{"id":231063,"links":{},"created":"2025-01-19T01:31:12.148828+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00231063","sets":["1164:1579:11081:11407"]},"path":["11407"],"owner":"44499","recid":"231063","title":["ベイズ深層学習のドロップアウトパターン事前決定による推論高速化手法の検討"],"pubdate":{"attribute_name":"公開日","attribute_value":"2023-11-28"},"_buckets":{"deposit":"4170cba2-888c-40fb-ac14-080efefff460"},"_deposit":{"id":"231063","pid":{"type":"depid","value":"231063","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"ベイズ深層学習のドロップアウトパターン事前決定による推論高速化手法の検討","author_link":["623210","623212","623211"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"ベイズ深層学習のドロップアウトパターン事前決定による推論高速化手法の検討"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"AIとグラフ","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2023-11-28","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"東京大学"},{"subitem_text_value":"東京大学"},{"subitem_text_value":"東京大学"},{"subitem_text_value":"東京大学"}]},"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/231063/files/IPSJ-ARC23255024.pdf","label":"IPSJ-ARC23255024.pdf"},"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-ARC23255024.pdf","filesize":[{"value":"990.1 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"0","billingrole":"16"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_login","version_id":"a5d5e775-8a5c-4542-9054-35e012624df3","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2023 by the Institute of Electronics, Information and Communication Engineers This SIG report is only available to those in membership of the SIG."}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"菅, 研吾"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"平山, 侑樹"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"金子, 竜也  高前田 伸也"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10096105","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_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8574","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"ベイズ深層学習 (BNN) は，ベイズ統計を利用する機械学習手法であり，予測の不確実性を捉えやすく，過学習を抑えるという利点がある．BNN を実現する方法として，推論時にドロップアウトを伴うフォワーディングを複数行い，得られた計算結果の平均と分散を出力とする手法が知られている．我々はハードウェア上でベイズ深層学習の推論を行うことを見据え，推論時のデータフローを最適化する手法を提案する．提案手法は，学習を行う前にドロップアウトパターンを事前決定し，これを学習と推論の両方に使用する．これにより推論時の乱数発生を不要にし，また低回数の順伝播で高精度な推論を可能にする．さらに，中間層の計算結果を効率的に使い回すため，順伝播のデータフローの構造を学習時に自動探索する．これにより計算回数をさらに削減し，推論プロセスの効率を向上させることを意図する．本稿では，提案手法の有用性を示すために，MNIST データセットを用いた実験を行った．その結果，既存手法と同程度の精度を保ちながら，途中計算に用いられる重みの要素数を平均で約 43% 削減できることが示された．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"6","bibliographic_titles":[{"bibliographic_title":"研究報告システム・アーキテクチャ（ARC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2023-11-28","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"24","bibliographicVolumeNumber":"2023-ARC-255"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"updated":"2025-01-19T10:53:25.287470+00:00"}