{"id":1061700,"date":"2021-02-26T01:28:00","date_gmt":"2021-02-26T09:28:00","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&p=1061700"},"modified":"2024-09-25T00:43:21","modified_gmt":"2024-09-25T07:43:21","slug":"li-dong","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/li-dong\/","title":{"rendered":"\u201c\u8bba\u6587\u5956\u9879\u9ad8\u4ea7\u6237\u201d\u8463\u529b\u7684\u81ea\u6211\u4fee\u517b\uff1aAim Higher"},"content":{"rendered":"\n

\u7f16\u8005\u6309\uff1a\u83b7\u5f97\u56fd\u9645\u9876\u4f1a\u7684\u6700\u4f73\u8bba\u6587\u5956\u662f\u6bcf\u4e00\u4f4d\u79d1\u7814\u4eba\u5458\u7684\u76ee\u6807\u3002\u535a\u58eb\u6bd5\u4e1a\u521a\u521a3\u5e74\u7684\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e3b\u7ba1\u7814\u7a76\u5458\u8463\u529b\u57282021\u5e74\u4f0a\u59cb\u8fde\u7eed\u83b7\u5f97\u4e86 AI \u9886\u57df\u7684\u4e24\u4e2a\u6700\u4f73\u8bba\u6587 Runner-Up \u5956\u3002\u4eca\u5929\u5c31\u8ba9\u6211\u4eec\u4e00\u8d77\u6765\u770b\u770b\uff0c\u4ed6\u62e5\u6709\u4e86\u600e\u6837\u7684\u5929\u65f6\u3001\u5730\u5229\u4e0e\u4eba\u548c\u3002<\/p>\n\n\n\n

\n\n\n\n

\u8fd1\u671f\uff0c\u4eba\u5de5\u667a\u80fd\u4fc3\u8fdb\u534f\u4f1a AAAI \u8054\u5408\u56fd\u9645\u8ba1\u7b97\u673a\u5b66\u4f1a ACM SIGAI \u9996\u6b21\u53d1\u5e03\u4e86\u535a\u58eb\u8bba\u6587\u5956\uff0c\u5e76\u5728\u4eca\u5e742\u6708\u4e3e\u529e\u7684\u7b2c35\u5c4a\u4eba\u5de5\u667a\u80fd\u56fd\u9645\u4f1a\u8bae AAAI \u4e0a\u6b63\u5f0f\u516c\u5e03\u4e86\u4e09\u4f4d\u83b7\u5956\u8005\u7684\u540d\u5355\u3002\u8be5\u5956\u9879\u65e8\u5728\u53d1\u73b0\u548c\u9f13\u52b1\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u4f18\u79c0\u535a\u58eb\u7814\u7a76\u548c\u8bba\u6587\uff0c\u5e76\u7531 AI \u9886\u57df\u7684\u56fd\u9645\u77e5\u540d\u9ad8\u6821\u6bcf\u5e74\u63a8\u4e3e\u4e00\u4f4d\u5019\u9009\u4eba\u53c2\u52a0\uff0c\u5956\u9879\u7684\u542b\u91d1\u91cf\u53ef\u89c1\u4e00\u6591\u3002\u66fe\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5b9e\u4e60\u3001\u535a\u58eb\u6bd5\u4e1a\u4e8e\u7231\u4e01\u5821\u5927\u5b66\u7684\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e3b\u7ba1\u7814\u7a76\u5458\u8463\u529b\u83b7\u5f97\u4e86\u9996\u5c4a AAAI\/ACM SIGAI \u535a\u58eb\u8bba\u6587\u5956\u7684 Runner-Up \u5956\u3002<\/p>\n\n\n\n

\"\u9996\u5c4a<\/figure>\n\n\n\n

\u8463\u529b\u7684\u83b7\u5956\u8bba\u6587\uff1aLearning Natural Language Interfaces with Neural Models [1]\uff08\u7528\u795e\u7ecf\u6a21\u578b\u6765\u5b66\u4e60\u81ea\u7136\u8bed\u8a00\u63a5\u53e3\uff09\u7684\u4e3b\u9898\u662f\u5173\u4e8e\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u89e3\u6790\u2014\u2014\u5c06\u81ea\u7136\u8bed\u8a00\u8868\u8fbe\u8f6c\u5316\u6210\u673a\u5668\u53ef\u4ee5\u6267\u884c\u7684\u7b26\u53f7\u8868\u793a\uff0c\u4ece\u800c\u901a\u8fc7\u4f7f\u7528\u795e\u7ecf\u7f51\u7edc\u53bb\u6784\u5efa\u66f4\u597d\u7684\u4eba\u673a\u4ea4\u4e92\u81ea\u7136\u8bed\u8a00\u754c\u9762\u3002\u6b64\u5916\uff0c\u8463\u529b\u7684\u53e6\u4e00\u7bc7\u8bba\u6587\uff1aSelf-Attention Attribution: Interpreting Information Interaction Inside Transformer [2]\uff0c\u4e5f\u83b7\u5f97\u4e86 AAAI 2021 Best Paper Runner-Up \u5956\u3002\u540c\u65f6\u83b7\u5f97\u4e24\u4e2a\u5956\u9879\uff0c\u53ef\u4ee5\u8bf4\u662f\u5bf9\u8463\u529b\u8fd1\u5341\u5e74\u7814\u7a76\u79ef\u7d2f\u7684\u8ba4\u53ef\u548c\u8912\u5956\uff0c\u5176\u80cc\u540e\u7684\u652f\u6491\u662f\u4e00\u7cfb\u5217\u7814\u7a76\u7406\u8bba\u548c\u6280\u672f\u521b\u65b0\uff0c\u4ee5\u53ca\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u6df1\u539a\u6e0a\u6e90\u3002<\/p>\n\n\n\n

8\u5e74\u524d\u5c31\u5f00\u59cb\u7528\u6df1\u5ea6\u5b66\u4e60\u63a2\u7d22NLP\uff0c\u62a2\u5f97\u5148\u673a<\/h2>\n\n\n\n

2012\u5e74\uff0c\u6b63\u5728\u8bfb\u5927\u56db\u7684\u8463\u529b\u5f00\u59cb\u4e86\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e09\u5e74\u7684\u5b9e\u4e60\u751f\u6d3b\u3002\u5f7c\u65f6\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u5df2\u7ecf\u7528 AI \u6280\u672f\u53bb\u63a2\u7d22\u6700\u5177\u6311\u6218\u6027\u7684\u95ee\u9898\uff0c\u5e76\u57282012\u5e74\u7684\u201c\u4e8c\u5341\u4e00\u4e16\u7eaa\u7684\u8ba1\u7b97\u201d\u5927\u4f1a\u4e0a\u5c55\u793a\u4e86\u5fae\u8f6f\u5b9e\u65f6\u8bed\u97f3\u673a\u5668\u7ffb\u8bd1\u7cfb\u7edf\uff0c\u5b9e\u73b0\u4e86\u8bed\u97f3\u8bc6\u522b\u3001\u6587\u672c\u7ffb\u8bd1\u53ca\u8bed\u97f3\u5408\u6210\u4e09\u9879\u6280\u672f\u4e0a\u7684\u5de8\u5927\u7a81\u7834\u3002\u8fd9\u71c3\u8d77\u4e86\u8463\u529b\u5bf9 NLP\uff08\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff09\u7684\u5174\u8da3\u3002<\/p>\n\n\n\n

\u5f53\u65f6\u6df1\u5ea6\u5b66\u4e60\u521d\u9732\u950b\u8292\uff0c\u5fae\u8f6f\u7684\u9093\u529b\u7814\u7a76\u5458\u5229\u7528\u76f8\u5173\u6280\u672f\u5728\u8bed\u97f3\u8bc6\u522b\u7b49\u95ee\u9898\u4e0a\u83b7\u5f97\u4e86\u7a81\u7834\u3002\u8463\u529b\u7684 Mentor \u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u9996\u5e2d\u7814\u7a76\u5458\u97e6\u798f\u5982\u654f\u9510\u5730\u5bdf\u89c9\u5230\u4e86\u8fd9\u4e00\u6280\u672f\u53d8\u9769\uff0c\u9f13\u52b1\u8463\u529b\u4ece\u96f6\u5f00\u59cb\u642d\u5efa\u6df1\u5ea6\u5b66\u4e60\u5e93\uff0c\u5e76\u4e00\u8d77\u63a8\u52a8\u4e86\u6df1\u5ea6\u5b66\u4e60\u5728\u81ea\u52a8\u95ee\u7b54\u3001\u4eba\u673a\u5bf9\u8bdd\u3001\u60c5\u611f\u5206\u6790\u7b49\u95ee\u9898\u4e0a\u7684\u843d\u5730\u3002\u8fd9\u4e2a\u5951\u673a\u8ba9\u8463\u529b\u5bf9 NLP \u7684\u7ecf\u5178\u65b9\u6cd5\u548c\u524d\u6cbf\u7814\u7a76\u90fd\u6709\u4e86\u7cfb\u7edf\u7684\u8ba4\u8bc6\u3002<\/p>\n\n\n\n

\"\u8463\u529b\uff08\u53f3\uff09\u57282015\u5e74\u201c\u4e8c\u5341\u4e00\u4e16\u7eaa\u7684\u8ba1\u7b97\u201d\u5927\u4f1a\u4e0a\u83b7\u5f97\u5fae\u8f6f\u5b66\u8005\u79f0\u53f7\uff0c\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66\u6570\u636e\u79d1\u5b66\u7814\u7a76\u6240\u6240\u957f\u5468\u4ee5\u771f\u6559\u6388\uff08\u5de6\uff09\u4e3a\u5176\u9881\u5956\"
\u8463\u529b\uff08\u53f3\uff09\u57282015\u5e74\u201c\u4e8c\u5341\u4e00\u4e16\u7eaa\u7684\u8ba1\u7b97\u201d\u5927\u4f1a\u4e0a\u83b7\u5f97\u5fae\u8f6f\u5b66\u8005\u79f0\u53f7\uff0c\u54e5\u4f26\u6bd4\u4e9a\u5927\u5b66\u6570\u636e\u79d1\u5b66\u7814\u7a76\u6240\u6240\u957f\u5468\u4ee5\u771f\u6559\u6388\uff08\u5de6\uff09\u4e3a\u5176\u9881\u5956<\/em><\/figcaption><\/figure>\n\n\n\n

\u201c\u6b63\u662f\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5bf9\u524d\u6cbf\u6280\u672f\u7684\u524d\u77bb\u6027\uff0c\u8ba9\u6211\u53ef\u4ee5\u5f88\u65e9\u5c31\u63a5\u89e6\u5230\u6df1\u5ea6\u5b66\u4e60\uff0c\u6709\u673a\u4f1a\u63a2\u7a76 NLP \u7814\u7a76\u65b9\u5411\u7684\u66f4\u591a\u6f5c\u529b\uff0c\u201d\u8463\u529b\u8868\u793a\u3002\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u201c\u7ed3\u8bc6\u201d\u7684 NLP \u548c\u6df1\u5ea6\u5b66\u4e60\uff0c\u6210\u4e3a\u4e86\u8463\u529b\u6b64\u540e\u7684\u7814\u7a76\u4e3b\u7ebf\uff0c\u5e76\u4e14\u57fa\u4e8e\u8fd9\u4e2a\u9886\u57df\u8d8a\u6316\u8d8a\u6df1\uff0c\u4e0d\u4ec5\u6210\u4e3a\u201c\u8bba\u6587\u9ad8\u4ea7\u6237\u201d\uff0c\u4e5f\u6210\u5c31\u4e86\u6b64\u6b21 AAAI\/ACM SIGAI \u7684\u535a\u58eb\u8bba\u6587\u5956\u3002<\/p>\n\n\n\n

\u8bba\u6587\u9ad8\u4ea7\u6237\uff1a\u9009\u5bf9\u65b9\u5411\uff0c\u6301\u7eed\u53d1\u529b<\/h2>\n\n\n\n

\u8463\u529b\u7684\u535a\u58eb\u8bba\u6587 Learning Natural Language Interfaces with Neural Models\uff0c\u5176\u6838\u5fc3\u662f\u8ba9\u673a\u5668\u53ef\u4ee5\u201c\u7406\u89e3\u201d\u4eba\u7c7b\u7684\u8bed\u8a00\u2014\u2014\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u5316\u6210\u4e3a\u673a\u5668\u53ef\u4ee5\u7406\u89e3\u7684\u7ed3\u6784\u5316\u3001\u5f62\u5f0f\u5316\u7684\u8bed\u8a00\uff0c\u5b9e\u73b0\u66f4\u81ea\u7136\u7684\u4eba\u673a\u4ea4\u4e92\u63a5\u53e3\u3002\u5728\u6b64\u4e4b\u524d\uff0c\u5b66\u672f\u754c\u666e\u904d\u901a\u8fc7\u6784\u5efa\u591a\u6a21\u5757\u6d41\u6c34\u7ebf\u6a21\u578b\u8fdb\u884c\u81ea\u7136\u8bed\u8a00\u7684\u8bed\u4e49\u89e3\u6790\u3002\u56e0\u4e3a\u53d7\u9650\u4e8e\u6a21\u5757\u4e4b\u95f4\u7684\u9519\u8bef\u6269\u6563\uff0c\u4ee5\u53ca\u5404\u79cd\u7e41\u6742\u7684\u4eba\u5de5\u89c4\u5219\uff0c\u6240\u4ee5\u8bed\u4e49\u89e3\u6790\u53ea\u80fd\u5728\u5c0f\u90e8\u5206\u73af\u5883\u4e2d\u5e94\u7528\u3002\u4f46\u8463\u529b\u7684\u7814\u7a76\u5c06\u7b26\u53f7\u5316\u7684\u4fe1\u606f\u4e0e\u795e\u7ecf\u7f51\u7edc\u6709\u673a\u5730\u7ed3\u5408\uff0c\u7834\u89e3\u4e86\u4eba\u4eec\u5bf9\u795e\u7ecf\u7f51\u7edc\u80fd\u5426\u51c6\u786e\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u7684\u7591\u60d1\uff0c\u83b7\u5f97\u4e86\u5b66\u672f\u754c\u7684\u5173\u6ce8\uff0c\u540c\u65f6\u4e5f\u63a8\u52a8\u4e86\u901a\u7528 AI \u7814\u7a76\u7684\u8fdb\u6b65\u3002\u8fd9\u4e4b\u540e\uff0c\u8463\u529b\u8fd8\u9488\u5bf9\u673a\u5668\u591a\u8bed\u4e49\u7406\u89e3\u8fdb\u884c\u4e86\u7814\u7a76\uff0c\u5e76\u5b9e\u73b0\u4e86\u673a\u5668\u5bf9\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u7f6e\u4fe1\u5ea6\u7684\u5224\u65ad\u3002<\/p>\n\n\n\n

\u8fd9\u7bc7\u83b7\u5956\u7684\u535a\u58eb\u8bba\u6587\u662f\u8463\u529b\u6b64\u524d\u4e00\u7cfb\u5217\u7814\u7a76\u5de5\u4f5c\u7684\u4e00\u4e2a\u603b\u7ed3\u30022016\u5e74\uff0c\u8463\u529b\u5728 ACL \u5927\u4f1a\u4e0a\u53d1\u8868\u7684\u8bba\u6587 Language to Logical Form with Neural Attention [3]\uff0c\u63d0\u51fa\u4e86\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u5316\u4e3a\u5f62\u5f0f\u5316\u8868\u793a\u7684\u65b9\u6cd5\uff0c\u5e76\u6784\u5efa\u51fa\u4e86\u76f8\u5e94\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u6765\u878d\u5165\u5f62\u5f0f\u5316\u8bed\u8a00\u7684\u8bed\u6cd5\u4fe1\u606f\u30022017\u5e74\uff0c\u4ed6\u5728 EMNLP \u5927\u4f1a\u4e0a\u53d1\u8868\u7684\u8bba\u6587 Learning to Paraphrase for Question Answering [4]\uff0c\u5219\u63d0\u51fa\u4e86\u5bf9\u540c\u4e00\u4e2a\u542b\u4e49\u6709\u591a\u79cd\u4e0d\u540c\u8bf4\u6cd5\u7684\u53e5\u5b50\u8fdb\u884c\u5efa\u6a21\u30022018\u5e74\uff0c\u4ed6\u5728 ACL \u4e0a\u53d1\u8868\u7684\u4e24\u7bc7\u8bba\u6587 Confidence Modeling for Neural Semantic Parsing [5] \u548c Coarse-to-Fine Decoding for Neural Semantic Parsing [6]\uff0c\u5206\u522b\u5bf9\u8bed\u4e49\u89e3\u6790\u7684\u7f6e\u4fe1\u5ea6\u4f30\u8ba1\u548c\u591a\u7c92\u5ea6\u89e3\u7801\u8fdb\u884c\u4e86\u7814\u7a76\uff0c\u5176\u4e2d\u524d\u4e00\u7bc7\u662f\u5728\u5fae\u8f6f\u96f7\u5fb7\u8499\u7814\u7a76\u9662\u5b9e\u4e60\u671f\u95f4\u5b8c\u6210\u7684\u3002\u540e\u4e00\u7bc7\u4e5f\u83b7\u5f97\u4e86\u5f53\u5e74\u7684 ACL Best Paper Honorable Mention\u3002<\/p>\n\n\n\n

\"\u8463\u529b\u5728\u7231\u4e01\u5821\u5927\u5b66\u535a\u58eb\u6bd5\u4e1a\u7559\u5f71\"
\u8463\u529b\u5728\u7231\u4e01\u5821\u5927\u5b66\u535a\u58eb\u6bd5\u4e1a\u7559\u5f71<\/em><\/figcaption><\/figure>\n\n\n\n

\u4e00\u5207\u90fd\u6709\u8ff9\u53ef\u5faa\uff0c\u8bba\u6587\u5956\u9879\u5e76\u975e\u51ed\u7a7a\u800c\u6765\u3002\u8463\u529b\u8868\u793a\uff0c\u201c\u6211\u5f88\u5e78\u8fd0\u80fd\u591f\u6210\u4e3a\u4e1a\u5185\u6700\u65e9\u4e00\u6279\u63a5\u89e6 NLP \u548c\u6df1\u5ea6\u5b66\u4e60\u8fd9\u4e2a\u4ea4\u53c9\u9886\u57df\u7684\u7814\u7a76\u4eba\u5458\uff0c\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u6211\u8ddf\u968f\u7740\u5168\u7403\u9876\u5c16\u7684\u7814\u7a76\u5458\u4eec\uff0c\u8e0f\u4e0a\u4e86\u8fd9\u4e2a\u5168\u65b0\u7684\u8d5b\u9053\u5e76\u6253\u4e0b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\uff0c\u6240\u4ee5\u5728\u535a\u58eb\u671f\u95f4\u6211\u624d\u5f97\u4ee5\u8fdb\u4e00\u6b65\u52a0\u901f\u3002\u201d2018\u5e74\u8463\u529b\u5728\u5b8c\u6210\u4e86\u535a\u58eb\u5b66\u4e1a\u540e\uff0c\u518d\u6b21\u56de\u5230\u7814\u7a76\u9662\uff0c\u6210\u4e3a\u4e86\u81ea\u7136\u8bed\u8a00\u8ba1\u7b97\u7ec4\u7684\u4e00\u540d\u7814\u7a76\u5458\u3002<\/p>\n\n\n\n

Aim higher, take a risk!<\/h2>\n\n\n\n

\u4ece\u5b66\u751f\u5230\u7814\u7a76\u5458\u7684\u8f6c\u53d8\uff0c\u4e0d\u4ec5\u4ec5\u662f\u89d2\u8272\u4e0a\u53d1\u751f\u4e86\u6539\u53d8\uff0c\u201c\u601d\u7ef4\u65b9\u5f0f\u4e5f\u5f00\u59cb\u6709\u6240\u4e0d\u540c\u201d\u3002\u8463\u529b\u7528\u8ba1\u7b97\u673a\u9886\u57df\u79d1\u7814\u4eba\u5458\u7279\u6709\u7684\u65b9\u5f0f\u5c06\u8fd9\u4e2a\u8f6c\u53d8\u505a\u4e86\u4e00\u4e2a\u7c7b\u6bd4\uff1a\u5b66\u751f\u548c\u5bfc\u5e08\u5c31\u50cf\u751f\u6210\u5f0f\u5bf9\u6297\u7f51\u7edc\uff08GAN\uff09\u4e2d\u7684\u751f\u6210\u6a21\u578b\uff08Generative Model\uff09\u548c\u5224\u522b\u6a21\u578b\uff08Discriminative Model\uff09\u3002\u5f53\u201c\u751f\u6210\u5668\u201d\u6709\u4e86\u65b0\u7684\u60f3\u6cd5\uff0c\u8981\u4ed8\u8bf8\u884c\u52a8\u65f6\uff0c\u201c\u5224\u522b\u5668\u201d\u4f1a\u7ed9\u51fa\u6307\u5bfc\u548c\u53c2\u8003\uff0c\u4ee5\u907f\u514d\u8d70\u5f2f\u8def\uff0c\u4ece\u800c\u8ba9\u201c\u751f\u6210\u5668\u201d\u53ef\u4ee5\u5728\u6f5c\u79fb\u9ed8\u5316\u4e2d\uff0c\u901a\u8fc7\u5927\u91cf\u7684\u4ea4\u6d41\u5207\u78cb\u4e0d\u65ad\u6210\u957f\u3002<\/p>\n\n\n\n

\u800c\u5de5\u4f5c\u4e2d\u4e0d\u518d\u6709\u201c\u5224\u522b\u5668\u201d\u8fd9\u4e2a\u89d2\u8272\uff0c\u4ece\u60f3\u6cd5\u5230\u5b9e\u8df5\uff0c\u518d\u5230\u8d70\u5b8c\u5168\u8fc7\u7a0b\uff0c\u66f4\u591a\u7684\u4e8b\u7269\u90fd\u9700\u8981\u81ea\u5df1\u627f\u62c5\u3002\u201c\u4f46\u662f\uff0c\u4e5f\u53ea\u6709\u5728\u5de5\u4f5c\u73af\u5883\u4e2d\uff0c\u4f60\u624d\u80fd\u63a5\u89e6\u5230\u66f4\u591a\u5b9e\u9645\u7684\u573a\u666f\uff0c\u53d1\u73b0\u66f4\u591a\u73b0\u5b9e\u4e2d\u5b58\u5728\u7684\u95ee\u9898\u3002\u6bd4\u5982\uff0c\u5bf9\u8bed\u4e49\u5206\u6790\u4e2d\u7f6e\u4fe1\u5ea6\u4f30\u8ba1\u7684\u8fd9\u4e2a\u9898\u76ee\uff0c\u5c31\u662f\u5f53\u65f6\u5728\u5fae\u8f6f\u96f7\u5fb7\u8499\u7814\u7a76\u9662\u5b9e\u4e60\u65f6\uff0c\u6211\u7684 Mentor Chris Quirk \u7814\u7a76\u5458\u7ed3\u5408\u53ef\u4fe1\u8d56 AI \u7cfb\u7edf\u7684\u9700\u6c42\uff0c\u63d0\u51fa\u6211\u4eec\u5e94\u8be5\u5728\u8fd9\u4e2a\u65b9\u5411\u63a2\u7d22\u7684\uff0c\u201d\u8463\u529b\u8bf4\u3002<\/p>\n\n\n\n

\"a
\u8463\u529b\uff08\u540e\u6392\u53f3\u4e09\uff09\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u81ea\u7136\u8bed\u8a00\u8ba1\u7b97\u7ec4\u7684\u90e8\u5206\u540c\u4e8b\uff0c\u5728\u62ff\u5230\u7b2c\u516d\u5c4a\u4e16\u754c\u4e92\u8054\u7f51\u5927\u4f1a\u201c\u4e16\u754c\u4e92\u8054\u7f51\u9886\u5148\u79d1\u6280\u6210\u679c\u201d\u5956\u676f\u540e\u5408\u5f71<\/em><\/figcaption><\/figure>\n\n\n\n

\u6b63\u5f0f\u52a0\u5165\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u8fd9\u4e24\u5e74\uff0c\u8463\u529b\u4ecd\u7136\u7ee7\u7eed\u4ece\u4e8b NLP \u9886\u57df\u7684\u7814\u7a76\uff0c\u76ee\u524d\u4ed6\u4e3b\u8981\u5173\u6ce8\u5927\u89c4\u6a21\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u3002\u4ed6\u548c\u56e2\u961f\u6240\u505a\u7684 UniLM\uff08Unified Language Model\uff0c\u7edf\u4e00\u8bed\u8a00\u6a21\u578b\uff09\uff0c\u53ef\u4ee5\u5c06\u81ea\u7136\u8bed\u8a00\u7406\u89e3\uff08NLU\uff09\u4e0e\u81ea\u7136\u8bed\u8a00\u751f\u6210\uff08NLG\uff09\u8fd9\u4e24\u79cd\u4e0d\u540c\u7684\u4efb\u52a1\u7edf\u4e00\u5230\u4e00\u4e2a\u6a21\u578b\u4e2d\uff0c\u4ece\u800c\u66f4\u597d\u5730\u652f\u6301\u4e0a\u5c42\u5e94\u7528\u3002\u4ed6\u4eec\u8fd8\u5c06\u76f8\u5173\u65b9\u6cd5\u6269\u5c55\u5230\u4e86\u591a\u8bed\u8a00\uff0c\u63d0\u51fa\u4e86InfoXLM \u6a21\u578b\u3002\u8fd9\u4e9b\u52aa\u529b\u90fd\u4ece\u5e95\u5c42\u7814\u7a76\u5f88\u597d\u5730\u652f\u6491\u4e86\u5fae\u8f6f\u7684\u5404\u9879 NLP \u5e94\u7528\uff0c\u4e5f\u8fdb\u4e00\u6b65\u63a8\u52a8\u4e86 NLP \u79d1\u7814\u9886\u57df\u548c\u4e1a\u754c\u76f8\u5173\u5e94\u7528\u7684\u53d1\u5c55\u3002<\/p>\n\n\n\n

\u5bf9\u4e8e\u7814\u7a76\u65b9\u6cd5\uff0c\u8463\u529b\u4e5f\u6709\u4e00\u4e9b\u81ea\u5df1\u7684\u5fc3\u5f97\uff0c\u4ed6\u4ecb\u7ecd\u9053\uff1a\u9996\u5148\uff0c\u8981\u5c3d\u529b\u6253\u9020\u81ea\u5df1\u7684\u4ee3\u8868\u4f5c\uff0c\u4e00\u4e2a milestone\u3001\u4e00\u4e2a milestone \u5730\u653b\u514b\uff0c\u7136\u540e\u5c06\u8fd9\u4e9b\u70b9\u4e32\u8054\u8d77\u6765\u5c31\u53ef\u4ee5\u5f62\u6210\u81ea\u5df1\u7684\u7814\u7a76\u4e3b\u7ebf\uff1b\u5176\u6b21\uff0c\u4e00\u5b9a\u8981\u534f\u8c03\u597d\u957f\u671f\u76ee\u6807\u548c\u77ed\u671f\u76ee\u6807\u4e4b\u95f4\u7684\u5173\u7cfb\uff0c\u4e0d\u80fd\u88ab\u77ed\u671f\u76ee\u6807\u5185\u5377\uff0c\u4e0d\u7136\u522b\u4eba\u6362\u4e2a\u8def\u5f84\u5c31\u53ef\u4ee5\u5bf9\u4f60\u8fdb\u884c\u964d\u7ef4\u6253\u51fb\uff0c\u8981\u9636\u6bb5\u6027\u5730\u4ece\u81ea\u5df1\u7684\u8303\u56f4\u91cc\u8df3\u51fa\u6765\u8fdb\u884c\u5ba1\u89c6\uff0c\u4ece\u4e0d\u540c\u89d2\u5ea6\u53bb\u8003\u5bdf\u81ea\u5df1\u7684\u7814\u7a76\uff1b\u6700\u540e\uff0c\u662f\u53e5\u8001\u8bdd\u2014\u2014\u8981\u4e0d\u65ad\u521b\u65b0\uff0c\u800c\u5b9e\u73b0\u521b\u65b0\u7684\u4e00\u4e2a\u6709\u6548\u65b9\u6cd5\u5c31\u662f aim higher\uff0c\u7ed9\u81ea\u5df1\u5236\u5b9a\u4e00\u4e2a\u7a0d\u5fae\u9ad8\u4e00\u4e9b\u7684\u76ee\u6807\uff0c\u4e0d\u8981\u5bb3\u6015\u5931\u8d25\uff0c\u56e0\u4e3a\u5982\u679c\u6bcf\u4e2a\u9879\u76ee\u90fd\u6210\u529f\u4e86\uff0c\u90a3\u53ef\u80fd\u610f\u5473\u7740\u5e76\u6ca1\u6709\u521b\u65b0\u800c\u662f\u5728\u6c42\u7a33\uff0c\u4ece\u800c\u4f1a\u9519\u5931\u4e00\u4e9b\u673a\u4f1a\u3002<\/p>\n\n\n\n

\u53c2\u8003\u6587\u732e\uff1a<\/p>\n\n\n\n

[1] Learning Natural Language Interfaces with Neural Models<\/p>\n\n\n\n

https:\/\/era.ed.ac.uk\/handle\/1842\/35587 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n

[2] Self-Attention Attribution: Interpreting Information Interactions Inside Transformer<\/p>\n\n\n\n

https:\/\/arxiv.org\/abs\/2004.11207 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n

[3] Language to Logical Form with Neural Attention<\/p>\n\n\n\n

https:\/\/www.aclweb.org\/anthology\/P16-1004 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n

[4] Learning to Paraphrase for Question Answering<\/p>\n\n\n\n

https:\/\/www.aclweb.org\/anthology\/D17-1091 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n

[5] Coarse-to-Fine Decoding for Neural Semantic Parsing<\/p>\n\n\n\n

https:\/\/www.aclweb.org\/anthology\/P18-1068 (opens in new tab)<\/span><\/a><\/p>\n\n\n\n

[6] Confidence Modeling for Neural Semantic Parsing<\/p>\n\n\n\n

https:\/\/www.aclweb.org\/anthology\/P18-1069 (opens in new tab)<\/span><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u7f16\u8005\u6309\uff1a\u83b7\u5f97\u56fd\u9645\u9876\u4f1a\u7684\u6700\u4f73\u8bba\u6587\u5956\u662f\u6bcf\u4e00\u4f4d\u79d1\u7814\u4eba\u5458\u7684\u76ee\u6807\u3002\u535a\u58eb\u6bd5\u4e1a\u521a\u521a3\u5e74\u7684\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e3b\u7ba1\u7814\u7a76\u5458\u8463\u529b\u57282021\u5e74\u4f0a\u59cb\u8fde\u7eed\u83b7\u5f97\u4e86 AI \u9886\u57df\u7684\u4e24\u4e2a\u6700\u4f73\u8bba\u6587 Runner-Up \u5956\u3002\u4eca\u5929\u5c31\u8ba9\u6211\u4eec\u4e00\u8d77\u6765\u770b\u770b\uff0c\u4ed6\u62e5\u6709\u4e86\u600e\u6837\u7684\u5929\u65f6\u3001\u5730\u5229\u4e0e\u4eba\u548c\u3002 \u8fd1\u671f\uff0c\u4eba\u5de5\u667a\u80fd\u4fc3\u8fdb\u534f\u4f1a AAAI \u8054\u5408\u56fd\u9645\u8ba1\u7b97\u673a\u5b66\u4f1a ACM SIGAI \u9996\u6b21\u53d1\u5e03\u4e86\u535a\u58eb\u8bba\u6587\u5956\uff0c\u5e76\u5728\u4eca\u5e742\u6708\u4e3e\u529e\u7684\u7b2c35\u5c4a\u4eba\u5de5\u667a\u80fd\u56fd\u9645\u4f1a\u8bae AAAI \u4e0a\u6b63\u5f0f\u516c\u5e03\u4e86\u4e09\u4f4d\u83b7\u5956\u8005\u7684\u540d\u5355\u3002\u8be5\u5956\u9879\u65e8\u5728\u53d1\u73b0\u548c\u9f13\u52b1\u4eba\u5de5\u667a\u80fd\u9886\u57df\u7684\u4f18\u79c0\u535a\u58eb\u7814\u7a76\u548c\u8bba\u6587\uff0c\u5e76\u7531 AI \u9886\u57df\u7684\u56fd\u9645\u77e5\u540d\u9ad8\u6821\u6bcf\u5e74\u63a8\u4e3e\u4e00\u4f4d\u5019\u9009\u4eba\u53c2\u52a0\uff0c\u5956\u9879\u7684\u542b\u91d1\u91cf\u53ef\u89c1\u4e00\u6591\u3002\u66fe\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5b9e\u4e60\u3001\u535a\u58eb\u6bd5\u4e1a\u4e8e\u7231\u4e01\u5821\u5927\u5b66\u7684\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e3b\u7ba1\u7814\u7a76\u5458\u8463\u529b\u83b7\u5f97\u4e86\u9996\u5c4a AAAI\/ACM SIGAI \u535a\u58eb\u8bba\u6587\u5956\u7684 Runner-Up \u5956\u3002 \u8463\u529b\u7684\u83b7\u5956\u8bba\u6587\uff1aLearning Natural Language Interfaces with Neural Models [1]\uff08\u7528\u795e\u7ecf\u6a21\u578b\u6765\u5b66\u4e60\u81ea\u7136\u8bed\u8a00\u63a5\u53e3\uff09\u7684\u4e3b\u9898\u662f\u5173\u4e8e\u57fa\u4e8e\u795e\u7ecf\u7f51\u7edc\u7684\u8bed\u4e49\u89e3\u6790\u2014\u2014\u5c06\u81ea\u7136\u8bed\u8a00\u8868\u8fbe\u8f6c\u5316\u6210\u673a\u5668\u53ef\u4ee5\u6267\u884c\u7684\u7b26\u53f7\u8868\u793a\uff0c\u4ece\u800c\u901a\u8fc7\u4f7f\u7528\u795e\u7ecf\u7f51\u7edc\u53bb\u6784\u5efa\u66f4\u597d\u7684\u4eba\u673a\u4ea4\u4e92\u81ea\u7136\u8bed\u8a00\u754c\u9762\u3002\u6b64\u5916\uff0c\u8463\u529b\u7684\u53e6\u4e00\u7bc7\u8bba\u6587\uff1aSelf-Attention Attribution: Interpreting Information Interaction Inside Transformer [2]\uff0c\u4e5f\u83b7\u5f97\u4e86 AAAI 2021 Best Paper Runner-Up \u5956\u3002\u540c\u65f6\u83b7\u5f97\u4e24\u4e2a\u5956\u9879\uff0c\u53ef\u4ee5\u8bf4\u662f\u5bf9\u8463\u529b\u8fd1\u5341\u5e74\u7814\u7a76\u79ef\u7d2f\u7684\u8ba4\u53ef\u548c\u8912\u5956\uff0c\u5176\u80cc\u540e\u7684\u652f\u6491\u662f\u4e00\u7cfb\u5217\u7814\u7a76\u7406\u8bba\u548c\u6280\u672f\u521b\u65b0\uff0c\u4ee5\u53ca\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u6df1\u539a\u6e0a\u6e90\u3002 2012\u5e74\uff0c\u6b63\u5728\u8bfb\u5927\u56db\u7684\u8463\u529b\u5f00\u59cb\u4e86\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u4e09\u5e74\u7684\u5b9e\u4e60\u751f\u6d3b\u3002\u5f7c\u65f6\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u5df2\u7ecf\u7528 AI \u6280\u672f\u53bb\u63a2\u7d22\u6700\u5177\u6311\u6218\u6027\u7684\u95ee\u9898\uff0c\u5e76\u57282012\u5e74\u7684\u201c\u4e8c\u5341\u4e00\u4e16\u7eaa\u7684\u8ba1\u7b97\u201d\u5927\u4f1a\u4e0a\u5c55\u793a\u4e86\u5fae\u8f6f\u5b9e\u65f6\u8bed\u97f3\u673a\u5668\u7ffb\u8bd1\u7cfb\u7edf\uff0c\u5b9e\u73b0\u4e86\u8bed\u97f3\u8bc6\u522b\u3001\u6587\u672c\u7ffb\u8bd1\u53ca\u8bed\u97f3\u5408\u6210\u4e09\u9879\u6280\u672f\u4e0a\u7684\u5de8\u5927\u7a81\u7834\u3002\u8fd9\u71c3\u8d77\u4e86\u8463\u529b\u5bf9 NLP\uff08\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff09\u7684\u5174\u8da3\u3002 \u5f53\u65f6\u6df1\u5ea6\u5b66\u4e60\u521d\u9732\u950b\u8292\uff0c\u5fae\u8f6f\u7684\u9093\u529b\u7814\u7a76\u5458\u5229\u7528\u76f8\u5173\u6280\u672f\u5728\u8bed\u97f3\u8bc6\u522b\u7b49\u95ee\u9898\u4e0a\u83b7\u5f97\u4e86\u7a81\u7834\u3002\u8463\u529b\u7684 Mentor \u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u9996\u5e2d\u7814\u7a76\u5458\u97e6\u798f\u5982\u654f\u9510\u5730\u5bdf\u89c9\u5230\u4e86\u8fd9\u4e00\u6280\u672f\u53d8\u9769\uff0c\u9f13\u52b1\u8463\u529b\u4ece\u96f6\u5f00\u59cb\u642d\u5efa\u6df1\u5ea6\u5b66\u4e60\u5e93\uff0c\u5e76\u4e00\u8d77\u63a8\u52a8\u4e86\u6df1\u5ea6\u5b66\u4e60\u5728\u81ea\u52a8\u95ee\u7b54\u3001\u4eba\u673a\u5bf9\u8bdd\u3001\u60c5\u611f\u5206\u6790\u7b49\u95ee\u9898\u4e0a\u7684\u843d\u5730\u3002\u8fd9\u4e2a\u5951\u673a\u8ba9\u8463\u529b\u5bf9 NLP \u7684\u7ecf\u5178\u65b9\u6cd5\u548c\u524d\u6cbf\u7814\u7a76\u90fd\u6709\u4e86\u7cfb\u7edf\u7684\u8ba4\u8bc6\u3002 \u201c\u6b63\u662f\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5bf9\u524d\u6cbf\u6280\u672f\u7684\u524d\u77bb\u6027\uff0c\u8ba9\u6211\u53ef\u4ee5\u5f88\u65e9\u5c31\u63a5\u89e6\u5230\u6df1\u5ea6\u5b66\u4e60\uff0c\u6709\u673a\u4f1a\u63a2\u7a76 NLP \u7814\u7a76\u65b9\u5411\u7684\u66f4\u591a\u6f5c\u529b\uff0c\u201d\u8463\u529b\u8868\u793a\u3002\u5728\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u201c\u7ed3\u8bc6\u201d\u7684 NLP \u548c\u6df1\u5ea6\u5b66\u4e60\uff0c\u6210\u4e3a\u4e86\u8463\u529b\u6b64\u540e\u7684\u7814\u7a76\u4e3b\u7ebf\uff0c\u5e76\u4e14\u57fa\u4e8e\u8fd9\u4e2a\u9886\u57df\u8d8a\u6316\u8d8a\u6df1\uff0c\u4e0d\u4ec5\u6210\u4e3a\u201c\u8bba\u6587\u9ad8\u4ea7\u6237\u201d\uff0c\u4e5f\u6210\u5c31\u4e86\u6b64\u6b21 […]<\/p>\n","protected":false},"author":34512,"featured_media":1061703,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1061700","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1061700","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":1,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1061700\/revisions"}],"predecessor-version":[{"id":1061718,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1061700\/revisions\/1061718"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1061703"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1061700"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1061700"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1061700"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1061700"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}