{"id":"https://openalex.org/W4401409218","doi":"https://doi.org/10.1145/3673038.3673140","title":"Sparse Gradient Communication with AlltoAll for Accelerating Distributed Deep Learning","display_name":"Sparse Gradient Communication with AlltoAll for Accelerating Distributed Deep Learning","publication_year":2024,"publication_date":"2024-08-08","ids":{"openalex":"https://openalex.org/W4401409218","doi":"https://doi.org/10.1145/3673038.3673140"},"language":"en","primary_location":{"id":"doi:10.1145/3673038.3673140","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3673038.3673140","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 53rd International Conference on Parallel Processing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3673038.3673140","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Jing Peng","orcid":"https://orcid.org/0000-0002-7221-1216"},"institutions":[{"id":"https://openalex.org/I24201400","display_name":"Chengdu University of Information Technology","ror":"https://ror.org/01yxwrh59","country_code":"CN","type":"education","lineage":["https://openalex.org/I24201400"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jing Peng","raw_affiliation_strings":["Chengdu University of Information Technology, China"],"raw_orcid":"https://orcid.org/0000-0002-7221-1216","affiliations":[{"raw_affiliation_string":"Chengdu University of Information Technology, China","institution_ids":["https://openalex.org/I24201400"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104171291","display_name":"Zihan Li","orcid":"https://orcid.org/0000-0002-1817-9059"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Zihan Li","raw_affiliation_strings":["The Hong Kong University of Science and Technology, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0002-1817-9059","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016836702","display_name":"Shaohuai Shi","orcid":"https://orcid.org/0000-0002-1418-5160"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shaohuai Shi","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-1418-5160","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5060171629","display_name":"Bo Li","orcid":"https://orcid.org/0000-0003-2955-750X"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Bo Li","raw_affiliation_strings":["The Hong Kong University of Science and Technology, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0003-2955-750X","affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology, Hong Kong","institution_ids":["https://openalex.org/I200769079"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":3,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"148","last_page":"157"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9994999766349792,"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":0.9994999766349792,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9986000061035156,"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"}},{"id":"https://openalex.org/T10502","display_name":"Advanced Memory and Neural Computing","score":0.9973000288009644,"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":0.7024816274642944},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.4401940703392029},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.37232398986816406}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7024816274642944},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.4401940703392029},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37232398986816406}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3673038.3673140","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3673038.3673140","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 53rd International Conference on Parallel Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:repository.hkust.edu.hk:1783.1-142282","is_oa":false,"landing_page_url":"http://repository.hkust.edu.hk/ir/Record/1783.1-142282","pdf_url":null,"source":{"id":"https://openalex.org/S4306401796","display_name":"Rare & Special e-Zone (The Hong Kong University of Science and Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I200769079","host_organization_name":"Hong Kong University of Science and Technology","host_organization_lineage":["https://openalex.org/I200769079"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference paper"}],"best_oa_location":{"id":"doi:10.1145/3673038.3673140","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3673038.3673140","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 53rd International Conference on Parallel Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":23,"referenced_works":["https://openalex.org/W95608104","https://openalex.org/W1512808804","https://openalex.org/W1572016165","https://openalex.org/W2018346931","https://openalex.org/W2108598243","https://openalex.org/W2131613942","https://openalex.org/W2407022425","https://openalex.org/W2912500072","https://openalex.org/W2963786636","https://openalex.org/W2975712713","https://openalex.org/W2982475424","https://openalex.org/W2985108934","https://openalex.org/W3047357290","https://openalex.org/W3094326709","https://openalex.org/W3124349893","https://openalex.org/W3193250980","https://openalex.org/W3204921648","https://openalex.org/W4221008227","https://openalex.org/W4372267133","https://openalex.org/W4381327474","https://openalex.org/W4387490420","https://openalex.org/W4387544248","https://openalex.org/W4387544258"],"related_works":["https://openalex.org/W2731899572","https://openalex.org/W3215138031","https://openalex.org/W3009238340","https://openalex.org/W4321369474","https://openalex.org/W4360585206","https://openalex.org/W4285208911","https://openalex.org/W3082895349","https://openalex.org/W4213079790","https://openalex.org/W2248239756","https://openalex.org/W4323565446"],"abstract_inverted_index":{"Synchronous":[0],"stochastic":[1],"gradient":[2],"descent":[3],"(S-SGD)":[4],"with":[5,66,110,164],"data":[6],"parallelism":[7],"has":[8,51],"become":[9],"a":[10,80,99,121,141,194],"de-facto":[11],"approach":[12,103],"in":[13,58,75,120,124],"training":[14],"large-scale":[15],"deep":[16],"neural":[17],"networks":[18],"(DNNs)":[19],"on":[20,131,149,193],"multi-GPU":[21],"systems.":[22],"However,":[23],"S-SGD":[24],"requires":[25],"iteratively":[26],"synchronizing":[27],"gradients":[28],"from":[29,71],"all":[30],"workers,":[31],"which":[32,106],"incurs":[33],"excessive":[34],"communication":[35,61,73,102,112,125,159,181],"costs":[36],"and":[37,156,184],"limits":[38],"the":[39,60,91,132],"scaling":[40],"efficiency":[41,160,182],"of":[42,158],"GPU":[43],"clusters.":[44],"Gradient":[45],"sparsification":[46,49,68,109],"like":[47],"top-k":[48,67,108],"techniques":[50],"been":[52],"shown":[53],"to":[54,84,113,187],"be":[55,137],"potentially":[56],"effective":[57,144],"reducing":[59],"volume.":[62],"Yet,":[63],"existing":[64,191],"approaches":[65],"still":[69],"suffer":[70],"high":[72],"complexity":[74],"that":[76,134,177],"they":[77],"often":[78],"need":[79],"very":[81],"low":[82],"density":[83],"achieve":[85],"better":[86],"performance":[87],"while":[88],"easily":[89],"sacrificing":[90],"model":[92,172],"accuracy.":[93,173],"To":[94],"this":[95],"end,":[96],"we":[97,139],"propose":[98],"novel":[100],"sparse":[101,115],"called":[104],"TopKA2A,":[105],"integrates":[107],"AlltoAll":[111],"exchange":[114],"tensors":[116],"among":[117],"GPUs,":[118],"resulting":[119],"notable":[122],"reduction":[123],"complexity.":[126],"With":[127],"rigorous":[128],"theoretical":[129],"analysis":[130,155],"conditions":[133],"TopKA2A":[135,163,178],"can":[136],"applied,":[138],"design":[140],"simple":[142],"yet":[143],"tensor":[145],"fusion":[146],"algorithm":[147],"based":[148],"binary":[150],"search.":[151],"We":[152],"perform":[153],"in-depth":[154],"evaluation":[157],"by":[161],"comparing":[162],"state-of-the-art":[165],"solutions":[166],"over":[167],"popular":[168],"models":[169],"without":[170],"compromising":[171],"Experimental":[174],"results":[175],"demonstrate":[176],"yields":[179],"substantial":[180],"gains":[183],"runs":[185],"up":[186],"73%":[188],"faster":[189],"than":[190],"algorithms":[192],"32-GPU":[195],"cluster.":[196]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":2}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}