{"id":3689,"date":"2023-08-21T05:35:00","date_gmt":"2023-08-20T20:35:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=3689"},"modified":"2023-08-21T05:35:00","modified_gmt":"2023-08-20T20:35:00","slug":"causallm-is-not-optimal-for-in-context-learning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=3689","title":{"rendered":"CausalLM is not optimal for in-context learning"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>CausalLM is not optimal for in-context learning\u00a0<\/strong>[21.6]<br>\u6700\u8fd1\u306e\u7d4c\u9a13\u7684\u8a3c\u62e0\u306f\u3001\u30d7\u30ec\u30d5\u30a3\u30c3\u30af\u30b9\u8a00\u8a9e\u30e2\u30c7\u30eb(LM)\u3092\u7528\u3044\u308b\u5834\u5408\u3001\u30b3\u30f3\u30c6\u30af\u30b9\u30c8\u5185\u5b66\u7fd2\u306b\u57fa\u3065\u304f\u30c8\u30e9\u30f3\u30b9\u30d5\u30a9\u30fc\u30de\u30fc\u304c\u3088\u308a\u512a\u308c\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u308b\u3002 \u3053\u306e\u7d50\u679c\u306f\u76f4\u611f\u7684\u3067\u3042\u308b\u304c\u3001\u7406\u8ad6\u7684\u306b\u306f\u7406\u89e3\u3055\u308c\u3066\u3044\u306a\u3044\u3002 \u672c\u7814\u7a76\u3067\u306f,\u3042\u308b\u30d1\u30e9\u30e1\u30fc\u30bf\u69cb\u6210\u306e\u4e0b\u3067,\u30d7\u30ec\u30d5\u30a3\u30c3\u30af\u30b9LM\u3068\u56e0\u679cLM\u306e\u53ce\u675f\u6319\u52d5\u3092\u7406\u8ad6\u7684\u306b\u89e3\u6790\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2308.06912v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2308.06912v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 14 Aug 2023 03:14:38 GMT)<\/li>\n\n\n\n<li>in-context learning\u306e\u7406\u8ad6\u7684\u89e3\u6790\u3068\u5b9f\u8a3c\u5b9f\u9a13\u3001\u300cIn particular, the stationary points of prefixLM coincides with the optimal least square solution; while the ones of causalLM is equivalent to the weights of an online learning system, that is not guaranteed to converge to the optimal solution.\u300d\u3068\u306e\u3053\u3068\u3067\u300cOur experiments verify that causalLM consistently underperforms prefixLM in all settings.\u300d\u3068\u3044\u3046\u7d50\u8ad6<\/li>\n\n\n\n<li>LLM\u5229\u7528\u3067ICL\u306f\u304b\u306a\u308a\u6709\u529b\uff06causalLM\u306e\u6d41\u884c\u3092\u8003\u3048\u308b\u3068\u9762\u767d\u3044\u7d50\u679c\u3002<\/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":[200],"class_list":["post-3689","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-in-context-learning"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/3689","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=3689"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/3689\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3689"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3689"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3689"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}