{"id":5463,"date":"2024-09-17T03:41:00","date_gmt":"2024-09-16T18:41:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5463"},"modified":"2024-09-17T03:41:00","modified_gmt":"2024-09-16T18:41:00","slug":"source2synth-synthetic-data-generation-and-curation-grounded-in-real-data-sources","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5463","title":{"rendered":"Source2Synth: Synthetic Data Generation and Curation Grounded in Real Data Sources\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Source2Synth: Synthetic Data Generation and Curation Grounded in Real Data Sources\u00a0<\/strong>[38.3]<br>\u6211\u3005\u306f\u3001\u30b3\u30b9\u30c8\u306e\u304b\u304b\u308b\u4eba\u7684\u30a2\u30ce\u30c6\u30fc\u30b7\u30e7\u30f3\u306b\u983c\u308b\u3053\u3068\u306a\u304f\u3001LLM\u306b\u65b0\u3057\u3044\u30b9\u30ad\u30eb\u3092\u6559\u3048\u308b\u305f\u3081\u306b\u4f7f\u7528\u3067\u304d\u308b\u65b0\u3057\u3044\u65b9\u6cd5\u3001Source2 Synth\u3092\u63d0\u6848\u3059\u308b\u3002 Source2 Synth\u306f\u30ab\u30b9\u30bf\u30e0\u30c7\u30fc\u30bf\u30bd\u30fc\u30b9\u3092\u5165\u529b\u3068\u3057\u3066\u3001\u5b9f\u4e16\u754c\u306e\u30bd\u30fc\u30b9\u3092\u30d9\u30fc\u30b9\u3068\u3057\u305f\u4e2d\u9593\u7684\u63a8\u8ad6\u30b9\u30c6\u30c3\u30d7\u3092\u5099\u3048\u305f\u5408\u6210\u30c7\u30fc\u30bf\u30dd\u30a4\u30f3\u30c8\u3092\u751f\u6210\u3059\u308b\u3002 \u30de\u30eb\u30c1\u30db\u30c3\u30d7\u8cea\u554f\u5fdc\u7b54(MHQA)\u3068\u30c4\u30fc\u30eb\u8cea\u554f\u5fdc\u7b54(TQA)\u306e\u63a8\u8ad6\u80fd\u529b\u3092\u30c6\u30b9\u30c8\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2409.08239v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2409.08239v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 12 Sep 2024 17:39:08 GMT)<\/li>\n\n\n\n<li>\u300cwe propose Source2Synth, a general approach to generate synthetic data grounded in external real-world sources.\u300d\u3067Dataset generation \u2192 Dataset Curation \u2192 Fine tuning\u306b\u9032\u3080\u306a\u304c\u308c\u3002<\/li>\n\n\n\n<li>\u30ad\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u30d5\u30a7\u30fc\u30ba\u3067\u300cThis is achieved by slicing the dataset in two and using one slice to fine-tune the LLM (LLMSynth).\u300d\u3001\u300cData filtering During filtering, LLMSynth is used to predict the output of the given synthetic example using k tries. If the output cannot be predicted at least once, it is assumed the example is low quality and is not included in the final curated dataset.\u300d\u3068\u306e\u3053\u3068\u3002\u6975\u7aef\u306a\u30c7\u30fc\u30bf\u3060\u3051\u30d5\u30a3\u30eb\u30bf\u3059\u308b\u3088\u3046\u306a\u610f\u56f3\u306a\u306e\u3060\u308d\u3046\u304b\u3002\uff08at least one\u3060\u3068\u554f\u984c\u306a\u3044\u304b\u3082\u3060\u304c\u3001\u95be\u5024\u306b\u3088\u3063\u3066\u30e2\u30c7\u30eb\u5d29\u58ca\u3092\u62db\u304f\u306e\u304b\u3069\u3046\u304b\u306a\u3069\u6c17\u306b\u306a\u308b\u3068\u3053\u308d\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":[356,390],"class_list":["post-5463","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-self-x","tag-synthetic-data"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5463","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=5463"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5463\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5463"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5463"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5463"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}