{"id":5502,"date":"2024-09-24T06:23:00","date_gmt":"2024-09-23T21:23:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5502"},"modified":"2024-09-24T06:23:00","modified_gmt":"2024-09-23T21:23:00","slug":"autoregressive-chain-of-thought-cot-%e2%89%83-recurrent%e3%80%81to-cot-or-not-to-cot","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5502","title":{"rendered":"Autoregressive + Chain of Thought (CoT) \u2243 Recurrent\u3001To CoT or not to CoT"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Chain of Thought\u306e\u691c\u8a3c\u3092\u884c\u3063\u305f\u8ad6\u6587\u304c\u51fa\u3066\u3044\u305f\u3002\uff11\u3064\u76ee\u306f\u52d5\u4f5c\u9762\u304b\u3089\u306e\u691c\u8a3c\u30672\u3064\u76ee\u306f\u30e1\u30bf\u5206\u6790\u306b\u3088\u308b\u3082\u306e\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Autoregressive + Chain of Thought (CoT) $\\simeq$ Recurrent: Recurrence&#8217;s Role in Language Models and a Revist of Recurrent Transformer&nbsp;<\/strong>[30.0]<br>\u8a00\u8a9e\u30e2\u30c7\u30eb\u306b\u304a\u3051\u308b\u7e70\u308a\u8fd4\u3057\u69cb\u9020\u304c\u63a8\u8ad6\u80fd\u529b\u306b\u4e0e\u3048\u308b\u5f71\u97ff\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002 \u7dda\u5f62\u5909\u63db\u5668\u3084RWKV\u306e\u3088\u3046\u306a\u30e2\u30c7\u30eb\u306b\u304a\u3051\u308b\u91cd\u8981\u306a\u7406\u8ad6\u7684\u9650\u754c\u3092\u540c\u5b9a\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2409.09239v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2409.09239v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Sat, 14 Sep 2024 00:30:57 GMT)<\/li>\n\n\n\n<li>\u300cWe explained that CoT approximates recurrence in Transformer-based autoregressive LLMs from a computational standpoint.\u300d\u3068\u306e\u3053\u3068\u3002\u9014\u4e2d\u306e\u300cRecurrent Neural Networks (RNNs) sacrifice parallel training for recurrent connections, while Transformers trade recurrence for parallelism.\u300d\u3082\u91cd\u8981\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning&nbsp;<\/strong>[55.5]<br>Chain-of-Thought (CoT) \u306f,\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb (LLM) \u304b\u3089\u63a8\u8ad6\u80fd\u529b\u3092\u5f15\u304d\u51fa\u3059\u30c7\u30d5\u30a1\u30af\u30c8\u624b\u6cd5\u3067\u3042\u308b\u3002 \u79c1\u305f\u3061\u306f\u3001CoT\u304c\u4e3b\u306b\u6570\u5b66\u3084\u8ad6\u7406\u5b66\u3092\u542b\u3080\u30bf\u30b9\u30af\u306b\u5f37\u3044\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u4e0a\u306e\u5229\u70b9\u3092\u3082\u305f\u3089\u3057\u3001\u4ed6\u306e\u30bf\u30b9\u30af\u3088\u308a\u3082\u306f\u308b\u304b\u306b\u5c11\u306a\u3044\u5229\u76ca\u3092\u3082\u305f\u3089\u3059\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u307e\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2409.12183v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2409.12183v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Wed, 18 Sep 2024 17:55:00 GMT)<\/li>\n\n\n\n<li>\u300cFinding 1: CoT only helps substantially on problems requiring mathematical, logical, or algorithmic reasoning.\u300d\u306f\u3088\u3044\u3068\u3057\u3066\u3001\u300cFinding 2: CoT primarily helps with the execution step that performs computation and symbolic manipulation, but falls short of what LLMs with tool augmentation can do.\u300d\u306fAgentic\u306a\u30a2\u30d7\u30ed\u30fc\u30c1\u306e\u307b\u3046\u304c\u6709\u671b\u3068\u3044\u3046\u3053\u3068\u306a\u3093\u3060\u308d\u3046\u304b\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Chain of Thought\u306e\u691c\u8a3c\u3092\u884c\u3063\u305f\u8ad6\u6587\u304c\u51fa\u3066\u3044\u305f\u3002\uff11\u3064\u76ee\u306f\u52d5\u4f5c\u9762\u304b\u3089\u306e\u691c\u8a3c\u30672\u3064\u76ee\u306f\u30e1\u30bf\u5206\u6790\u306b\u3088\u308b\u3082\u306e\u3002<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[59],"class_list":["post-5502","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-chain-of-thought"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5502","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=5502"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5502\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5502"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5502"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5502"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}