{"created":"2025-01-19T01:42:46.383376+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00239244","sets":["6164:6165:6522:11751"]},"path":["11751"],"owner":"44499","recid":"239244","title":["画像生成モデルの弱点検出タスクへの適用可能性調査"],"pubdate":{"attribute_name":"公開日","attribute_value":"2024-09-10"},"_buckets":{"deposit":"50836358-606f-4f4b-83eb-a016671d8c44"},"_deposit":{"id":"239244","pid":{"type":"depid","value":"239244","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"画像生成モデルの弱点検出タスクへの適用可能性調査","author_link":["655500","655499"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"画像生成モデルの弱点検出タスクへの適用可能性調査"},{"subitem_title":"Investigating the Applicability of Image Generative Models to Weakness Detection Tasks ","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"テスト","subitem_subject_scheme":"Other"}]},"item_type_id":"18","publish_date":"2024-09-10","item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_18_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"富士通株式会社／現在，電気通信大学"},{"subitem_text_value":"国立情報学研究所"}]},"item_18_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Fujitsu Limited / Presently with The University of Electro-Communications","subitem_text_language":"en"},{"subitem_text_value":"National Institute of Informatics","subitem_text_language":"en"}]},"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/239244/files/IPSJ-SES2024016.pdf","label":"IPSJ-SES2024016.pdf"},"date":[{"dateType":"Available","dateValue":"2026-09-10"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-SES2024016.pdf","filesize":[{"value":"554.2 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":"12"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"68a423e0-1eac-4857-b260-fb6cdffe86f6","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2024 by the Information Processing Society of Japan"}]},"item_18_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"横山, 晴樹"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"石川, 冬樹"}],"nameIdentifiers":[{}]}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_5794","resourcetype":"conference paper"}]},"item_18_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"ML モデルの振る舞いの安全性違反は社会的影響が大きいため，実世界での稼働に向けて十分なテストが必要である．既存手法では，組合せテストモデリングの考え方に基づき，データに付随するラベリング可能な特徴を用いて ML モデルの弱点を検出していたが，特に画像データに関しては，実世界からの収集が難しい希少な特徴が含まれることがあり，十分なテストが困難となる課題が生じる．本論文では，この希少性の課題に対して，画像生成モデルを活用するアプローチを検討した．実験では，「生成データはテストデータと比較して因子水準別の弱点をどの程度検出できるのか？」という問いに対し，生成データで検出できた弱点のうち，76%～80% はテストデータでも同様に検出できることを明らかにした．また，学習において Fine Tuning を使用せず Data Augmentation のみを用いたモデルの弱点が特に検出されやすいことを明らかにした．","subitem_description_type":"Other"}]},"item_18_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"71","bibliographic_titles":[{"bibliographic_title":"ソフトウェアエンジニアリングシンポジウム2024論文集"}],"bibliographicPageStart":"63","bibliographicIssueDates":{"bibliographicIssueDate":"2024-09-10","bibliographicIssueDateType":"Issued"},"bibliographicVolumeNumber":"2024"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":239244,"updated":"2025-01-19T08:21:00.729029+00:00","links":{}}