{"updated":"2025-01-19T23:59:17.764920+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00192832","sets":["1164:2240:9411:9646"]},"path":["9646"],"owner":"44499","recid":"192832","title":["GraphCNN向けの疎行列積計算Batch最適化"],"pubdate":{"attribute_name":"公開日","attribute_value":"2018-12-10"},"_buckets":{"deposit":"e27cd651-1b7c-46e7-85ae-c52e797abb7a"},"_deposit":{"id":"192832","pid":{"type":"depid","value":"192832","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"GraphCNN向けの疎行列積計算Batch最適化","author_link":["450987","450986","450984","450985"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"GraphCNN向けの疎行列積計算Batch最適化"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"アクセラレータ","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2018-12-10","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"東京工業大学"},{"subitem_text_value":"東京工業大学"},{"subitem_text_value":"京都大学"},{"subitem_text_value":"(国立研究開発法人)理化学研究所計算科学研究センター／東京工業大学"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Tokyo Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"Tokyo Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"Kyoto University","subitem_text_language":"en"},{"subitem_text_value":"RIKEN Center for Computational Science (R-CCS) / Tokyo Institute of Technology","subitem_text_language":"en"}]},"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/192832/files/IPSJ-HPC18167007.pdf","label":"IPSJ-HPC18167007.pdf"},"date":[{"dateType":"Available","dateValue":"2020-12-10"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-HPC18167007.pdf","filesize":[{"value":"704.5 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":"14"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"2d50549a-c2b2-4c4a-be83-6f9fb7cc1b75","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2018 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"長坂, 侑亮"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"額田, 彰"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"小島, 諒介"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"松岡, 聡"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10463942","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":"2188-8841","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"バイオインフォマティクス等における深層学習的手法の適用として，高い認識精度を得ることが可能である Graph Convolutional Network (GCN) が近年注目を集めている．グラフ構造を持つデータに対する畳込み演算が可能である GCN の処理では，疎行列計算 （SpMM） を含む膨大な演算を処理するために GPU が用いられている．しかしながら，GCN で扱われるデータのグラフ構造にはノード数が数十程度の小さいものが含まれており，小行列に対する SpMM は GPU の並列性の活用が困難であるために，疎行列計算が GCN の学習や推論の性能のボトルネックとなっている．GCN の処理性能向上のために，複数のデータに対する SpMM 計算を一つのカーネルで行うことで GPU の高い並列性と演算能力を活用可能にする Batched SpMM と，GPU のメモリ階層を活用した Batched SpMM Dynamic を提案する．NVIDIA  Tesla P100 GPU を搭載する TSUBAME 3.0 にて評価実験を行い，GCN アプリケーションに Batched 手法を適用することによって学習と推論の双方において高速化を実現し，学習性能は最大 1.64 倍，推論性能は最大 1.38 倍の性能向上を達成した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"9","bibliographic_titles":[{"bibliographic_title":"研究報告ハイパフォーマンスコンピューティング（HPC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2018-12-10","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"7","bibliographicVolumeNumber":"2018-HPC-167"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"created":"2025-01-19T00:58:30.976485+00:00","id":192832,"links":{}}