{"id":7574,"date":"2025-10-17T06:11:00","date_gmt":"2025-10-16T21:11:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7574"},"modified":"2025-10-12T10:19:11","modified_gmt":"2025-10-12T01:19:11","slug":"self-improving-llm-agents-at-test-time","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7574","title":{"rendered":"Self-Improving LLM Agents at Test-Time"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Self-Improving LLM Agents at Test-Time\u00a0<\/strong>[49.9]<br>\u8a00\u8a9e\u30e2\u30c7\u30eb(LM)\u306e1\u3064\u306e\u30d1\u30e9\u30c0\u30a4\u30e0\u306f\u3001\u5927\u898f\u6a21\u306a\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u4f5c\u6210\u306b\u4f9d\u5b58\u3057\u3066\u3044\u308b\u3002 \u5b9f\u969b\u306b\u306f\u3001\u5927\u91cf\u306e\u30c7\u30fc\u30bf\u3092\u96c6\u3081\u308b\u3053\u3068\u306f\u975e\u52b9\u7387\u3067\u3042\u308a\u3001\u305d\u308c\u3089\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306f\u9055\u6cd5\u306b\u9ad8\u4fa1\u3067\u3042\u308b\u3002 \u30c6\u30b9\u30c8\u6642\u9593\u81ea\u5df1\u6539\u5584(TT-SI)\u3068\u30c6\u30b9\u30c8\u6642\u9593\u84b8\u7559(TT-D)\u306e2\u3064\u306e\u30d0\u30ea\u30a8\u30fc\u30b7\u30e7\u30f3\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2510.07841v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2510.07841v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 09 Oct 2025 06:37:35 GMT)<\/li>\n\n\n\n<li>\u300c(i) identify uncertain samples via a novel uncertainty estimator, (ii) generate new training instances similar to these samples, and (iii) update the model online.\u300d\u3068\u3044\u3046\u30b9\u30c6\u30c3\u30d7\u304b\u3089\u306a\u308bself improvement\u3002\u300cTest-Time Self-Improvement (TT-SI), where the model trains on self-generated samples using parameter efficient fine-tuning techniques (PEFT) (Hu et al , 2022), and Test-Time Distillation (TT-D) where adaptation is guided by supervision from samples synthesized by a more capable teacher model.\u300d\u306e2\u7a2e\u985e\u3092\u691c\u8a0e\u3057\u3066\u3044\u308b\uff08\u5f8c\u8005\u306fself-improving\u306a\u306e\u304b\u82e5\u5e72\u7591\u554f\u3067\u306f\u3042\u308b\u304c\u3002\u3002\uff09<\/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":[356],"class_list":["post-7574","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-self-x"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7574","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=7574"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7574\/revisions"}],"predecessor-version":[{"id":7575,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7574\/revisions\/7575"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7574"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7574"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7574"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}