{"id":"https://openalex.org/W4391053210","doi":"https://doi.org/10.1145/3635035.3635045","title":"AshPipe: Asynchronous Hybrid Pipeline Parallel for DNN Training","display_name":"AshPipe: Asynchronous Hybrid Pipeline Parallel for DNN Training","publication_year":2024,"publication_date":"2024-01-18","ids":{"openalex":"https://openalex.org/W4391053210","doi":"https://doi.org/10.1145/3635035.3635045"},"language":"en","primary_location":{"id":"doi:10.1145/3635035.3635045","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3635035.3635045","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference on High Performance Computing in Asia-Pacific Region","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5093752369","display_name":"Ryubu Hosoki","orcid":"https://orcid.org/0000-0002-8435-7829"},"institutions":[{"id":"https://openalex.org/I114531698","display_name":"Tokyo Institute of Technology","ror":"https://ror.org/0112mx960","country_code":"JP","type":"education","lineage":["https://openalex.org/I114531698"]}],"countries":["JP"],"is_corresponding":true,"raw_author_name":"Ryubu Hosoki","raw_affiliation_strings":["Tokyo Institute of Technology, Japan"],"affiliations":[{"raw_affiliation_string":"Tokyo Institute of Technology, Japan","institution_ids":["https://openalex.org/I114531698"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011254074","display_name":"Toshio Endo","orcid":"https://orcid.org/0000-0001-7297-6211"},"institutions":[{"id":"https://openalex.org/I114531698","display_name":"Tokyo Institute of Technology","ror":"https://ror.org/0112mx960","country_code":"JP","type":"education","lineage":["https://openalex.org/I114531698"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Toshio Endo","raw_affiliation_strings":["Tokyo Institute of Technology, Japan"],"affiliations":[{"raw_affiliation_string":"Tokyo Institute of Technology, Japan","institution_ids":["https://openalex.org/I114531698"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063117669","display_name":"Takahiro Hirofuchi","orcid":"https://orcid.org/0000-0002-1253-6625"},"institutions":[{"id":"https://openalex.org/I73613424","display_name":"National Institute of Advanced Industrial Science and Technology","ror":"https://ror.org/01703db54","country_code":"JP","type":"government","lineage":["https://openalex.org/I73613424"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Takahiro Hirofuchi","raw_affiliation_strings":["National Institute of Advanced Industrial Science and Technology, Japan"],"affiliations":[{"raw_affiliation_string":"National Institute of Advanced Industrial Science and Technology, Japan","institution_ids":["https://openalex.org/I73613424"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5004964792","display_name":"Tsutomu Ikegami","orcid":null},"institutions":[{"id":"https://openalex.org/I73613424","display_name":"National Institute of Advanced Industrial Science and Technology","ror":"https://ror.org/01703db54","country_code":"JP","type":"government","lineage":["https://openalex.org/I73613424"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Tsutomu Ikegami","raw_affiliation_strings":["National Institute of Advanced Industrial Science and Technology, Japan"],"affiliations":[{"raw_affiliation_string":"National Institute of Advanced Industrial Science and Technology, Japan","institution_ids":["https://openalex.org/I73613424"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5093752369"],"corresponding_institution_ids":["https://openalex.org/I114531698"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.00502948,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"117","last_page":"126"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":1.0,"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":1.0,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9977999925613403,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9965999722480774,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8614460229873657},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.8565982580184937},{"id":"https://openalex.org/keywords/data-parallelism","display_name":"Data parallelism","score":0.8241345882415771},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.8076112270355225},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.7811645269393921},{"id":"https://openalex.org/keywords/asynchronous-communication","display_name":"Asynchronous communication","score":0.7782787680625916},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.6862499117851257},{"id":"https://openalex.org/keywords/instruction-level-parallelism","display_name":"Instruction-level parallelism","score":0.6753661632537842},{"id":"https://openalex.org/keywords/task-parallelism","display_name":"Task parallelism","score":0.6161090135574341},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.45913487672805786}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8614460229873657},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.8565982580184937},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.8241345882415771},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.8076112270355225},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.7811645269393921},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.7782787680625916},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.6862499117851257},{"id":"https://openalex.org/C140763907","wikidata":"https://www.wikidata.org/wiki/Q2714055","display_name":"Instruction-level parallelism","level":3,"score":0.6753661632537842},{"id":"https://openalex.org/C42992933","wikidata":"https://www.wikidata.org/wiki/Q691169","display_name":"Task parallelism","level":3,"score":0.6161090135574341},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.45913487672805786},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.0},{"id":"https://openalex.org/C555944384","wikidata":"https://www.wikidata.org/wiki/Q249","display_name":"Wireless","level":2,"score":0.0},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.0},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3635035.3635045","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3635035.3635045","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference on High Performance Computing in Asia-Pacific Region","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[{"score":0.41999998688697815,"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure"}],"awards":[{"id":"https://openalex.org/G1069223013","display_name":null,"funder_award_id":"JSPS KAKENHI","funder_id":"https://openalex.org/F4320334764","funder_display_name":"Japan Society for the Promotion of Science"},{"id":"https://openalex.org/G2997173747","display_name":"Scalable System Software for Machine Learning on Heterogeneous Parallel Computing Environments","funder_award_id":"20H04165","funder_id":"https://openalex.org/F4320334764","funder_display_name":"Japan Society for the Promotion of Science"},{"id":"https://openalex.org/G3459562248","display_name":null,"funder_award_id":"Grant","funder_id":"https://openalex.org/F4320334764","funder_display_name":"Japan Society for the Promotion of Science"},{"id":"https://openalex.org/G4636223006","display_name":null,"funder_award_id":"JSPS KAK","funder_id":"https://openalex.org/F4320334764","funder_display_name":"Japan Society for the Promotion of Science"},{"id":"https://openalex.org/G5315456625","display_name":null,"funder_award_id":"20H04165","funder_id":"https://openalex.org/F4320323817","funder_display_name":"Universitas Brawijaya"},{"id":"https://openalex.org/G8430481527","display_name":null,"funder_award_id":"Number","funder_id":"https://openalex.org/F4320334764","funder_display_name":"Japan Society for the Promotion of Science"}],"funders":[{"id":"https://openalex.org/F4320323817","display_name":"Universitas Brawijaya","ror":"https://ror.org/01wk3d929"},{"id":"https://openalex.org/F4320334764","display_name":"Japan Society for the Promotion of Science","ror":"https://ror.org/00hhkn466"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":21,"referenced_works":["https://openalex.org/W1967134278","https://openalex.org/W2117539524","https://openalex.org/W2194775991","https://openalex.org/W2618530766","https://openalex.org/W2734941459","https://openalex.org/W2901299405","https://openalex.org/W2963446712","https://openalex.org/W2963703197","https://openalex.org/W2969388332","https://openalex.org/W2980268386","https://openalex.org/W3031276512","https://openalex.org/W3037116551","https://openalex.org/W3130395060","https://openalex.org/W3130614105","https://openalex.org/W3132107458","https://openalex.org/W3137550677","https://openalex.org/W3188270315","https://openalex.org/W3202134982","https://openalex.org/W3206832494","https://openalex.org/W6600392996","https://openalex.org/W6778883912"],"related_works":["https://openalex.org/W2950520577","https://openalex.org/W2003935582","https://openalex.org/W74409296","https://openalex.org/W1991844655","https://openalex.org/W2105992728","https://openalex.org/W1229628","https://openalex.org/W2009213655","https://openalex.org/W2494130044","https://openalex.org/W2593878938","https://openalex.org/W305742777"],"abstract_inverted_index":{"Deep":[0],"Neural":[1],"Networks":[2],"(DNNs)":[3],"have":[4,10],"become":[5],"increasingly":[6],"computationally":[7],"intensive":[8],"and":[9,33,48,87,104,112],"larger":[11],"parameters,":[12],"requiring":[13],"efficient":[14,93],"parallelization":[15],"or":[16],"distribution":[17],"using":[18],"multiple":[19],"accelerators.":[20],"Pipeline":[21],"parallelism":[22,42,86,90,139,142,159,185],"has":[23],"been":[24],"proposed":[25,98,126],"as":[26,127],"an":[27,128],"effective":[28],"way":[29],"to":[30,65,91,130,140,183],"distribute":[31],"models":[32],"improve":[34],"hardware":[35],"utilization.":[36],"However,":[37],"the":[38,44,101,144,155,168,189],"problem":[39],"with":[40,107,179],"pipeline":[41,89],"is":[43,124,165],"trade-off":[45],"between":[46],"speedup":[47,94],"accuracy:":[49],"synchronous":[50],"approaches":[51,59],"do":[52],"not":[53],"provide":[54],"sufficient":[55],"speedup,":[56],"while":[57],"asynchronous":[58,88],"suffer":[60],"from":[61,69,137],"accuracy":[62,132,181],"degradation":[63],"due":[64],"a":[66,70,79,173],"different":[67],"scheme":[68],"single":[71],"worker.":[72],"In":[73],"this":[74],"paper,":[75],"we":[76],"propose":[77],"AshPipe,":[78],"hybrid":[80,141],"parallel":[81],"framework":[82],"that":[83,151],"combines":[84],"data":[85,105,138,158,184],"achieve":[92],"for":[95,160],"training.":[96,147],"The":[97],"runtime":[99],"uses":[100],"1F1B":[102],"schedule":[103],"parallelism,":[106],"non-uniform":[108],"numbers":[109],"of":[110,146,157],"workers":[111],"identical":[113],"global":[114],"batch":[115],"sizes":[116],"across":[117],"stages.":[118],"A":[119],"Switch":[120],"Parallelism":[121],"(SP)":[122],"mechanism":[123],"also":[125],"option":[129],"mitigate":[131],"degradation,":[133],"which":[134],"switches":[135],"over":[136],"in":[143,176],"course":[145],"Experimental":[148],"results":[149],"show":[150],"AshPipe":[152,171],"achieves":[153],"1.844\u00d7":[154],"throughput":[156],"ViT-H/14":[161],"whose":[162],"parameter":[163],"size":[164],"632M.":[166],"With":[167],"SP":[169],"mechanism,":[170],"achieved":[172],"30.2%":[174],"reduction":[175],"training":[177,187],"time":[178],"comparable":[180],"compared":[182],"when":[186],"on":[188],"CIFAR100":[190],"dataset.":[191]},"counts_by_year":[],"updated_date":"2026-04-10T15:06:20.359241","created_date":"2025-10-10T00:00:00"}