{"id":1064457,"date":"2020-11-04T01:34:00","date_gmt":"2020-11-04T09:34:00","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&#038;p=1064457"},"modified":"2024-09-25T00:52:07","modified_gmt":"2024-09-25T07:52:07","slug":"t-ulrv2","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/t-ulrv2\/","title":{"rendered":"\u5fae\u8f6f\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578bT-ULRv2\u767b\u9876XTREME\u6392\u884c\u699c"},"content":{"rendered":"\n<p>\u7f16\u8005\u6309\uff1a\u4e3a\u8fdb\u4e00\u6b65\u5b9e\u73b0\u7528 AI \u8d4b\u80fd\u7528\u6237\u4f53\u9a8c\uff0c\u5fae\u8f6f\u6b63\u5728\u4e0d\u65ad\u62d3\u5c55\u591a\u8bed\u8a00\u6a21\u578b\u7684\u8fb9\u754c\u3002\u8fd1\u65e5\uff0c\u7531\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8054\u5408\u63a2\u7d22\u7684\u6700\u65b0\u8de8\u8bed\u8a00\u7814\u7a76\u6210\u679c\u2014\u2014\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b T-ULRv2\uff0c\u767b\u9876 XTREME \u6392\u884c\u699c\uff0cT-ULRv2 \u53ef\u4ee5\u5728\u76f8\u540c\u5411\u91cf\u7a7a\u95f4\u8868\u793a\u548c\u7406\u89e394\u79cd\u8bed\u8a00\uff0c\u63d0\u5347\u6240\u6709\u8bed\u8a00\u7684\u4ea7\u54c1\u4f53\u9a8c\u3002\u672c\u6587\u7f16\u8bd1\u81ea\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u526f\u9662\u957f\u5468\u660e\u548c\u5fae\u8f6f\u526f\u603b\u88c1\u3001\u6770\u51fa\u5de5\u7a0b\u5e08 Saurabh Tiwary \u5171\u540c\u53d1\u8868\u7684\u5fae\u8f6f\u7814\u7a76\u9662\u535a\u5ba2\u6587\u7ae0 \u201cMicrosoft Turing Universal Language Representation model, T-ULRv2, tops XTREME leaderboard\u201d\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u8fd1\u65e5\uff0c\u7531\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8054\u5408\u5f00\u53d1\u7684\u56fe\u7075\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b T-ULRv2 \u8363\u81ba XTREME \u516c\u5171\u6392\u884c\u699c\u51a0\u519b\uff0c\u8be5\u6a21\u578b\u6bd4\u4e4b\u524d\u6700\u597d\u7684\u6a21\u578b VECO \u6709\u4e863.5\u7684\u5e73\u5747\u5206\u63d0\u5347\u3002\u5728\u6b64\u6b21\u7a81\u7834\u4e2d\uff0c\u9664\u4e86\u9884\u8bad\u7ec3\u7684\u6a21\u578b\u5916\uff0c\u5fae\u8f6f\u7684\u7814\u7a76\u5458\u4eec\u8fd8\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e\u7a33\u5b9a\u6027\u8bad\u7ec3\u7684\u65b0\u578b\u591a\u8bed\u8a00\u5fae\u8c03\u6280\u672f StableTune\u3002\u5165\u56f4\u672c\u6b21\u6392\u884c\u699c\u7684\u5176\u4ed6\u6a21\u578b\u8fd8\u5305\u62ec\uff1aXLM-R\u3001mBERT \u548c XLM \u7b49\u3002\u6b64\u524d\u7684\u6700\u597d\u6a21\u578b FILTER \u4e5f\u662f\u7531\u5fae\u8f6f\u56e2\u961f\u63d0\u4ea4\u7684\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"638\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-1.jpg\" alt=\"table\" class=\"wp-image-1064541\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-1.jpg 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-1-300x187.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-1-768x479.jpg 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-1-240x150.jpg 240w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578bt-ulrv2-\u652f\u630194\u79cd\u8bed\u8a00\">\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578bT-ULRv2\uff1a\u652f\u630194\u79cd\u8bed\u8a00<\/h2>\n\n\n\n<p>\u957f\u671f\u4ee5\u6765\uff0c\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u575a\u4fe1\u6587\u672c\u7684\u8bed\u4e49\u8868\u793a\u662f\u8bed\u8a00\u65e0\u5173\u7684\u3002\u57282018\u5e74\u53d1\u8868\u7684\u8bba\u6587 Towards Language Agnostic Universal Representations\u4e2d\uff08https:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/towards-language-agnostic-universal-representations\/\uff09\uff0c\u5fae\u8f6f\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b9\u6cd5\uff0c\u4ee5\u65e0\u76d1\u7763\u65b9\u5f0f\u8bad\u7ec3\u8bed\u8a00\u65e0\u5173\u7684\u8bed\u4e49\u8868\u793a\u3002\u6b64\u65b9\u6cd5\u53ef\u4ee5\u8ba9\u7ecf\u8fc7\u8bad\u7ec3\u7684\u6a21\u578b\u5728\u9488\u5bf9\u67d0\u4e00\u4e2a\u8bed\u8a00\u52a0\u4ee5\u5fae\u8c03\u540e\uff0c\u7acb\u5373\u5e94\u7528\u4e8e\u53e6\u4e00\u4e2a\u8bed\u8a00\u3002\u4ee5\u5f80\u9488\u5bf9\u6bcf\u4e2a\u8bed\u8a00\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\u90fd\u8981\u6c42\u5177\u5907\u8be5\u8bed\u8a00\u7684\u6807\u6ce8\u6570\u636e\uff0c\u800c\u6b64\u65b9\u6cd5\u6709\u6548\u5730\u89e3\u51b3\u4e86\u8fd9\u4e00\u96be\u9898\u3002\u65e0\u76d1\u7763\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u5df2\u6210\u4e3a\u6240\u6709 NLP \u6a21\u578b\u7684\u57fa\u7840\uff0c\u57fa\u4e8e Transformer \u7684\u6a21\u578b\u5219\u662f\u6240\u6709\u6b64\u7c7b\u521b\u65b0\u7684\u6838\u5fc3\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"576\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-1024x576.jpg\" alt=\"text, whiteboard\" class=\"wp-image-1064544\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-1024x576.jpg 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-300x169.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-768x432.jpg 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-1066x600.jpg 1066w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-655x368.jpg 655w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-240x135.jpg 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-640x360.jpg 640w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2-960x540.jpg 960w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-2.jpg 1247w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u4f5c\u4e3a\u5fae\u8f6f AI at Scale\uff08AI \u89c4\u6a21\u5316\uff09\u7814\u53d1\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\u4e4b\u4e00\uff0c\u56fe\u7075\u7cfb\u5217\u7684 NLP \u6a21\u578b\u4e00\u76f4\u5728\u4e3a\u5fae\u8f6f\u4ea7\u54c1\u7684\u4e0b\u4e00\u4ee3 AI \u4f53\u9a8c\u63d0\u4f9b\u539f\u52a8\u529b\u3002T-ULRv2 \u662f\u8de8\u8bed\u8a00\u7814\u7a76\u7684\u6700\u65b0\u6210\u679c\uff0c\u5b83\u878d\u5408\u4e86\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8fd1\u671f\u5728 InfoXLM \u8bba\u6587\uff08\u70b9\u51fb\u9605\u8bfb\u539f\u6587\uff0c\u4e86\u89e3\u8bba\u6587\u8be6\u60c5\uff09\u4e2d\u7684\u521b\u65b0\uff0c\u5176\u6240\u5f00\u53d1\u7684\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b\u53ef\u4ee5\u7528\u4e8e94\u79cd\u8bed\u8a00\u7684\u6587\u672c\u7684\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\u3002\u901a\u8fc7 T-ULR \u53ef\u4ee5\u5c06\u5fae\u8f6f\u5fc5\u5e94\u7684\u667a\u80fd\u95ee\u9898\u89e3\u7b54\u670d\u52a1\u6269\u5c55\u5230\u6240\u6709\u53d7\u652f\u6301\u7684\u8bed\u8a00\u548c\u5730\u533a\u3002\u800c\u4e14\u8fd9\u4e2a\u6a21\u578b\u8fd8\u88ab\u7528\u4e8e\u5c06 Word \u8bed\u4e49\u641c\u7d22\u529f\u80fd\u6269\u5c55\u5230\u82f1\u8bed\u4e4b\u5916\u7684\u8bed\u8a00\uff0c\u5e76\u4e14\u5b9e\u73b0\u5bf9\u5fae\u8f6f Outlook \u548c\u5fae\u8f6f Teams \u4e2d\u201c\u5efa\u8bae\u7b54\u590d\u201d\u529f\u80fd\u5730\u666e\u904d\u652f\u6301\u3002\u8fd9\u4e9b\u529f\u80fd\u7684\u4f53\u9a8c\u5c06\u4f1a\u5f88\u5feb\u4e0e\u7528\u6237\u89c1\u9762\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"286\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-3.jpg\" alt=\"\u7531 T-ULR \u9a71\u52a8\u7684\u897f\u73ed\u7259\u6587\u53ca\u963f\u62c9\u4f2f\u6587\u7248\u5fae\u8f6f\u5fc5\u5e94\u667a\u80fd\u95ee\u9898\u89e3\u7b54\u793a\u4f8b\" class=\"wp-image-1064547\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-3.jpg 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-3-300x84.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-3-768x215.jpg 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-3-240x67.jpg 240w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u7531 T-ULR \u9a71\u52a8\u7684\u897f\u73ed\u7259\u6587\u53ca\u963f\u62c9\u4f2f\u6587\u7248\u5fae\u8f6f\u5fc5\u5e94\u667a\u80fd\u95ee\u9898\u89e3\u7b54\u793a\u4f8b<\/em><\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"616\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4-1024x616.jpg\" alt=\"graphical user interface, text, application, Word\" class=\"wp-image-1064550\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4-1024x616.jpg 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4-300x180.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4-768x462.jpg 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4-240x144.jpg 240w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-4.jpg 1172w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\"><em>\u7531 T-ULR \u9a71\u52a8\u7684\u6cd5\u6587\u7248\u5fae\u8f6f Word \u8bed\u4e49\u641c\u7d22\u793a\u4f8b<\/em><\/figcaption><\/figure>\n\n\n\n<p>\u8fd9\u4e9b\u771f\u5b9e\u7684\u4ea7\u54c1\u573a\u666f\u5bf9\u7b97\u6cd5\u8d28\u91cf\u7684\u8981\u6c42\u6781\u9ad8\uff0c\u56e0\u6b64\u4e5f\u4e3a\u7814\u7a76\u5458\u4eec\u7684 AI \u6a21\u578b\u63d0\u4f9b\u4e86\u7406\u60f3\u7684\u6d4b\u8bd5\u5e73\u53f0\u3002\u6839\u636e\u7ed3\u679c\u663e\u793a\uff0cT-ULR \u5728\u591a\u8bed\u8a00 NLP \u4efb\u52a1\u4e2d\u7684\u51c6\u786e\u7387\u548c\u6027\u80fd\u90fd\u8fbe\u5230\u4e86\u975e\u5e38\u53ef\u89c2\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"xtreme\u57fa\u51c6\u6570\u636e\u96c6\">XTREME\u57fa\u51c6\u6570\u636e\u96c6<\/h2>\n\n\n\n<p>\u591a\u8bed\u8a00\u7f16\u7801\u5668\u7684\u8de8\u8bed\u8a00\u8fc1\u79fb\u80fd\u529b\u8bc4\u4f30 XTREME \u57fa\u51c6\u6db5\u76d6\u4e86\u6a2a\u8de812\u4e2a\u8bed\u7cfb\u768440\u79cd\u8bed\u8a00\uff0c\u5176\u4e2d\u5305\u542b\u76849\u9879\u4efb\u52a1\u90fd\u9700\u8981\u5bf9\u4e0d\u540c\u5c42\u7ea7\u7684\u8bed\u6cd5\u6216\u8bed\u4e49\u8fdb\u884c\u63a8\u7406\u3002XTREME \u4e2d\u7684\u8bed\u8a00\u9009\u62e9\u65e8\u5728\u6700\u5927\u7a0b\u5ea6\u5730\u63d0\u9ad8\u8bed\u8a00\u591a\u6837\u6027\u3001\u8986\u76d6\u73b0\u6709\u7684\u4efb\u52a1\u8303\u56f4\uff0c\u5e76\u8003\u8651\u5230\u8bad\u7ec3\u6570\u636e\u7684\u53ef\u7528\u6027\u3002<\/p>\n\n\n\n<p>XTREME \u4e2d\u5305\u542b\u7684\u4efb\u52a1\u6db5\u76d6\u4e86\u5305\u62ec\u8bed\u53e5\u6587\u672c\u5206\u7c7b\u3001\u7ed3\u6784\u5316\u9884\u6d4b\u3001\u8bed\u53e5\u68c0\u7d22\u548c\u8de8\u8bed\u8a00\u95ee\u7b54\u7b49\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\u3002\u56e0\u6b64\uff0c\u4e3a\u4e86\u6210\u529f\u5b8c\u6210 XTREME \u57fa\u51c6\u6d4b\u8bd5\uff0c\u6a21\u578b\u5fc5\u987b\u5b66\u4f1a\u53ef\u4ee5\u8de8\u8bed\u8a00\u8fc1\u79fb\u7684\u8868\u793a\u3002<\/p>\n\n\n\n<p>\u6709\u5173\u57fa\u51c6\u6d4b\u8bd5\u548c\u4efb\u52a1\u7684\u5b8c\u6574\u8bf4\u660e\uff0c\u8bf7\u53c2\u89c1\u300aXTREME\uff1aA Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalization\u300b(https:\/\/arxiv.org\/abs\/2003.11080) \u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"t-ulrv2\u9884\u8bad\u7ec3\u7684-\u94c1\u4eba\u4e09\u9879-mmlm-tlm\u548cxlco\">T-ULRv2\u9884\u8bad\u7ec3\u7684\u201c\u94c1\u4eba\u4e09\u9879\u201d\uff1aMMLM\u3001TLM\u548cXLCo<\/h2>\n\n\n\n<p>T-ULRv2 \u91c7\u7528\u7684\u662f24\u5c42\u7684 Transformer \u67b6\u6784\uff0c\u53c2\u6570\u603b\u91cf\u591a\u8fbe5.5\u4ebf\u4e2a\u3002T-ULRv2 \u9884\u8bad\u7ec3\u8981\u5b8c\u6210\u4e09\u9879\u4e0d\u540c\u7684\u4efb\u52a1\uff1a\u591a\u8bed\u8a00\u63a9\u7801\u8bed\u8a00\u6a21\u578b\uff08MMLM\uff09\u3001\u7ffb\u8bd1\u8bed\u8a00\u6a21\u578b\uff08TLM\uff09\u548c\u5fae\u8f6f\u7814\u7a76\u56e2\u961f\u63d0\u51fa\u7684\u8de8\u8bed\u8a00\u5bf9\u6bd4\u5b66\u4e60\uff08XLCo\uff09\u3002MMLM \u4efb\u52a1\uff08\u4e5f\u79f0\u4e3a\u5b8c\u5f62\u586b\u7a7a\u4efb\u52a1\uff09\u7684\u76ee\u6807\u662f\u6839\u636e\u8f93\u5165\u7684\u4e0d\u540c\u8bed\u8a00\u4fe1\u606f\u9884\u6d4b\u51fa\u88ab\u906e\u853d\u7684\u5b57\u7b26\u3002T-ULRv2 \u4f7f\u7528\u4e86\u6765\u81ea\u7f51\u7edc\u7684\u591a\u8bed\u8a00\u6570\u636e\u8bed\u6599\u5e93\uff0c\u8be5\u8bed\u6599\u5e93\u753194\u79cd\u8bed\u8a00\u7ec4\u6210\uff0c\u7528\u4e8e MMLM \u4efb\u52a1\u8bad\u7ec3\u3002<\/p>\n\n\n\n<p>\u4e0e MMLM \u76f8\u4f3c\uff0cTLM \u4efb\u52a1\u4e5f\u662f\u9884\u6d4b\u88ab\u906e\u853d\u7684\u5b57\u7b26\uff0c\u4f46\u8be5\u9884\u6d4b\u662f\u9488\u5bf9\u62fc\u63a5\u5728\u4e00\u8d77\u7684\u7684\u5e73\u884c\u53e5\u5bf9\u3002\u4f8b\u5982\uff0c\u7ed9\u5b9a\u4e00\u5bf9\u82f1\u8bed\u548c\u6cd5\u8bed\u7684\u5e73\u884c\u53e5\u5b50\uff0c\u6a21\u578b\u53ef\u4ee5\u901a\u8fc7\u89c2\u5bdf\u88ab\u906e\u853d\u82f1\u8bed\u5b57\u7b26\u5468\u8fb9\u7684\u82f1\u8bed\u5b57\u7b26\u6216\u8005\u5b83\u7684\u6cd5\u8bed\u7ffb\u8bd1\u7248\u6765\u9884\u6d4b\u8fd9\u4e2a\u88ab\u906e\u853d\u7684\u82f1\u8bed\u5b57\u7b26\u3002\u5b83\u80fd\u591f\u5e2e\u52a9\u6a21\u578b\u201c\u5bf9\u9f50\u201d\u4e0d\u540c\u8bed\u8a00\u7684\u8868\u793a\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"584\" height=\"322\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-5.jpg\" alt=\"\u906e\u853d\u8bed\u8a00\u5efa\u6a21\uff08MLM\uff09\u7684\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u4ee5\u53ca TLM \u4efb\u52a1\uff08\u6765\u6e90\uff1aXLM\uff09\" class=\"wp-image-1064553\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-5.jpg 584w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-5-300x165.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-5-240x132.jpg 240w\" sizes=\"auto, (max-width: 584px) 100vw, 584px\" \/><figcaption class=\"wp-element-caption\"><em>\u906e\u853d\u8bed\u8a00\u5efa\u6a21\uff08MLM\uff09\u7684\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u4ee5\u53ca TLM \u4efb\u52a1\uff08\u6765\u6e90\uff1aXLM\uff09<\/em><\/figcaption><\/figure>\n\n\n\n<p>XLCo \u4efb\u52a1\u5219\u53ef\u4ee5\u4f7f\u7528\u5e73\u884c\u8bad\u7ec3\u6570\u636e\uff0c\u5176\u76ee\u7684\u662f\u8ba9\u5e73\u884c\u53e5\u5b50\u8868\u793a\u4e4b\u95f4\u7684\u4e92\u4fe1\u606f\u6700\u5927\u5316\u3002MMLM \u548c TLM \u7684\u76ee\u6807\u662f\u4f7f\u5b57\u7b26-\u5e8f\u5217\u4e92\u4fe1\u606f\u6700\u5927\u5316\uff0c\u800c XLCo \u7684\u76ee\u6807\u662f\u5e8f\u5217\u7ea7\u522b\u7684\u8de8\u8bed\u8a00\u4e92\u4fe1\u606f\u6700\u5927\u5316\u3002\u5bf9\u4e8e TLM \u548c XLCo \u4efb\u52a1\uff0cT-ULRv2 \u4f1a\u7528\u5230\u542b\u670914\u4e2a\u8bed\u8a00\u5bf9\u7684\u5e73\u884c\u6570\u636e\u3002XLCo \u7684\u635f\u5931\u51fd\u6570\u5982\u4e0b\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"768\" height=\"178\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-6.jpg\" alt=\"formular\" class=\"wp-image-1064556\" style=\"width:285px;height:auto\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-6.jpg 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-6-300x70.jpg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-6-240x56.jpg 240w\" sizes=\"auto, (max-width: 768px) 100vw, 768px\" \/><\/figure>\n\n\n\n<p>\u968f\u540e\uff0c\u5b83\u5c06\u8ba1\u5165 MMLM \u548c TLM \u7684\u635f\u5931\uff0c\u5f97\u51fa\u8de8\u8bed\u8a00\u9884\u8bad\u7ec3\u7684\u603b\u4f53\u635f\u5931\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"274\" height=\"50\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-7.jpg\" alt=\"formular\" class=\"wp-image-1064559\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-7.jpg 274w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/t-ulrv2-7-240x44.jpg 240w\" sizes=\"auto, (max-width: 274px) 100vw, 274px\" \/><\/figure>\n\n\n\n<p>\u6709\u5173 T-ULRv2 \u7684\u66f4\u591a\u4fe1\u606f\uff0c\u8bf7\u53c2\u89c1\u8bba\u6587\u300aInfoXLM: An Information-Theoretic Framework for Cross-Lingual Language Model Pre-Training\u300b(https:\/\/arxiv.org\/abs\/2007.07834) \u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u6784\u5efa\u5b9a\u5236\u5316\u5e94\u7528-t-ulrv2\u5c06\u53d1\u6325\u91cd\u8981\u4f5c\u7528\">\u6784\u5efa\u5b9a\u5236\u5316\u5e94\u7528\uff1aT-ULRv2\u5c06\u53d1\u6325\u91cd\u8981\u4f5c\u7528<\/h2>\n\n\n\n<p>\u5728\u5fae\u8f6f Ignite 2020 \u5927\u4f1a\u4e0a\uff0c\u5fae\u8f6f\u5ba3\u5e03\u56fe\u7075\u6a21\u578b\u5c06\u4f5c\u4e3a\u79c1\u4eba\u9884\u89c8\u7684\u4e00\u90e8\u5206\uff0c\u7528\u4e8e\u6784\u5efa\u5b9a\u5236\u5316\u5e94\u7528\uff0c\u800c T-ULRv2 \u4e5f\u5c06\u6210\u4e3a\u8be5\u8ba1\u5212\u7684\u4e00\u90e8\u5206\u3002\u76ee\u524d\uff0c\u7814\u7a76\u56e2\u961f\u6b63\u5728\u4e0e\u5fae\u8f6f Azure \u8ba4\u77e5\u670d\u52a1\u90e8\u95e8\u7d27\u5bc6\u5408\u4f5c\uff0c\u4e3a\u5f53\u524d\u53ca\u672a\u6765\u5e26\u6709\u56fe\u7075\u6a21\u578b\u7684\u8bed\u8a00\u670d\u52a1\u63d0\u4f9b\u652f\u6301\u3002\u5fae\u8f6f Azure \u8ba4\u77e5\u670d\u52a1\u7684\u73b0\u6709\u5ba2\u6237\u5c06\u901a\u8fc7 API \u81ea\u52a8\u4eab\u53d7\u8fd9\u4e9b\u5347\u7ea7\u670d\u52a1\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai\u4f53\u9a8c\u666e\u53ca\u5316-\u4ee5\u6539\u5584\u6240\u6709\u8bed\u8a00\u7684\u4ea7\u54c1\u4f53\u9a8c\u4e3a\u5148\">AI\u4f53\u9a8c\u666e\u53ca\u5316\uff1a\u4ee5\u6539\u5584\u6240\u6709\u8bed\u8a00\u7684\u4ea7\u54c1\u4f53\u9a8c\u4e3a\u5148<\/h2>\n\n\n\n<p>\u5fae\u8f6f\u8ba4\u4e3a\uff0c\u5168\u7403\u5316\u53d1\u5c55\u4e0d\u4ec5\u4ec5\u662f\u4e2a\u7814\u7a76\u8bfe\u9898\uff0c\u4e5f\u662f\u4e00\u9879\u5fc5\u987b\u76f4\u9762\u7684\u6311\u6218\u3002\u76ee\u524d Windows \u5728\u5168\u4e16\u754c\u8303\u56f4\u5185\u88ab\u4f7f\u7528\uff0c\u5fae\u8f6f Office \u548c\u5fae\u8f6f\u5fc5\u5e94\u4e5f\u5df2\u5728200\u4e2a\u5730\u533a\u4f7f\u7528\uff0c\u4e14\u63d0\u4f9b\u4e86\u8d85\u8fc7100\u4e2a\u8bed\u8a00\u7248\u672c\u3002\u5fae\u8f6f\u7684\u5ba2\u6237\u904d\u53ca\u5168\u7403\u6bcf\u4e2a\u89d2\u843d\uff0c\u4ed6\u4eec\u90fd\u5728\u4ee5\u5404\u81ea\u7684\u6bcd\u8bed\u4f7f\u7528\u7740\u5fae\u8f6f\u7684\u4ea7\u54c1\u3002\u5f53\u4e0b\uff0c\u5fae\u8f6f\u7684\u7814\u7a76\u5458\u4eec\u6b63\u5728\u79ef\u6781\u5730\u62d3\u5c55\u591a\u8bed\u8a00\u6a21\u578b\u7684\u8fb9\u754c\uff0c\u50cf\u7c7b\u4f3c\u4e8e T-ULRv2 \u8fd9\u6837\u65e0\u8bed\u8a00\u5dee\u5f02\u7684\u8868\u793a\u4fbf\u662f\u62d3\u5c55\u4e4b\u8def\u4e0a\u7684\u53c8\u4e00\u6b21\u7a81\u7834\u3002\u5b83\u5c06\u53ef\u4ee5\u63d0\u5347\u6240\u6709\u8bed\u8a00\u7684\u4ea7\u54c1\u4f53\u9a8c\uff0c\u8d4b\u80fd\u6240\u6709\u7528\u6237\uff0c\u771f\u6b63\u5b9e\u73b0 AI \u4f53\u9a8c\u7684\u666e\u53ca\u5316\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7f16\u8005\u6309\uff1a\u4e3a\u8fdb\u4e00\u6b65\u5b9e\u73b0\u7528 AI \u8d4b\u80fd\u7528\u6237\u4f53\u9a8c\uff0c\u5fae\u8f6f\u6b63\u5728\u4e0d\u65ad\u62d3\u5c55\u591a\u8bed\u8a00\u6a21\u578b\u7684\u8fb9\u754c\u3002\u8fd1\u65e5\uff0c\u7531\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8054\u5408\u63a2\u7d22\u7684\u6700\u65b0\u8de8\u8bed\u8a00\u7814\u7a76\u6210\u679c\u2014\u2014\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b T-ULRv2\uff0c\u767b\u9876 XTREME \u6392\u884c\u699c\uff0cT-ULRv2 \u53ef\u4ee5\u5728\u76f8\u540c\u5411\u91cf\u7a7a\u95f4\u8868\u793a\u548c\u7406\u89e394\u79cd\u8bed\u8a00\uff0c\u63d0\u5347\u6240\u6709\u8bed\u8a00\u7684\u4ea7\u54c1\u4f53\u9a8c\u3002\u672c\u6587\u7f16\u8bd1\u81ea\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u526f\u9662\u957f\u5468\u660e\u548c\u5fae\u8f6f\u526f\u603b\u88c1\u3001\u6770\u51fa\u5de5\u7a0b\u5e08 Saurabh Tiwary \u5171\u540c\u53d1\u8868\u7684\u5fae\u8f6f\u7814\u7a76\u9662\u535a\u5ba2\u6587\u7ae0 \u201cMicrosoft Turing Universal Language Representation model, T-ULRv2, tops XTREME leaderboard\u201d\u3002 \u8fd1\u65e5\uff0c\u7531\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u4e0e\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8054\u5408\u5f00\u53d1\u7684\u56fe\u7075\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b T-ULRv2 \u8363\u81ba XTREME \u516c\u5171\u6392\u884c\u699c\u51a0\u519b\uff0c\u8be5\u6a21\u578b\u6bd4\u4e4b\u524d\u6700\u597d\u7684\u6a21\u578b VECO \u6709\u4e863.5\u7684\u5e73\u5747\u5206\u63d0\u5347\u3002\u5728\u6b64\u6b21\u7a81\u7834\u4e2d\uff0c\u9664\u4e86\u9884\u8bad\u7ec3\u7684\u6a21\u578b\u5916\uff0c\u5fae\u8f6f\u7684\u7814\u7a76\u5458\u4eec\u8fd8\u63d0\u51fa\u4e86\u4e00\u79cd\u57fa\u4e8e\u7a33\u5b9a\u6027\u8bad\u7ec3\u7684\u65b0\u578b\u591a\u8bed\u8a00\u5fae\u8c03\u6280\u672f StableTune\u3002\u5165\u56f4\u672c\u6b21\u6392\u884c\u699c\u7684\u5176\u4ed6\u6a21\u578b\u8fd8\u5305\u62ec\uff1aXLM-R\u3001mBERT \u548c XLM \u7b49\u3002\u6b64\u524d\u7684\u6700\u597d\u6a21\u578b FILTER \u4e5f\u662f\u7531\u5fae\u8f6f\u56e2\u961f\u63d0\u4ea4\u7684\u3002 \u957f\u671f\u4ee5\u6765\uff0c\u5fae\u8f6f\u56fe\u7075\u56e2\u961f\u575a\u4fe1\u6587\u672c\u7684\u8bed\u4e49\u8868\u793a\u662f\u8bed\u8a00\u65e0\u5173\u7684\u3002\u57282018\u5e74\u53d1\u8868\u7684\u8bba\u6587 Towards Language Agnostic Universal Representations\u4e2d\uff08https:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/towards-language-agnostic-universal-representations\/\uff09\uff0c\u5fae\u8f6f\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b9\u6cd5\uff0c\u4ee5\u65e0\u76d1\u7763\u65b9\u5f0f\u8bad\u7ec3\u8bed\u8a00\u65e0\u5173\u7684\u8bed\u4e49\u8868\u793a\u3002\u6b64\u65b9\u6cd5\u53ef\u4ee5\u8ba9\u7ecf\u8fc7\u8bad\u7ec3\u7684\u6a21\u578b\u5728\u9488\u5bf9\u67d0\u4e00\u4e2a\u8bed\u8a00\u52a0\u4ee5\u5fae\u8c03\u540e\uff0c\u7acb\u5373\u5e94\u7528\u4e8e\u53e6\u4e00\u4e2a\u8bed\u8a00\u3002\u4ee5\u5f80\u9488\u5bf9\u6bcf\u4e2a\u8bed\u8a00\u8fdb\u884c\u6a21\u578b\u8bad\u7ec3\u90fd\u8981\u6c42\u5177\u5907\u8be5\u8bed\u8a00\u7684\u6807\u6ce8\u6570\u636e\uff0c\u800c\u6b64\u65b9\u6cd5\u6709\u6548\u5730\u89e3\u51b3\u4e86\u8fd9\u4e00\u96be\u9898\u3002\u65e0\u76d1\u7763\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u5df2\u6210\u4e3a\u6240\u6709 NLP \u6a21\u578b\u7684\u57fa\u7840\uff0c\u57fa\u4e8e Transformer \u7684\u6a21\u578b\u5219\u662f\u6240\u6709\u6b64\u7c7b\u521b\u65b0\u7684\u6838\u5fc3\u3002 \u4f5c\u4e3a\u5fae\u8f6f AI at Scale\uff08AI \u89c4\u6a21\u5316\uff09\u7814\u53d1\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\u4e4b\u4e00\uff0c\u56fe\u7075\u7cfb\u5217\u7684 NLP \u6a21\u578b\u4e00\u76f4\u5728\u4e3a\u5fae\u8f6f\u4ea7\u54c1\u7684\u4e0b\u4e00\u4ee3 AI \u4f53\u9a8c\u63d0\u4f9b\u539f\u52a8\u529b\u3002T-ULRv2 \u662f\u8de8\u8bed\u8a00\u7814\u7a76\u7684\u6700\u65b0\u6210\u679c\uff0c\u5b83\u878d\u5408\u4e86\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8fd1\u671f\u5728 InfoXLM \u8bba\u6587\uff08\u70b9\u51fb\u9605\u8bfb\u539f\u6587\uff0c\u4e86\u89e3\u8bba\u6587\u8be6\u60c5\uff09\u4e2d\u7684\u521b\u65b0\uff0c\u5176\u6240\u5f00\u53d1\u7684\u591a\u8bed\u8a00\u9884\u8bad\u7ec3\u6a21\u578b\u53ef\u4ee5\u7528\u4e8e94\u79cd\u8bed\u8a00\u7684\u6587\u672c\u7684\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\u3002\u901a\u8fc7 [&hellip;]<\/p>\n","protected":false},"author":34512,"featured_media":1064562,"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-1064457","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\/1064457","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":2,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1064457\/revisions"}],"predecessor-version":[{"id":1064565,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1064457\/revisions\/1064565"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1064562"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1064457"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1064457"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1064457"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1064457"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}