279\u5361\u7f57\u6cd5\uff0c\u5305\u62ec VMC\u3001DMC\uff0c\u662f\u91cf\u5b50\u5316\u5b66\u7684\u6700\u6709\u6548\u7684\u4e00\u7cfb\u5217\u7b97\u6cd5 [9]\u3002\u6211\u4eec\u8fd8\u63d0\u51fa\u4e86\u5c06\u91cf\u5b50\u8ba1\u7b97\u548c\u91cf\u5b50\u8499\u7279\u5361\u7f57\u6cd5\u7ed3\u5408\u7684\u65b0\u65b9\u6cd5\u3002\u8fd9\u4e2a\u65b9\u6cd5\u53ef\u4ee5\u4f53\u73b0\u91cf\u5b50\u8ba1\u7b97\u5bf9\u91cf\u5b50\u5316\u5b66\u7684\u4e00\u4e9b\u4f18\u52bf\u3002\u5177\u4f53\u5730\uff0c\u91cf\u5b50\u8ba1\u7b97\u53ef\u4ee5\u90e8\u5206\u89e3\u51b3\u91cf\u5b50\u8499\u7279\u5361\u6d1b\u6cd5\u4e2d\u7684\u7b26\u53f7\u95ee\u9898\u3002<\/p>\n

AI \u5236\u836f<\/strong><\/p>\n

\u4f7f\u7528 AI \u6280\u672f\u8f85\u52a9\u836f\u7269\u53d1\u73b0\u5df2\u7ecf\u6210\u4e3a\u88ab\u4e1a\u754c\u5e7f\u4e3a\u63a5\u53d7\u7684\u65b0\u8303\u5f0f\u3002\u8fd1\u5e74\u6709\u5927\u91cf\u7684\u7814\u7a76\uff0c\u4e5f\u6709\u4e00\u4e9b\u6280\u672f\u5e94\u7528\u4e8e\u5b9e\u9645\u7684\u573a\u666f\u3002\u6211\u4eec\u8fdb\u884c\u4e86\u57fa\u4e8e AI \u6280\u672f\u7684\u836f\u7269\u8bbe\u8ba1\u7684\u7814\u7a76\u548c\u5f00\u53d1\uff0c\u5305\u62ec\u5c0f\u5206\u5b50\u836f\u7269\u548c\u5927\u5206\u5b50\u836f\u7269\uff08\u6297\u4f53\u836f\u7269\uff09\u3002<\/p>\n

\u5c0f\u5206\u5b50\u836f\u7269\u8bbe\u8ba1\u8fc7\u7a0b\u5305\u62ec\u86cb\u767d\u8d28\u9776\u70b9\u7684\u53d1\u73b0\uff0c\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u7684\u751f\u6210\uff0c\u5019\u9009\u4e0e\u9776\u70b9\uff08target\uff09\u7684\u4eb2\u548c\u6027\u3001\u5019\u9009\u7684\u6210\u836f\u6027\u3001\u65e0\u6bd2\u6027\u7b49\u7684\u5224\u65ad\u3002\u76ee\u524d\u6709 AI \u6280\u672f\u53ef\u4ee5\u5b9e\u73b0\u8fd9\u4e9b\u836f\u7269\u7684\u5f00\u53d1\u6b65\u9aa4\u3002\u6211\u4eec\u5f00\u53d1\u4e86\u57fa\u4e8e\u673a\u5668\u5b66\u4e60\u7684\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u751f\u6210\u7684\u65b9\u6cd5\uff0cMARS \u662f\u5229\u7528\u6253\u5206\u51fd\u6570\u81ea\u52a8\u751f\u6210\u5019\u9009\u7684\u65b9\u6cd5 [10]\uff0cDESERT \u662f\u6839\u636e\u9776\u70b9\u7684\u5f62\u72b6\u81ea\u52a8\u751f\u6210\u5019\u9009\u7684\u65b9\u6cd5 [11]\u3002<\/p>\n

MARS \u4ece\u79cd\u5b50\u5206\u5b50\u5f00\u59cb\uff0c\u4e0d\u65ad\u7f16\u8f91\u5206\u5b50\uff0c\u76f4\u5230\u6700\u540e<\/span>\u5f97\u5230\u6700\u4f18<\/span>\u7684\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u3002\u751f\u6210\u7684\u8fc7\u7a0b\u4e2d\u4f7f\u7528\u9a6c\u5c14\u53ef\u592b\u94fe\u8499\u7279\u5361\u6d1b\u6cd5\uff08MCMC\uff09\uff0c\u5176\u5e73\u7a33\u5206\u5e03\u662f\u7531\u591a\u4e2a\u6253\u5206\u51fd\u6570\u6784\u6210\u7684\u6982\u7387\u5206\u5e03\u3002\u6253\u5206\u51fd\u6570\u8868\u793a\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u7684\u4eb2\u548c\u6027\u3001\u6210\u836f\u6027\u3001\u65e0\u6bd2\u6027\u7b49\u3002\u5efa\u8bae\u5206\u5e03\u8868\u793a\u57fa\u4e8e\u56fe\u795e\u7ecf\u7f51\u7edc(MPNN)\u8fdb\u884c\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u7684\u7f16\u8f91\u524d\u540e\u7684\u6761\u4ef6\u6982\u7387\u5206\u5e03\u3002\u56fe\u795e\u7ecf\u7f51\u7edc\u8868\u793a\u5c0f\u5206\u5b50\u5316\u5408\u7269\u7684\u5206\u5b50\u5f0f\uff0c\u7ed3\u70b9\u662f\u539f\u5b50\uff0c\u8fb9\u662f\u5316\u5b66\u952e\u3002\u5bf9\u5c0f\u5206\u5b50\u7684\u7f16\u8f91\u5305\u62ec\u589e\u52a0\u65b0\u7684\u7ed3\u70b9\u548c\u5220\u9664\u5df2\u6709\u7684\u7ed3\u70b9\u3002\u56fe\u795e\u7ecf\u7f51\u7edc\u4e0a\u53ef\u4ee5\u9884\u6d4b\u5bf9\u5c0f\u5206\u5b50\u7684\u53ef\u80fd\u7684\u7f16\u8f91\u64cd\u4f5c(\u589e\u52a0\u6216\u5220\u9664)\uff0c\u5176\u53c2\u6570\u901a\u8fc7\u5b66\u4e60\u5f97\u5230\u3002MARS \u53ea\u9700\u8981\u6709\u6253\u5206\u51fd\u6570\u3001\u5c0f\u5206\u5b50\u836f\u7269\u7684\u6570\u636e\u5e93(\u5206\u5b50\u5f0f)\u5c31\u53ef\u4ee5\u751f\u6210\u5168\u65b0\u7684\u548c\u591a\u6837\u7684\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u3002\u76ee\u524d MARS \u5df2\u7ecf\u88ab\u7528\u4e8e\u5b9e\u9645\u7684\u5c0f\u5206\u5b50\u836f\u7269\u8bbe\u8ba1\u5de5\u4f5c\u4e2d\u3002<\/p>\n

DESERT \u901a\u8fc7\u4e24\u4e2a\u6b65\u9aa4\u8fdb\u884c\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u7684\u751f\u6210\u3002Sketching:\u91c7\u6837\u4e0e\u86cb\u767d\u8d28\u9776\u70b9\u7684\u53e3\u888b\uff08pocket\uff09\u5f62\u72b6\u4e92\u8865\u7684\u836f\u7269\u5019\u9009\u7684\u5f62\u72b6\uff0cGenerating:\u57fa\u4e8e\u836f\u7269\u5019\u9009\u7684\u5f62\u72b6\u81ea\u52a8\u751f\u6210\u836f\u7269\u5019\u9009\u7684\u5206\u5b50\u5f0f\u3002\u56fe5\u663e\u793a\u8fd9\u4e2a\u8fc7\u7a0b\u3002<\/p>\n

\n

\u56fe5. DESERT:\u81ea\u52a8\u751f\u6210\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009<\/p>\n

\u5019\u9009\u548c\u9776\u70b9\u7ed3\u5408\uff0c\u5176\u5fc5\u8981\u6761\u4ef6\u662f\u4e24\u8005\u7684\u5f62\u72b6\u80fd\u591f\u8fdb\u884c\u5f88\u597d\u7684\u5bf9\u63a5\u3002\u5728 Sketching \u9636\u6bb5\uff0c\u6839\u636e\u86cb\u767d\u8d28\u9776\u70b9\u7684\u5f62\u72b6\uff0c\u7528\u542f\u53d1\u5f0f\u7684\u65b9\u6cd5\u4ea7\u751f\u5019\u9009\u7684\u5f62\u72b6\u3002\u5728 Generating \u9636\u6bb5\uff0c\u7528\u4e8b\u5148\u5b66\u597d\u7684\u5f62\u72b6\u5230\u5206\u5b50\u7684\u751f\u6210\u6a21\u578b Shape2Mol \u6839\u636e\u5f62\u72b6\u81ea\u52a8\u751f\u6210\u5206\u5b50\u5f0f\u3002\u53ef\u4ee5\u5229\u7528\u5206\u5b50\u5e93\u91cc\u7684\u5927\u91cf\u7684\u836f\u7269\u7684\u5206\u5b50\u5f0f\u548c\u5f62\u72b6\uff0c\u5b66\u4e60\u8fd9\u4e2a\u751f\u6210\u6a21\u578b\u3002\u5982\u56fe6\u6240\u793a\uff0cShape2Mol \u4e2d\uff0c\u7f16\u7801\u5668\u5c06\u5206\u5b503D \u5f62\u72b6\u8fdb\u884c\u7f16\u7801\uff0c\u4ea7\u751f\u4e2d\u95f4\u8868\u793a\uff0c\u89e3\u7801\u5668\u6839\u636e\u4e2d\u95f4\u8868\u793a\u751f\u6210\u5206\u5b50\u5f0f\u30023D \u5f62\u72b6\u4f7f\u7528\u4f53\u7d20\u8868\u793a\uff0c\u5206\u5b50\u5f0f\u4f7f\u7528\u7b26\u53f7\u5e8f\u5217\u8868\u793a\u3002DESERT \u662f2022\u5e74\u5c0f\u5206\u5b50\u836f\u7269\u5019\u9009\u751f\u6210\u7ed3\u5408\u6027\u80fd\u6700\u597d<\/span>\u7684\u65b9\u6cd5\u3002<\/p>\n

\n

\u56fe6. \u5f62\u72b6\u5230\u5206\u5b50\u7684\u751f\u6210\u6a21\u578b Shape2Mol \u7684\u793a\u610f<\/p>\n

\u6700\u8fd1\u6211\u4eec\u805a\u7126\u5728\u5927\u5206\u5b50\u836f\u7269\u8bbe\u8ba1\u4e0a\uff0c\u66f4\u4e00\u822c\u7684\u95ee\u9898\u662f\u86cb\u767d\u8d28\u8bbe\u8ba1\u3002\u86cb\u767d\u8d28\u8bbe\u8ba1\u5305\u62ec\u6297\u4f53\u836f\u7269\u8bbe\u8ba1\uff0c\u591a\u80bd\u836f\u7269\u8bbe\u8ba1\u7b49\u3002\u5982\u679c\u77e5\u9053\u4e86\u86cb\u767d\u8d28\u5e8f\u5217\uff08\u6c28\u57fa\u9178\u5e8f\u5217\uff09\uff0c\u5c31\u53ef\u4ee5\u9884\u6d4b\u5176\u7ed3\u6784\uff0c\u77e5\u9053\u4e86\u86cb\u767d\u8d28\u7ed3\u6784\u4e5f\u5c31\u80fd\u9884\u6d4b\u5176\u529f\u80fd\u3002\u8fd9\u5c31\u662f\u8457\u540d<\/span>\u7684 AlphaFold \u505a\u7684\u4e8b\u60c5\u3002\u86cb\u767d\u8d28\u8bbe\u8ba1\u53ef\u4ee5\u770b\u4f5c\u662f\u4e00\u4e2a\u53cd\u5411\u7684\u8fc7\u7a0b\u3002\u4e00\u822c\u4ece\u529f\u80fd\u51fa\u53d1\u51b3\u5b9a\u5bf9\u5e94\u7684\u86cb\u767d\u8d28\u7ed3\u6784\uff0c\u518d\u6839\u636e\u86cb\u767d\u8d28\u7ed3\u6784\u51b3\u5b9a\u5bf9\u5e94\u7684\u86cb\u767d\u8d28\u5e8f\u5217\u3002\u6211\u4eec\u5f00\u53d1\u4e86\u4ece\u86cb\u767d\u8d28\u7ed3\u6784\u751f\u6210\u86cb\u767d\u8d28\u5e8f\u5217\u7684\u6a21\u578b LM-Design\u3002<\/p>\n

LM-Design \u7684\u8f93\u5165\u662f\u86cb\u767d\u8d28\u7ed3\u6784\uff0c\u8f93\u51fa\u662f\u5bf9\u5e94\u7684\u86cb\u767d\u8d28\u5e8f\u5217\u3002LM-Design \u7531\u7ed3\u6784\u7f16\u7801\u5668\u548c\u5e8f\u5217\u89e3\u7801\u5668\u7ec4\u6210\u3002\u5176\u4e2d\uff0c\u7ed3\u6784\u7f16\u7801\u5668\u662f\u5df2\u8bad\u7ec3\u597d\u7684\u8868\u793a\u86cb\u767d\u8d28\u7ed3\u6784\u7684\u56fe\u795e\u7ecf\u7f51\u7edc;\u5e8f\u5217\u89e3\u7801\u5668\u57fa\u4e8e\u5df2\u9884\u8bad\u7ec3\u597d\u7684\u5927\u89c4\u6a21\u86cb\u767d\u8d28\u8bed\u8a00\u6a21\u578b\uff08Protein Language Model\uff09\uff0c\u4e0e BERT\/Transformer Encoder \u76f8\u4f3c(\u4f7f\u7528\u53cc\u5411\u81ea\u6ce8\u610f\u529b)\uff0c\u5728\u6700\u540e<\/span>\u4e00\u5c42\u63d2\u5165\u4e00\u4e2a\u7ed3\u6784\u9002\u914d\u5668(Structural Adaptor)\u3002\u7ed3\u6784\u9002\u914d\u5668\u7684\u53c2\u6570\u662f\u5f85\u5b66\u4e60\u7684\u3002\u56fe7\u7ed9\u51fa LM-Design \u7684\u6a21\u578b\u67b6\u6784\u3002<\/p>\n

\n

\u56fe7. \u86cb\u767d\u8d28\u7ed3\u6784\u5230\u5e8f\u5217\u7684\u751f\u6210\u6a21\u578b LM-Design \u7684\u67b6\u6784<\/p>\n

LM-Design \u7684\u5b66\u4e60\u548c\u9884\u6d4b\u662f\u63a9\u7801\u8bed\u8a00\u5efa\u6a21\uff08Masked Language Modeling\uff09\uff0c\u4e0e BERT \u6a21\u578b\u7684\u8bad\u7ec3\u76f8\u4f3c\uff0c\u5176\u76ee\u6807\u662f\u591a\u6b21\u8fd8\u539f\u88ab\u63a9\u7801\u7684\u5e8f\u5217\u4e2d\u7684\u7b26\u53f7(\u6c28\u57fa\u9178)\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u57fa\u4e8e\u5df2\u8bad\u7ec3\u597d\u7684\u86cb\u767d\u8d28\u8bed\u8a00\u6a21\u578b\u4e2d\u7684\u4fe1\u606f\uff0c\u4ee5\u53ca\u5f53\u524d\u7684\u86cb\u767d\u8d28\u7ed3\u6784\u4fe1\u606f\uff0c\u5bf9\u86cb\u767d\u8d28\u5e8f\u5217\u8fdb\u884c\u591a\u6b21\u6539\u5199\u3002LM-Design \u57fa\u4e8e\u5168\u5c40\u5e8f\u5217\u4fe1\u606f\u5bf9\u5176\u4e2d\u5f88\u5c11\u4e00\u90e8\u5206\u7b26\u53f7(\u6c28\u57fa\u9178)\u8fdb\u884c\u6539\u5199\uff0c\u6240\u4ee5\u5bf9\u86cb\u767d\u8d28\u8fdc\u8ddd\u79bb\u4f9d\u5b58\u5173\u7cfb\u80fd\u591f\u8fdb\u884c\u5f88\u597d\u7684\u8868\u793a\u548c\u9884\u6d4b\u3002\u6ce8:\u86cb\u767d\u8d28\u6298\u53e0\u4e4b\u540e\uff0c\u5e8f\u5217\u4e0a\u8ddd\u79bb\u5f88\u8fdc\u7684\u6c28\u57fa\u9178\u5728\u7ed3\u6784\u4e0a\u4e5f\u53ef\u80fd\u5f88\u8fd1\u3002<\/p>\n

\u73b0\u5b9e\u4e2d\u6709\u5927\u91cf\u7ecf\u8fc7\u6d4b\u5e8f\u7684\u86cb\u767d\u8d28\u5e8f\u5217\u6570\u636e\uff0c\u4f46\u53ea\u6709\u5c11\u91cf\u7684\u86cb\u767d\u8d28\u7ed3\u6784\u4e0e\u5e8f\u5217\u5bf9\u9f50\u7684\u6570\u636e\u3002LM-Design \u7684\u4e00\u4e2a\u4f18\u52bf\u662f\u53ef\u4ee5\u5229\u7528\u6d77\u91cf\u7684\u86cb\u767d\u8d28\u5e8f\u5217\u6570\u636e\uff0c\u5145\u5206\u5b66\u4e60\u548c\u5229\u7528\u86cb\u767d\u8d28\u5e8f\u5217\u4e4b\u95f4\u7684\u8fdb\u5316\u4e2d\u4ea7\u751f\u7684\u5173\u8054\u5173\u7cfb\uff0c\u5927\u5e45\u63d0\u9ad8\u4ece\u86cb\u767d\u8d28\u7ed3\u6784\u5230\u5e8f\u5217\u751f\u6210\u7684\u9884\u6d4b\u51c6\u786e\u7387\u3002\u6b64\u5916\uff0c\u6211\u4eec\u53d1\u73b0\u589e\u5927\u9884\u8bad\u7ec3\u86cb\u767d\u8d28\u6a21\u578b\u7684\u89c4\u6a21\u53ef\u4ee5\u8fdb\u4e00\u6b65\u63d0\u5347\u51c6\u786e\u7387\u3002\u5982\u56fe8\u6240\u793a\uff0cLM-Design \u662f\u76ee\u524d\u6548\u679c\u6700\u597d<\/span>\u7684\u86cb\u767d\u8d28\u5e8f\u5217\u751f\u6210\u6a21\u578b\uff0c\u56fe\u4e2d\u5706\u7684\u5927\u5c0f\u8868\u793a\u6a21\u578b\u7684\u53c2\u6570\u91cf\u3002<\/p>\n

\n

\u56fe8. \u86cb\u767d\u8d28\u5e8f\u5217\u751f\u6210\u65b9\u6cd5\u7684\u7cbe\u5ea6<\/p>\n

\u81f4\u8c22<\/strong><\/p>\n

\u611f\u8c22\u4efb\u7ef4\u7edc\u3001\u5415\u5b9a\u987a\u3001\u987e\u5168\u5168\u3001\u5434\u51ef\u3001\u90d1\u5728\u7fd4\u3001\u5468\u5955\u3001\u7f57\u66fc\u5e73\u3001\u5f20\u9707\u5b87\u4e3a\u672c\u6587\u64b0\u5199\u63d0\u4f9b\u7684\u5efa\u8bae\u548c\u5e2e\u52a9\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

\u58f0\u660e:\u672c\u6587\u6765\u81ea\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7 \u673a\u5668\u4e4b\u5fc3\uff08ID:almosthuman2...<\/p>\n","protected":false},"author":1,"featured_media":831,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[19],"tags":[],"class_list":["post-3605","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-dongtai"],"_links":{"self":[{"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/posts\/3605","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/comments?post=3605"}],"version-history":[{"count":0,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/posts\/3605\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/media\/831"}],"wp:attachment":[{"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/media?parent=3605"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/categories?post=3605"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.idea2003.com\/wp-json\/wp\/v2\/tags?post=3605"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}