{"created":"2025-01-19T01:27:56.862116+00:00","updated":"2025-01-19T11:42:38.389104+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00228811","sets":["6164:6165:6462:11379"]},"path":["11379"],"owner":"44499","recid":"228811","title":["人種間の公平性を考慮した顔認証距離学習"],"pubdate":{"attribute_name":"公開日","attribute_value":"2023-10-23"},"_buckets":{"deposit":"780a5c78-76ad-4c33-9f89-a669c7b32e44"},"_deposit":{"id":"228811","pid":{"type":"depid","value":"228811","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"人種間の公平性を考慮した顔認証距離学習","author_link":["614121","614119","614118","614123","614117","614122","614120","614124"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"人種間の公平性を考慮した顔認証距離学習"},{"subitem_title":"Metric Learning for Facial Recognition Considering Racial Fairness","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"SNS，テキスト分類","subitem_subject_scheme":"Other"}]},"item_type_id":"18","publish_date":"2023-10-23","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":"静岡大学"},{"subitem_text_value":"東北大学"},{"subitem_text_value":"静岡大学"},{"subitem_text_value":"静岡大学 ／理化学研究所 革新知能統合研究センター"}]},"item_18_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Shizuoka University","subitem_text_language":"en"},{"subitem_text_value":"Tohoku University","subitem_text_language":"en"},{"subitem_text_value":"Shizuoka University","subitem_text_language":"en"},{"subitem_text_value":"Shizuoka University / RIKEN AIP","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/228811/files/IPSJ-CSS2023198.pdf","label":"IPSJ-CSS2023198.pdf"},"date":[{"dateType":"Available","dateValue":"2025-10-23"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-CSS2023198.pdf","filesize":[{"value":"381.9 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":"30"},{"tax":["include_tax"],"price":"0","billingrole":"46"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"fa38b0da-ef4a-4d93-8672-63c97ead3a76","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2023 by the Information Processing Society of Japan"}]},"item_18_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"佐藤, 佑哉"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"伊藤, 康一"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"西垣, 正勝"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"大木, 哲史"}],"nameIdentifiers":[{}]}]},"item_18_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yuya, Sato","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Koichi, Ito","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Masakatsu, Nishigaki","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Tetsushi, Ohki","creatorNameLang":"en"}],"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":"学習データの人種割合が偏った顔画像データセットで学習された顔認証モデルは，人種によって認証精度が異なってしまい，公平性の観点から問題になることがある．一方，顔認証モデルを学習する際に広く使用される大規模な顔画像データセットは，インターネット上で自動的に収集しており，人種割合が偏っていることが指摘されている．顔画像データセットの人種割合を統一しながら大規模な顔画像データセットを構築することは難しく，アンダーサンプリングによって人種割合を統一した場合は，学習データ数の減少により認証精度の低下につながる．そこで，本稿では顔画像データセットの人種割合が偏ったまま，顔認証モデルにおける公平性のバランスを調整可能な新たなモデル学習法を提案する．提案手法は，損失関数のパラメータを学習段階や公平性を考慮しながら動的に変化させることで，学習を安定させながら公平性を向上させる．さらに，高精度な人種の精度低下の許容度を決定する調和パラメータによって，公平性と認証精度の重要度の比重に応じて最適なバランスで学習を行うことができる．","subitem_description_type":"Other"}]},"item_18_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Face recognition models trained on facial image datasets with a skewed racial distribution often display differing authentication accuracies depending on race, which can pose problems from a fairness standpoint. On the other hand, large-scale facial image datasets widely used for training face recognition models\nare automatically collected from the internet, and a racial imbalance in these datasets has been pointed out. Constructing a large-scale facial image dataset with a uniform racial distribution is challenging, and if the racial distribution is uniformed by undersampling, a reduction in the number of training data leads\nto a decrease in authentication accuracy. Therefore, this paper proposes a new model learning method that can adjust the balance of fairness in face recognition models, even with a skewed racial distribution in the facial image dataset. The proposed method dynamically changes the parameters of the loss function during\nthe learning stage and in consideration of fairness, thereby stabilizing the learning process while improving fairness. Furthermore, by using a harmony parameter that determines the allowable degree of accuracy degradation for highly accurate races, it is possible to perform learning at an optimal balance according to\nthe relative importance of fairness and authentication accuracy.","subitem_description_type":"Other"}]},"item_18_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"1464","bibliographic_titles":[{"bibliographic_title":"コンピュータセキュリティシンポジウム2023論文集"}],"bibliographicPageStart":"1459","bibliographicIssueDates":{"bibliographicIssueDate":"2023-10-23","bibliographicIssueDateType":"Issued"}}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":228811,"links":{}}