{"id":"https://openalex.org/W4400447476","doi":"https://doi.org/10.1109/access.2024.3425638","title":"Exploring Attention Sparsity to Accelerate Transformer Training on GPUs","display_name":"Exploring Attention Sparsity to Accelerate Transformer Training on GPUs","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4400447476","doi":"https://doi.org/10.1109/access.2024.3425638"},"language":"en","primary_location":{"id":"doi:10.1109/access.2024.3425638","is_oa":true,"landing_page_url":"http://dx.doi.org/10.1109/access.2024.3425638","pdf_url":null,"source":{"id":"https://openalex.org/S2485537415","display_name":"IEEE Access","issn_l":"2169-3536","issn":["2169-3536"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Access","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"http://dx.doi.org/10.1109/access.2024.3425638","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5002163407","display_name":"Bokyeong Yoon","orcid":"https://orcid.org/0009-0006-0175-0753"},"institutions":[{"id":"https://openalex.org/I148751991","display_name":"Sogang University","ror":"https://ror.org/056tn4839","country_code":"KR","type":"education","lineage":["https://openalex.org/I148751991"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Bokyeong Yoon","raw_affiliation_strings":["Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0006-0175-0753","affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I148751991"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102703030","display_name":"Ah-Hyun Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I148751991","display_name":"Sogang University","ror":"https://ror.org/056tn4839","country_code":"KR","type":"education","lineage":["https://openalex.org/I148751991"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Ah-Hyun Lee","raw_affiliation_strings":["Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I148751991"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100676208","display_name":"Jinsung Kim","orcid":"https://orcid.org/0000-0003-3751-8869"},"institutions":[{"id":"https://openalex.org/I67900169","display_name":"Chung-Ang University","ror":"https://ror.org/01r024a98","country_code":"KR","type":"education","lineage":["https://openalex.org/I67900169"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jinsung Kim","raw_affiliation_strings":["School of Computer Science and Engineering, Chung-Ang University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0003-3751-8869","affiliations":[{"raw_affiliation_string":"School of Computer Science and Engineering, Chung-Ang University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I67900169"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5023057376","display_name":"Gordon Euhyun Moon","orcid":"https://orcid.org/0000-0003-4992-6181"},"institutions":[{"id":"https://openalex.org/I148751991","display_name":"Sogang University","ror":"https://ror.org/056tn4839","country_code":"KR","type":"education","lineage":["https://openalex.org/I148751991"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Gordon Euhyun Moon","raw_affiliation_strings":["Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0003-4992-6181","affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering, Sogang University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I148751991"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":{"value":1850,"currency":"USD","value_usd":1850},"apc_paid":{"value":1850,"currency":"USD","value_usd":1850},"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.08764494,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"12","issue":null,"first_page":"131373","last_page":"131384"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9993000030517578,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9993000030517578,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12702","display_name":"Brain Tumor Detection and Classification","score":0.9927999973297119,"subfield":{"id":"https://openalex.org/subfields/2808","display_name":"Neurology"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T12676","display_name":"Machine Learning and ELM","score":0.989300012588501,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7913734912872314},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.7879171967506409},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6874892711639404},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.6305183172225952},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.44814276695251465},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4392455518245697},{"id":"https://openalex.org/keywords/skewness","display_name":"Skewness","score":0.42702800035476685},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.3400747776031494},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.25338587164878845},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.10286220908164978},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.10065022110939026}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7913734912872314},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.7879171967506409},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6874892711639404},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6305183172225952},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.44814276695251465},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4392455518245697},{"id":"https://openalex.org/C122342681","wikidata":"https://www.wikidata.org/wiki/Q330828","display_name":"Skewness","level":2,"score":0.42702800035476685},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3400747776031494},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.25338587164878845},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.10286220908164978},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.10065022110939026},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1109/access.2024.3425638","is_oa":true,"landing_page_url":"http://dx.doi.org/10.1109/access.2024.3425638","pdf_url":null,"source":{"id":"https://openalex.org/S2485537415","display_name":"IEEE Access","issn_l":"2169-3536","issn":["2169-3536"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Access","raw_type":"journal-article"},{"id":"pmh:oai:doaj.org/article:36d2203371f44d2d892e6e1f40d57a23","is_oa":true,"landing_page_url":"https://doaj.org/article/36d2203371f44d2d892e6e1f40d57a23","pdf_url":null,"source":{"id":"https://openalex.org/S4306401280","display_name":"DOAJ (DOAJ: Directory of Open Access Journals)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"IEEE Access, Vol 12, Pp 131373-131384 (2024)","raw_type":"article"}],"best_oa_location":{"id":"doi:10.1109/access.2024.3425638","is_oa":true,"landing_page_url":"http://dx.doi.org/10.1109/access.2024.3425638","pdf_url":null,"source":{"id":"https://openalex.org/S2485537415","display_name":"IEEE Access","issn_l":"2169-3536","issn":["2169-3536"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Access","raw_type":"journal-article"},"sustainable_development_goals":[{"score":0.6299999952316284,"display_name":"Affordable and clean energy","id":"https://metadata.un.org/sdg/7"}],"awards":[{"id":"https://openalex.org/G2287828947","display_name":null,"funder_award_id":"RS-2024-00259099","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"}],"funders":[{"id":"https://openalex.org/F4320322120","display_name":"National Research Foundation of Korea","ror":"https://ror.org/013aysd81"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":26,"referenced_works":["https://openalex.org/W1566289585","https://openalex.org/W2570841452","https://openalex.org/W2923014074","https://openalex.org/W3015468748","https://openalex.org/W3033529678","https://openalex.org/W3105238007","https://openalex.org/W3118608800","https://openalex.org/W3131922516","https://openalex.org/W3138516171","https://openalex.org/W4287748555","https://openalex.org/W4307537353","https://openalex.org/W4323654151","https://openalex.org/W4382334332","https://openalex.org/W4385245566","https://openalex.org/W4395066850","https://openalex.org/W6685053522","https://openalex.org/W6755207826","https://openalex.org/W6761628794","https://openalex.org/W6768021236","https://openalex.org/W6771626834","https://openalex.org/W6776048684","https://openalex.org/W6779163297","https://openalex.org/W6780069040","https://openalex.org/W6781533629","https://openalex.org/W6794559225","https://openalex.org/W6795842910"],"related_works":["https://openalex.org/W2058965144","https://openalex.org/W2164382479","https://openalex.org/W2146343568","https://openalex.org/W98480971","https://openalex.org/W2150291671","https://openalex.org/W2013643406","https://openalex.org/W2027972911","https://openalex.org/W2157978810","https://openalex.org/W2597809628","https://openalex.org/W3046370962"],"abstract_inverted_index":{"The":[0],"computational":[1],"complexity":[2],"required":[3],"for":[4,40,99],"training":[5,23,50,62,134,195],"a":[6,25,109,126],"Transformer":[7,27,49,112,213],"model":[8,88,102],"quadratically":[9],"increases":[10],"as":[11],"the":[12,15,22,36,41,47,58,66,70,76,80,94,121,132,136,139,146,151,156,164,176,181],"length":[13],"of":[14,24,38,129,131,135,158,163,178],"input":[16],"sequence":[17],"increases.":[18],"Therefore,":[19],"to":[20,34,56,187,210],"accelerate":[21],"large-scale":[26],"with":[28],"long":[29],"sequences,":[30],"it":[31],"is":[32,97,143],"crucial":[33],"reduce":[35],"number":[37,128,177],"operations":[39,179],"multi-head":[42,59,182],"attention":[43,60,71,81,122,165,183],"computations,":[44],"which":[45],"dominate":[46],"overall":[48],"process.":[51],"Previous":[52],"approaches":[53],"have":[54],"sought":[55],"sparsify":[57],"before":[61],"by":[63],"statically":[64],"selecting":[65],"critical":[67,77,95],"elements":[68,78,96],"in":[69,79,120,145,180],"score":[72,82,123,166],"matrix.":[73],"However,":[74],"since":[75],"matrix":[83,124],"can":[84],"vary":[85],"across":[86],"different":[87],"tasks":[89],"and":[90,116,160,206],"datasets,":[91],"dynamically":[92],"considering":[93],"essential":[98],"achieving":[100,185],"better":[101,207],"quality.":[103],"In":[104],"this":[105],"paper,":[106],"we":[107],"propose":[108],"new":[110],"sparsity-aware":[111],"that":[113,171],"captures":[114],"task-":[115],"input-dependent":[117],"sparsity":[118,141],"pattern":[119,142],"during":[125],"small":[127],"steps":[130],"standard":[133,152],"Transformer.":[137],"Then":[138],"identified":[140],"utilized":[144],"sparse":[147,212],"training,":[148,153],"transferred":[149],"from":[150],"based":[154],"on":[155],"degree":[157],"skewness":[159],"distance":[161],"values":[162],"matrices.":[167],"Experimental":[168],"results":[169],"demonstrate":[170],"our":[172],"approach":[173],"significantly":[174],"reduces":[175],"operations,":[184],"up":[186],"<inline-formula":[188,197],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[189,198],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">":[190,199],"<tex-math":[191,200],"notation=\"LaTeX\">$2.84\\times":[192],"$":[193,202],"</tex-math></inline-formula>":[194,203],"speedup,":[196],"notation=\"LaTeX\">$6.87\\times":[201],"memory":[204],"reduction":[205],"accuracy":[208],"compared":[209],"state-of-the-art":[211],"models.":[214]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}