{"id":5587,"date":"2024-10-14T05:34:00","date_gmt":"2024-10-13T20:34:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5587"},"modified":"2024-10-14T05:34:00","modified_gmt":"2024-10-13T20:34:00","slug":"data-selection-via-optimal-control-for-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5587","title":{"rendered":"Data Selection via Optimal Control for Language Models"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Data Selection via Optimal Control for Language Models\u00a0<\/strong>[134.7]<br>\u672c\u7814\u7a76\u306f,\u5927\u898f\u6a21\u30b3\u30fc\u30d1\u30b9\u304b\u3089\u9ad8\u54c1\u8cea\u306a\u4e8b\u524d\u5b66\u7fd2\u30c7\u30fc\u30bf\u3092\u9078\u629e\u3059\u308b\u3053\u3068\u306b\u3088\u308a,\u4e0b\u6d41\u5229\u7528\u306b\u304a\u3051\u308bLM\u306e\u80fd\u529b\u3092\u5411\u4e0a\u3055\u305b\u308b\u3053\u3068\u3092\u76ee\u7684\u3068\u3059\u308b\u3002 PMP\u6761\u4ef6\u3092\u89e3\u304f\u3053\u3068\u3067\u6700\u9069\u306a\u30c7\u30fc\u30bf\u9078\u629e\u3092\u8fd1\u4f3c\u3059\u308b\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308bPMP\u30d9\u30fc\u30b9\u306e\u30c7\u30fc\u30bf\u9078\u629e(PDS)\u3092\u5c0e\u5165\u3059\u308b\u3002 PDS\u306e\u5229\u70b9\u306f\u3001\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u6cd5\u5247\u306b\u5f93\u3063\u3066\u30c6\u30b9\u30c8\u640d\u5931\u66f2\u7dda\u306e\u5916\u633f\u306b\u3088\u3063\u3066\u8a3c\u660e\u3055\u308c\u305f\u3088\u3046\u306b\u300110T\u30c8\u30fc\u30af\u30f3\u3067\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3055\u308c\u305f400B\u30e2\u30c7\u30eb\u306b\u307e\u3067\u62e1\u5f35\u3055\u308c\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2410.07064v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2410.07064v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 09 Oct 2024 17:06:57 GMT)<\/li>\n\n\n\n<li>\u300cby treating data selection as the control variables (i.e., whether a data point is included in pre-training), the LM pre-training process as the dynamic system, and the LM\u2019s downstream performance as the objective, we leverage Pontryagin\u2019s Maximum Principle (PMP; 63) to derive the necessary conditions for optimal data selection in theory.\u300d\u3068\u3044\u3046\u5236\u5fa1\u7406\u8ad6\u3092\u5fdc\u7528\u3057\u305f\u30c7\u30fc\u30bf\u30bb\u30ec\u30af\u30b7\u30e7\u30f3\u624b\u6cd5\u306e\u63d0\u6848\u3002\u300cThe overhead of running PDS to select data is only about 1\/9 of that of pre-training a 1.7B model.\u300d\u3068\u5b9f\u7528\u7684\u306b\u601d\u3048\u308b\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/thegenerality.com\/agi\/\">Advancing AI for Humanity (thegenerality.com)<\/a>\u3001\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/microsoft\/LMOps\/tree\/main\/data_selection\">LMOps\/data_selection at main \u00b7 microsoft\/LMOps \u00b7 GitHub<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[102,550],"class_list":["post-5587","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-dataset-distillation","tag-550"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5587","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5587"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5587\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5587"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5587"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5587"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}