{"updated":"2025-01-21T19:13:50.478807+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00081890","sets":["1164:5159:6679:6767"]},"path":["6767"],"owner":"11","recid":"81890","title":["誤差逆伝播を利用した重み付き仮説推論の教師あり学習"],"pubdate":{"attribute_name":"公開日","attribute_value":"2012-05-03"},"_buckets":{"deposit":"b60d5a77-a0a7-4149-9505-816a536e6a9f"},"_deposit":{"id":"81890","pid":{"type":"depid","value":"81890","revision_id":0},"owners":[11],"status":"published","created_by":11},"item_title":"誤差逆伝播を利用した重み付き仮説推論の教師あり学習","author_link":["0","0"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"誤差逆伝播を利用した重み付き仮説推論の教師あり学習"},{"subitem_title":"Backpropagation Learning for Weighted Abduction","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"学習・システム","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2012-05-03","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"東北大学"},{"subitem_text_value":"東北大学"},{"subitem_text_value":"東北大学"},{"subitem_text_value":"東北大学／科学技術振興機構さきがけ"},{"subitem_text_value":"東北大学"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"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/81890/files/IPSJ-SLP12091009.pdf"},"date":[{"dateType":"Available","dateValue":"2014-05-03"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SLP12091009.pdf","filesize":[{"value":"863.0 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"22"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"88136e66-8a58-4308-9658-b1c28ac9b4a6","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2012 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"山本, 風人"},{"creatorName":"井之上, 直也"},{"creatorName":"渡邊, 陽太郎"},{"creatorName":"岡崎, 直観"},{"creatorName":"乾, 健太郎"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Kazeto, Yamamoto","creatorNameLang":"en"},{"creatorName":"Naoya, Inoue","creatorNameLang":"en"},{"creatorName":"Yotaro, Watanabe","creatorNameLang":"en"},{"creatorName":"Naoaki, Okazaki","creatorNameLang":"en"},{"creatorName":"Kentaro, Inui","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10442647","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"本稿では、重み付き仮説推論のパラメタを教師あり学習によって自動調整する手法を提案する。仮説推論は、与えられた観測に対して評価関数を最大化する最良の説明を求める推論であり、自然言語処理において文章に明示されていない情報の顕在化を行うことに対して有用な枠組みとして注目を浴びている。しかしその一方で、仮説推論の評価関数の学習手法が未だ提案されておらず、評価関数のパラメタの調整は、人手による調整やヒューリスティックな手法に頼らざるを得ないという問題があった。そこで我々は、仮説推論の拡張のひとつである重み付き仮説推論を対象として、仮説の証明木におけるリテラル間のコストの関係をフィードフォワードニューラルネットワークの形で表現することで、誤差に対する各パラメタの勾配を求め、評価関数のパラメタの識別学習を実現する。また、提案手法によって評価関数を学習できていることを確かめるために、既存のデータセットを用いて実験した結果についても報告する。","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"We explore a discourse processing framework for discovering implicit information in texts, based on Hobbs et al.'s weighted abduction [5]. Abduction is inference to the best explanation. In weighted abduction, the best explanation is defined as the explanation that minimizes a parametrized cost function. However, less attention has been paid to how to tune the parameters of the cost function automatically. In this paper, we propose a discriminative approach to learning parameters in weighted abduction. We represent the transition of costs in a proof tree as feed-forward neural networks, and calculate the gradients of parameters in a background knowledge base. Our experiments show that our method correctly learns parameters on the existing dataset of plan recognition.","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"8","bibliographic_titles":[{"bibliographic_title":"研究報告音声言語情報処理（SLP）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2012-05-03","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"9","bibliographicVolumeNumber":"2012-SLP-91"}]},"relation_version_is_last":true,"weko_creator_id":"11"},"created":"2025-01-18T23:35:55.402384+00:00","id":81890,"links":{}}