{"id":"https://openalex.org/W4221008227","doi":"https://doi.org/10.1145/3503221.3508399","title":"Near-optimal sparse allreduce for distributed deep learning","display_name":"Near-optimal sparse allreduce for distributed deep learning","publication_year":2022,"publication_date":"2022-03-28","ids":{"openalex":"https://openalex.org/W4221008227","doi":"https://doi.org/10.1145/3503221.3508399"},"language":"en","primary_location":{"id":"doi:10.1145/3503221.3508399","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3503221.3508399","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2201.07598","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5088172355","display_name":"Shigang Li","orcid":"https://orcid.org/0000-0003-0022-7865"},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":true,"raw_author_name":"Shigang Li","raw_affiliation_strings":["ETH Zurich, Switzerland"],"affiliations":[{"raw_affiliation_string":"ETH Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5026990786","display_name":"Torsten Hoefler","orcid":"https://orcid.org/0000-0002-1333-9797"},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Torsten Hoefler","raw_affiliation_strings":["ETH Zurich, Switzerland"],"affiliations":[{"raw_affiliation_string":"ETH Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5088172355"],"corresponding_institution_ids":["https://openalex.org/I35440088"],"apc_list":null,"apc_paid":null,"fwci":6.3847,"has_fulltext":false,"cited_by_count":49,"citation_normalized_percentile":{"value":0.96973365,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":94,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"135","last_page":"149"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9998999834060669,"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"}},"topics":[{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9998999834060669,"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/T10500","display_name":"Sparse and Compressive Sensing Techniques","score":0.9997000098228455,"subfield":{"id":"https://openalex.org/subfields/2206","display_name":"Computational Mechanics"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12676","display_name":"Machine Learning and ELM","score":0.9994000196456909,"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":0.6806386113166809},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4632483720779419},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.45223021507263184}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6806386113166809},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4632483720779419},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.45223021507263184}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3503221.3508399","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3503221.3508399","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2201.07598","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2201.07598","pdf_url":"https://arxiv.org/pdf/2201.07598","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2201.07598","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2201.07598","pdf_url":"https://arxiv.org/pdf/2201.07598","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":57,"referenced_works":["https://openalex.org/W192578068","https://openalex.org/W381682963","https://openalex.org/W1522301498","https://openalex.org/W1686810756","https://openalex.org/W2084557272","https://openalex.org/W2113911479","https://openalex.org/W2131613942","https://openalex.org/W2170796499","https://openalex.org/W2612387305","https://openalex.org/W2617766261","https://openalex.org/W2622263826","https://openalex.org/W2769644379","https://openalex.org/W2786602455","https://openalex.org/W2788193959","https://openalex.org/W2788653909","https://openalex.org/W2798445803","https://openalex.org/W2809617027","https://openalex.org/W2896457183","https://openalex.org/W2912083425","https://openalex.org/W2935990182","https://openalex.org/W2962747323","https://openalex.org/W2963086843","https://openalex.org/W2963433607","https://openalex.org/W2963766684","https://openalex.org/W2965658867","https://openalex.org/W2966527647","https://openalex.org/W2967558351","https://openalex.org/W2969388332","https://openalex.org/W2972087877","https://openalex.org/W2982475424","https://openalex.org/W2985108934","https://openalex.org/W2988070836","https://openalex.org/W2990352720","https://openalex.org/W2995435108","https://openalex.org/W3036879053","https://openalex.org/W3038104246","https://openalex.org/W3038581078","https://openalex.org/W3091097978","https://openalex.org/W3099748883","https://openalex.org/W3101036738","https://openalex.org/W3107401959","https://openalex.org/W3129251660","https://openalex.org/W3129329365","https://openalex.org/W3132107458","https://openalex.org/W3180473732","https://openalex.org/W3212284901","https://openalex.org/W3213513948","https://openalex.org/W4230874317","https://openalex.org/W4250599615","https://openalex.org/W4252163020","https://openalex.org/W4287363917","https://openalex.org/W4292779060","https://openalex.org/W4294527468","https://openalex.org/W4295312788","https://openalex.org/W4297685247","https://openalex.org/W4301239768","https://openalex.org/W4385245566"],"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":{"Communication":[0],"overhead":[1],"is":[2,18,29,81,96,155],"one":[3],"of":[4,38,42],"the":[5,24,40,53,85,100,106,119,146,150],"major":[6],"obstacles":[7],"to":[8,22,32,111,141],"train":[9],"large":[10],"deep":[11,129],"learning":[12,130],"models":[13,126],"at":[14],"scale.":[15],"Gradient":[16,89],"sparsification":[17,54,101],"a":[19,60,70],"promising":[20],"technique":[21],"reduce":[23,99],"communication":[25,78],"volume.":[26],"However,":[27],"it":[28],"very":[30],"challenging":[31],"obtain":[33],"real":[34],"performance":[35],"improvement":[36,165],"because":[37],"(1)":[39],"difficulty":[41],"achieving":[43],"an":[44,112],"scalable":[45,157],"and":[46,51,93,149,158],"efficient":[47],"sparse":[48,66,72,152],"allreduce":[49,73],"algorithm":[50,74],"(2)":[52],"overhead.":[55],"This":[56],"paper":[57],"proposes":[58],"Ok-Topk,":[59],"scheme":[61],"for":[62,166],"distributed":[63],"training":[64,161],"with":[65,84,123,145],"gradients.":[67],"Ok-Topk":[68,103,136,154],"integrates":[69],"novel":[71],"(less":[75],"than":[76],"6k":[77],"volume":[79],"which":[80],"asymptotically":[82],"optimal)":[83],"decentralized":[86],"parallel":[87],"Stochastic":[88],"Descent":[90],"(SGD)":[91],"optimizer,":[92],"its":[94],"convergence":[95],"proved.":[97],"To":[98],"overhead,":[102],"efficiently":[104],"selects":[105],"top-k":[107],"gradient":[108],"values":[109],"according":[110],"estimated":[113],"threshold.":[114],"Evaluations":[115],"are":[116],"conducted":[117],"on":[118,168],"Piz":[120],"Daint":[121],"supercomputer":[122],"neural":[124],"network":[125],"from":[127],"different":[128],"domains.":[131],"Empirical":[132],"results":[133],"show":[134],"that":[135],"achieves":[137],"similar":[138],"model":[139],"accuracy":[140],"dense":[142,148],"allreduce.":[143],"Compared":[144],"optimized":[147],"state-of-the-art":[151],"allreduces,":[153],"more":[156],"significantly":[159],"improves":[160],"throughput":[162],"(e.g.,":[163],"3.29x-12.95x":[164],"BERT":[167],"256":[169],"GPUs).":[170]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":19},{"year":2024,"cited_by_count":11},{"year":2023,"cited_by_count":14},{"year":2022,"cited_by_count":2}],"updated_date":"2026-04-10T15:06:20.359241","created_date":"2025-10-10T00:00:00"}