{"updated":"2025-01-19T22:05:46.405563+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00198145","sets":["1164:2240:9748:9861"]},"path":["9861"],"owner":"44499","recid":"198145","title":["Toward Training a Large 3D Cosmological CNN with Hybrid Parallelization"],"pubdate":{"attribute_name":"公開日","attribute_value":"2019-07-17"},"_buckets":{"deposit":"b4ba0c20-5ea5-4f3f-977b-0a81eb007c14"},"_deposit":{"id":"198145","pid":{"type":"depid","value":"198145","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"Toward Training a Large 3D Cosmological CNN with Hybrid Parallelization","author_link":["476667","476660","476659","476664","476665","476668","476662","476672","476669","476661","476657","476658","476663","476670","476671","476656","476673","476666"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"Toward Training a Large 3D Cosmological CNN with Hybrid Parallelization"},{"subitem_title":"Toward Training a Large 3D Cosmological CNN with Hybrid Parallelization","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"機械学習","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2019-07-17","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"Tokyo Institute of Technology／Lawrence Livermore National Laboratory"},{"subitem_text_value":"Lawrence Livermore National Laboratory"},{"subitem_text_value":"University of Illinois at Urbana-Champaign／Lawrence Livermore National Laboratory"},{"subitem_text_value":"Lawrence Berkeley National Laboratory"},{"subitem_text_value":"Lawrence Berkeley National Laboratory"},{"subitem_text_value":"RIKEN Center for Computational Science／Tokyo Institute of Technology"},{"subitem_text_value":"University of Illinois at Urbana-Champaign"},{"subitem_text_value":"Lawrence Berkeley National Laboratory"},{"subitem_text_value":"Lawrence Livermore National Laboratory"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Tokyo Institute of Technology / Lawrence Livermore National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"Lawrence Livermore National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"University of Illinois at Urbana-Champaign / Lawrence Livermore National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"Lawrence Berkeley National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"Lawrence Berkeley National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"RIKEN Center for Computational Science / Tokyo Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"University of Illinois at Urbana-Champaign","subitem_text_language":"en"},{"subitem_text_value":"Lawrence Berkeley National Laboratory","subitem_text_language":"en"},{"subitem_text_value":"Lawrence Livermore National Laboratory","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"eng"}]},"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/198145/files/IPSJ-HPC19170008.pdf","label":"IPSJ-HPC19170008.pdf"},"date":[{"dateType":"Available","dateValue":"2021-07-17"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-HPC19170008.pdf","filesize":[{"value":"678.3 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":"28907e5c-f1a8-43b6-8296-b9c61511830a","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2019 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yosuke, Oyama"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Naoya, Maruyama"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Nikoli, Dryden"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Peter, Harrington"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Jan, Balewski"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Satoshi, Matsuoka"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Marc, Snir"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Peter, Nugent"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Brian, Van Essen"}],"nameIdentifiers":[{}]}]},"item_4_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Yosuke, Oyama","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Naoya, Maruyama","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Nikoli, Dryden","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Peter, Harrington","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Jan, Balewski","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Satoshi, Matsuoka","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Marc, Snir","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Peter, Nugent","creatorNameLang":"en"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Brian, Van Essen","creatorNameLang":"en"}],"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":"We report our preliminary work on large-scale training of a 3D convolutional neural network model for cosmological analyses of dark matter distributions. Previous work showed promising results for predicting cosmological parameters using CNNs trained on a large-scale parallel computing platform. However, due to its weak scaling nature, there exists a trade-off of training performance and prediction accuracy. This paper extends the existing work for better prediction accuracy and performance by exploiting finer-grained parallelism in distributed convolutions. We show significant improvements using the latest complex cosmological dataset with a huge model that was previously unfeasible due to its memory pressure. We achieve 1.42 PFlop/s on a single training task with a mini-batch size of 128 by using 512 Tesla V100 GPUs. Our results imply that the state-of-the-art deep learning case study can be further advanced with HPC-based algorithms.","subitem_description_type":"Other"}]},"item_4_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"We report our preliminary work on large-scale training of a 3D convolutional neural network model for cosmological analyses of dark matter distributions. Previous work showed promising results for predicting cosmological parameters using CNNs trained on a large-scale parallel computing platform. However, due to its weak scaling nature, there exists a trade-off of training performance and prediction accuracy. This paper extends the existing work for better prediction accuracy and performance by exploiting finer-grained parallelism in distributed convolutions. We show significant improvements using the latest complex cosmological dataset with a huge model that was previously unfeasible due to its memory pressure. We achieve 1.42 PFlop/s on a single training task with a mini-batch size of 128 by using 512 Tesla V100 GPUs. Our results imply that the state-of-the-art deep learning case study can be further advanced with HPC-based algorithms.","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"8","bibliographic_titles":[{"bibliographic_title":"研究報告ハイパフォーマンスコンピューティング（HPC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2019-07-17","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"8","bibliographicVolumeNumber":"2019-HPC-170"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"created":"2025-01-19T01:02:28.618618+00:00","id":198145,"links":{}}