{"id":7646,"date":"2025-10-22T04:26:00","date_gmt":"2025-10-21T19:26:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7646"},"modified":"2025-10-18T16:30:50","modified_gmt":"2025-10-18T07:30:50","slug":"detecting-data-contamination-from-reinforcement-learning-post-training-for-large-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7646","title":{"rendered":"Detecting Data Contamination from Reinforcement Learning Post-training for Large Language Models"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Detecting Data Contamination from Reinforcement Learning Post-training for Large Language Models\u00a0<\/strong>[30.3]<br>\u30c7\u30fc\u30bf\u6c5a\u67d3\u306f\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306e\u4fe1\u983c\u6027\u8a55\u4fa1\u306b\u91cd\u5927\u306a\u8105\u5a01\u3092\u3082\u305f\u3089\u3059 \u3053\u306e\u554f\u984c\u306f\u3001\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u30b5\u30f3\u30d7\u30eb\u304c\u5fc5\u7136\u7684\u306b\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30bb\u30c3\u30c8\u306b\u73fe\u308c\u3001\u5831\u544a\u3055\u308c\u305f\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u6709\u52b9\u6027\u3092\u640d\u306a\u3046\u3053\u3068\u306b\u306a\u308b\u3002 \u672c\u7a3f\u3067\u306f,RL\u30dd\u30b9\u30c8\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306e\u305f\u3081\u306e\u7279\u6b8a\u6c5a\u67d3\u691c\u51fa\u624b\u6cd5\u3068\u3057\u3066,\u81ea\u5df1\u6279\u5224\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2510.09259v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2510.09259v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Fri, 10 Oct 2025 10:58:50 GMT)<\/li>\n\n\n\n<li>\u30b3\u30f3\u30bf\u30df\u30cd\u30fc\u30b7\u30e7\u30f3\u691c\u77e5\u306e\u305f\u3081\u306eSelf-Critique\u300c. The method compares token-level entropy sequences between the initial response and the self-critique response. High similarity in entropy space indicates contamination (policy collapse), while low similarity indicates clean samples.\u300d\u304c\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/yongding-tao\/RL-Data-Contamination\">GitHub &#8211; yongding-tao\/RL-Data-Contamination<\/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":[216,356],"class_list":["post-7646","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-leakage","tag-self-x"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7646","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=7646"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7646\/revisions"}],"predecessor-version":[{"id":7647,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7646\/revisions\/7647"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7646"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7646"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7646"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}