{"id":"https://openalex.org/W2942889782","doi":"https://doi.org/10.1109/tpds.2019.2913833","title":"Fast Deep Neural Network Training on Distributed Systems and Cloud TPUs","display_name":"Fast Deep Neural Network Training on Distributed Systems and Cloud TPUs","publication_year":2019,"publication_date":"2019-04-30","ids":{"openalex":"https://openalex.org/W2942889782","doi":"https://doi.org/10.1109/tpds.2019.2913833","mag":"2942889782"},"language":"en","primary_location":{"id":"doi:10.1109/tpds.2019.2913833","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tpds.2019.2913833","pdf_url":null,"source":{"id":"https://openalex.org/S97130795","display_name":"IEEE Transactions on Parallel and Distributed Systems","issn_l":"1045-9219","issn":["1045-9219","1558-2183","2161-9883"],"is_oa":false,"is_in_doaj":false,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Parallel and Distributed Systems","raw_type":"journal-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/A5100658705","display_name":"Yang You","orcid":"https://orcid.org/0000-0003-2816-4384"},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Yang You","raw_affiliation_strings":["Computer Science Division, University of California at Berkeley, Berkeley, CA, USA"],"affiliations":[{"raw_affiliation_string":"Computer Science Division, University of California at Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100422982","display_name":"Zhao Zhang","orcid":"https://orcid.org/0000-0001-5921-0035"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao Zhang","raw_affiliation_strings":["Texas Advanced Computing Center, University of Texas at Austin, Austin, TX, USA"],"affiliations":[{"raw_affiliation_string":"Texas Advanced Computing Center, University of Texas at Austin, Austin, TX, USA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010841999","display_name":"Cho\u2010Jui Hsieh","orcid":"https://orcid.org/0000-0002-3520-9627"},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Cho-Jui Hsieh","raw_affiliation_strings":["Department of Computer Science, University of California at Los Angeles, Los Angeles, CA, USA"],"affiliations":[{"raw_affiliation_string":"Department of Computer Science, University of California at Los Angeles, Los Angeles, CA, USA","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076825233","display_name":"James Demmel","orcid":"https://orcid.org/0000-0003-1145-3745"},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"James Demmel","raw_affiliation_strings":["Computer Science Division, University of California at Berkeley, Berkeley, CA, USA"],"affiliations":[{"raw_affiliation_string":"Computer Science Division, University of California at Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5047285420","display_name":"Kurt Keutzer","orcid":"https://orcid.org/0000-0003-3868-8501"},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kurt Keutzer","raw_affiliation_strings":["Computer Science Division, University of California at Berkeley, Berkeley, CA, USA"],"affiliations":[{"raw_affiliation_string":"Computer Science Division, University of California at Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I95457486"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5100658705"],"corresponding_institution_ids":["https://openalex.org/I95457486"],"apc_list":null,"apc_paid":null,"fwci":3.5833,"has_fulltext":false,"cited_by_count":60,"citation_normalized_percentile":{"value":0.94435943,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":91,"max":100},"biblio":{"volume":"30","issue":"11","first_page":"2449","last_page":"2462"},"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/T12702","display_name":"Brain Tumor Detection and Classification","score":0.9991999864578247,"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":0.9979000091552734,"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":0.8575468063354492},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.7511035203933716},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"Stochastic gradient descent","score":0.6560357213020325},{"id":"https://openalex.org/keywords/supercomputer","display_name":"Supercomputer","score":0.6528248190879822},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.5548073053359985},{"id":"https://openalex.org/keywords/cloud-computing","display_name":"Cloud computing","score":0.5316907167434692},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.5237981081008911},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.5086601376533508},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.4958120286464691},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.4707111716270447},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.46103906631469727},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.41966235637664795},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.4165947437286377},{"id":"https://openalex.org/keywords/computer-engineering","display_name":"Computer engineering","score":0.37785786390304565},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.3583618402481079},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.17104700207710266}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8575468063354492},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.7511035203933716},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.6560357213020325},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.6528248190879822},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5548073053359985},{"id":"https://openalex.org/C79974875","wikidata":"https://www.wikidata.org/wiki/Q483639","display_name":"Cloud computing","level":2,"score":0.5316907167434692},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.5237981081008911},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.5086601376533508},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.4958120286464691},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4707111716270447},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.46103906631469727},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41966235637664795},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.4165947437286377},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.37785786390304565},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3583618402481079},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.17104700207710266},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.0},{"id":"https://openalex.org/C120665830","wikidata":"https://www.wikidata.org/wiki/Q14620","display_name":"Optics","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0},{"id":"https://openalex.org/C13280743","wikidata":"https://www.wikidata.org/wiki/Q131089","display_name":"Geodesy","level":1,"score":0.0},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tpds.2019.2913833","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tpds.2019.2913833","pdf_url":null,"source":{"id":"https://openalex.org/S97130795","display_name":"IEEE Transactions on Parallel and Distributed Systems","issn_l":"1045-9219","issn":["1045-9219","1558-2183","2161-9883"],"is_oa":false,"is_in_doaj":false,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Parallel and Distributed Systems","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure","score":0.5600000023841858}],"awards":[{"id":"https://openalex.org/G6046368439","display_name":null,"funder_award_id":"HR0011-12- 2-0016","funder_id":"https://openalex.org/F4320332180","funder_display_name":"Defense Advanced Research Projects Agency"},{"id":"https://openalex.org/G8454444317","display_name":null,"funder_award_id":"OAC-1540931","funder_id":"https://openalex.org/F4320335353","funder_display_name":"National Science Foundation of Sri Lanka"}],"funders":[{"id":"https://openalex.org/F4320307798","display_name":"Nokia","ror":"https://ror.org/04pkc8m17"},{"id":"https://openalex.org/F4320307904","display_name":"Oracle","ror":"https://ror.org/006c77m33"},{"id":"https://openalex.org/F4320309327","display_name":"Google","ror":"https://ror.org/00njsd438"},{"id":"https://openalex.org/F4320309480","display_name":"Nvidia","ror":"https://ror.org/03jdj4y14"},{"id":"https://openalex.org/F4320322183","display_name":"Huawei Technologies","ror":"https://ror.org/00cmhce21"},{"id":"https://openalex.org/F4320322382","display_name":"LG Display","ror":"https://ror.org/03ddh2c27"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"},{"id":"https://openalex.org/F4320332195","display_name":"Samsung","ror":"https://ror.org/04w3jy968"},{"id":"https://openalex.org/F4320335353","display_name":"National Science Foundation of Sri Lanka","ror":"https://ror.org/010xaa060"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":49,"referenced_works":["https://openalex.org/W1518038022","https://openalex.org/W1549847926","https://openalex.org/W1572016165","https://openalex.org/W1598866093","https://openalex.org/W1815076433","https://openalex.org/W2064675550","https://openalex.org/W2081612620","https://openalex.org/W2094233035","https://openalex.org/W2108598243","https://openalex.org/W2110485445","https://openalex.org/W2131613942","https://openalex.org/W2138243089","https://openalex.org/W2155893237","https://openalex.org/W2168231600","https://openalex.org/W2193413348","https://openalex.org/W2194775991","https://openalex.org/W2287011250","https://openalex.org/W2336650964","https://openalex.org/W2525778437","https://openalex.org/W2606722458","https://openalex.org/W2622263826","https://openalex.org/W2749988060","https://openalex.org/W2769856846","https://openalex.org/W2770102447","https://openalex.org/W2962747323","https://openalex.org/W2962911728","https://openalex.org/W2963792515","https://openalex.org/W2963804082","https://openalex.org/W2963959597","https://openalex.org/W4254816979","https://openalex.org/W4294641903","https://openalex.org/W4298840580","https://openalex.org/W6635810480","https://openalex.org/W6638545294","https://openalex.org/W6638803421","https://openalex.org/W6680402377","https://openalex.org/W6684859321","https://openalex.org/W6684953698","https://openalex.org/W6687566353","https://openalex.org/W6696413332","https://openalex.org/W6703420464","https://openalex.org/W6726983090","https://openalex.org/W6727690538","https://openalex.org/W6730009442","https://openalex.org/W6739622702","https://openalex.org/W6743289643","https://openalex.org/W6743955621","https://openalex.org/W6746419290","https://openalex.org/W6746514494"],"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/W2955229517","https://openalex.org/W2966297898"],"abstract_inverted_index":{"Since":[0],"its":[1],"creation,":[2],"the":[3,17,27,40,75,78,95,108,112,121,127,158,174,185,203,242,264,282,288,315,328,342,361,381,397,429,445,474],"ImageNet-1k":[4,53],"benchmark":[5,14,42],"set":[6],"has":[7,36,191,377],"played":[8],"a":[9,13,51,58,194,214,236,291],"significant":[10,137],"role":[11],"as":[12,39,260],"for":[15,43,182,249,326],"ascertaining":[16],"accuracy":[18,212,304,448],"of":[19,107,111,147,217,226,252,290,384,399,432],"different":[20,45],"deep":[21],"neural":[22,268],"net":[23],"(DNN)":[24],"models":[25],"on":[26,57,134,187,196,266,305,409,422,473],"image":[28],"classification":[29],"problem.":[30],"Moreover,":[31],"in":[32,72,123,178,206,380,450],"recent":[33],"years":[34],"it":[35],"also":[37,172],"served":[38],"principal":[41],"assessing":[44],"approaches":[46,151],"to":[47,94,119,144,152,166,234,287,349,368,434,459,469],"DNN":[48,142,189,229],"training.":[49,230],"Finishing":[50],"90-epoch":[52,362],"training":[54,66,122,143,155,190,198,345,364],"with":[55,277],"ResNet-50":[56,363],"NVIDIA":[59],"M40":[60],"GPU":[61],"takes":[62],"14":[63],"days.":[64],"This":[65],"requires":[67],"1018":[68],"single":[69,88],"precision":[70,89],"operations":[71,90],"total.":[73],"On":[74],"other":[76],"hand,":[77],"world's":[79],"current":[80],"fastest":[81,113],"supercomputer":[82],"can":[83,103],"finish":[84,120],"3":[85],"x":[86],"1017":[87],"per":[91],"second":[92],"(according":[93],"Nov":[96],"2018":[97],"Top":[98],"500":[99],"results).":[100],"If":[101],"we":[102,115,222,261,313,340,359,405,427,455],"make":[104],"full":[105],"use":[106,235,314],"computing":[109,254],"capability":[110,225],"supercomputer,":[114],"should":[116],"be":[117],"able":[118],"several":[124],"seconds.":[125],"Over":[126],"last":[128],"two":[129],"years,":[130],"researchers":[131],"have":[132,156],"focused":[133],"closing":[135],"this":[136,220,394],"performance":[138],"gap":[139],"through":[140],"scaling":[141,153,188],"larger":[145,310],"numbers":[146],"processors.":[148],"Most":[149],"successful":[150],"ImageNet":[154],"used":[157,177],"synchronous":[159,168,208],"minibatch":[160],"stochastic":[161],"gradient":[162],"descent":[163],"(SGD).":[164],"However,":[165],"scale":[167,428],"SGD":[169,209],"one":[170],"must":[171],"increase":[173],"batch":[175,204,238,306,329,430],"size":[176,205,431],"each":[179],"iteration.":[180],"Thus,":[181],"many":[183],"researchers,":[184],"focus":[186,195],"translated":[192],"into":[193],"developing":[197],"algorithms":[199],"that":[200,308],"enable":[201],"increasing":[202],"data-parallel":[207],"without":[210],"losing":[211],"over":[213],"fixed":[215],"number":[216],"epochs.":[218],"In":[219],"paper,":[221],"investigate":[223],"supercomputers'":[224],"speeding":[227],"up":[228],"Our":[231,256,371],"approach":[232,257,300,408,458],"is":[233,258,373,442],"large":[237,278],"size,":[239],"powered":[240],"by":[241],"Layer-wise":[243],"Adaptive":[244],"Rate":[245],"Scaling":[246],"(LARS)":[247],"algorithm,":[248],"efficient":[250],"usage":[251],"massive":[253],"resources.":[255],"generic,":[259],"empirically":[262],"evaluate":[263],"effectiveness":[265],"five":[267],"networks:":[269],"AlexNet,":[270],"AlexNet-BN,":[271],"GNMT,":[272],"ResNet-50,":[273],"and":[274,376,388,396,424,436],"ResNet-50-v2":[275,433],"trained":[276],"datasets":[279],"while":[280],"preserving":[281],"state-of-the-art":[283],"test":[284,303],"accuracy.":[285],"Compared":[286],"baseline":[289],"previous":[292,420],"study":[293],"from":[294,347,366],"Goyal":[295,323],"et":[296,324],"al.":[297,325],"[1],":[298],"our":[299,318,400,407,419,451,457],"shows":[301],"higher":[302],"sizes":[307,330],"are":[309,320],"than":[311,322,444],"16K.When":[312],"same":[316],"baseline,":[317],"results":[319],"better":[321,443],"all":[327],"(Fig.":[331],"20).":[332],"Using":[333],"2,048":[334,353],"Intel":[335,354,382],"Xeon":[336,355],"Platinum":[337],"8160":[338],"processors,":[339],"reduce":[341,360],"100-epoch":[343],"AlexNet":[344],"time":[346,365],"hours":[348,367],"11":[350],"minutes.":[351,370],"With":[352],"Phi":[356],"7250":[357],"Processors,":[358],"20":[369],"implementation":[372],"open":[374],"source":[375],"been":[378],"released":[379],"distribution":[383],"Caffe,":[385],"Facebook's":[386],"PyTorch,":[387],"Google's":[389,410,460],"TensorFlow.":[390],"The":[391],"difference":[392],"between":[393],"paper":[395],"conference-version":[398],"work":[401],"[2]":[402],"includes:":[403],"(1)":[404],"implement":[406],"cloud":[411,475],"Tensor":[412],"Processing":[413],"Unit":[414],"(TPU)":[415],"platform,":[416],"which":[417,441,466],"verifies":[418],"success":[421],"CPUs":[423],"GPUs.":[425],"(2)":[426],"32K":[435],"achieve":[437],"76.3":[438],"percent":[439,447],"accuracy,":[440],"75.3":[446],"achieved":[449],"conference":[452],"paper.":[453],"(3)":[454],"apply":[456],"Neural":[461],"Machine":[462],"Translation":[463],"(GNMT)":[464],"application,":[465],"helps":[467],"us":[468],"achieves":[470],"4x":[471],"speedup":[472],"TPUs.":[476]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":8},{"year":2023,"cited_by_count":13},{"year":2022,"cited_by_count":9},{"year":2021,"cited_by_count":10},{"year":2020,"cited_by_count":12},{"year":2019,"cited_by_count":4}],"updated_date":"2026-04-02T15:55:50.835912","created_date":"2025-10-10T00:00:00"}