{"updated":"2025-01-20T06:03:47.815385+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00175956","sets":["581:8417:8429"]},"path":["8429"],"owner":"11","recid":"175956","title":["教師データが不足した環境での機械学習結果改善手法"],"pubdate":{"attribute_name":"公開日","attribute_value":"2016-11-15"},"_buckets":{"deposit":"d2fa1714-889b-4e62-9f7f-d19d0b621d28"},"_deposit":{"id":"175956","pid":{"type":"depid","value":"175956","revision_id":0},"owners":[11],"status":"published","created_by":11},"item_title":"教師データが不足した環境での機械学習結果改善手法","author_link":["368765","368764"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"教師データが不足した環境での機械学習結果改善手法"},{"subitem_title":"Refinement of Machine Learning Results Generated from Insufficient Sample Data","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"[特集：ゲームプログラミング] 教師あり機械学習，ヒューリスティクス，将棋","subitem_subject_scheme":"Other"}]},"item_type_id":"2","publish_date":"2016-11-15","item_2_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"株式会社富士通研究所"}]},"item_2_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Fujitsu Laboratories Ltd.","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/175956/files/IPSJ-JNL5711007.pdf","label":"IPSJ-JNL5711007.pdf"},"date":[{"dateType":"Available","dateValue":"2018-11-15"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-JNL5711007.pdf","filesize":[{"value":"1.1 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"8"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"af5a2d86-be27-421d-a0c2-0edfd43b1e05","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2016 by the Information Processing Society of Japan"}]},"item_2_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"金澤, 裕治"}],"nameIdentifiers":[{}]}]},"item_2_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yuzi, Kanazawa","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_2_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN00116647","subitem_source_identifier_type":"NCID"}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_6501","resourcetype":"journal article"}]},"item_2_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"1882-7764","subitem_source_identifier_type":"ISSN"}]},"item_2_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"厳密解を求めるのが困難でヒューリスティクスによって解かれている問題で，計算機が熟練者を上回ることが困難なものが存在する．そのような問題において，ヒューリスティクス手法を多数のパラメータで制御できるようにしておき，そのパラメータを機械学習によりチューニングすることで，熟練者の判断を再現できれば，解法の性能向上が期待できる．そのために解決しなければならない課題の1つが，教師データの不足である．本論文では，教師データが不足した環境で学習結果に含まれる誤りを改善する強化学習類似手法を提案する．提案手法を将棋プログラムBonanza 6.0の機械学習テーブル改善に適用し，1回の適用でイロレーティングが平均25程度，繰り返し適用することで，最終的には150程度向上した．","subitem_description_type":"Other"}]},"item_2_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"There are some problems where human experts can produce better result than heuristics methods on computers. Performance of such heuristics methods may be improved significantly by machine learning on result by human experts. An issue that must be solved to make it possible is sample data shortage. This paper proposes a reinforcement-learning-like method to fix errors in machine learning result generated from insufficient sample data. The method was applied to refine parameters used by the shogi program Bonanza 6.0. Experimental results show that Elo rating of Bonanza 6.0 with refined parameters was improved by 150 points.","subitem_description_type":"Other"}]},"item_2_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"2391","bibliographic_titles":[{"bibliographic_title":"情報処理学会論文誌"}],"bibliographicPageStart":"2382","bibliographicIssueDates":{"bibliographicIssueDate":"2016-11-15","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"11","bibliographicVolumeNumber":"57"}]},"relation_version_is_last":true,"weko_creator_id":"11"},"created":"2025-01-19T00:45:40.620896+00:00","id":175956,"links":{}}