{"id":"https://openalex.org/W4313965085","doi":"https://doi.org/10.48550/arxiv.2301.02654","title":"Does compressing activations help model parallel training?","display_name":"Does compressing activations help model parallel training?","publication_year":2023,"publication_date":"2023-01-06","ids":{"openalex":"https://openalex.org/W4313965085","doi":"https://doi.org/10.48550/arxiv.2301.02654"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2301.02654","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.02654","pdf_url":"https://arxiv.org/pdf/2301.02654","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2301.02654","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Bian, Song","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bian, Song","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Li, Dacheng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Dacheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100701041","display_name":"Hongyi Wang","orcid":"https://orcid.org/0009-0006-0034-0074","observed_orcids":["https://orcid.org/0009-0006-0034-0074"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Hongyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5009547049","display_name":"Eric P. Xing","orcid":"https://orcid.org/0009-0005-9158-4201","observed_orcids":["https://orcid.org/0009-0005-9158-4201","https://orcid.org/0000-0002-3683-4280"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xing, Eric P.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5082710354","display_name":"Shivaram Venkataraman","orcid":"https://orcid.org/0000-0001-9575-7935","observed_orcids":["https://orcid.org/0000-0001-9575-7935"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Venkataraman, Shivaram","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3959999978542328,"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"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3959999978542328,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.3230000138282776,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.1339000016450882,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/activation-compression","display_name":"activation compression","score":0.7860000133514404},{"id":"https://openalex.org/keywords/transformer-training","display_name":"transformer training","score":0.7129999995231628},{"id":"https://openalex.org/keywords/communication-efficiency","display_name":"communication efficiency","score":0.5740000009536743},{"id":"https://openalex.org/keywords/pruning","display_name":"pruning","score":0.46299999952316284},{"id":"https://openalex.org/keywords/quantization","display_name":"quantization","score":0.9800000190734863},{"id":"https://openalex.org/keywords/data-parallelism","display_name":"data parallelism","score":0.6790000200271606},{"id":"https://openalex.org/keywords/large-scale-neural-networks","display_name":"large-scale neural networks","score":0.5299999713897705},{"id":"https://openalex.org/keywords/model-scaling","display_name":"model scaling","score":0.45399999618530273}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.860545814037323},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.713111162185669},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.6431951522827148},{"id":"https://openalex.org/C42992933","wikidata":"https://www.wikidata.org/wiki/Q691169","display_name":"Task parallelism","level":3,"score":0.6161293387413025},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5462228655815125},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5172358155250549},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.47006234526634216},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.46773797273635864},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.444353312253952},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.41033878922462463},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34432244300842285},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33561789989471436},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2543378472328186},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"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/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2301.02654","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.02654","pdf_url":"https://arxiv.org/pdf/2301.02654","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"},{"id":"doi:10.48550/arxiv.2301.02654","doi":"https://doi.org/10.48550/arxiv.2301.02654","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2301.02654","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2301.02654","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2301.02654","pdf_url":"https://arxiv.org/pdf/2301.02654","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G2632089271","display_name":"Collaborative Research: SCH: Trustworthy and Explainable AI for Neurodegenerative Diseases","funder_award_id":"2123952","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G264567485","display_name":"III: Small: A New Approach to Latent Space Learning with Diversity-Inducing Regularization and Applications to Healthcare Data Analytics","funder_award_id":"1617583","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G5607104872","display_name":"XPS: FULL: Broad-Purpose, Aggressively Asynchronous and Theoretically Sound Parallel Large-scale Machine Learning","funder_award_id":"1629559","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G6555421410","display_name":"CNS Core: Small: Toward Globally-Optimal Resource Distribution and Computation Acceleration in Multi-Tenant and Heterogeneous Machine Learning Systems","funder_award_id":"2008248","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G7210974099","display_name":"ML Basis for Intelligence Augmentation:Toward Personalized Modeling, Reasoning under Data-Knowledge Symbiosis, and Interpretable Interaction for AI-assisted Human Decision-making","funder_award_id":"2040381","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G7792429826","display_name":"RI: III: Medium: Scalable Machine Learning for Automating Scientific Discovery in Astrophysics","funder_award_id":"1563887","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G8765148894","display_name":"Collaborative Research: RI: Medium: A Rigorous, General Framework for Tractable Learning of Large-Scale DAGs from Data","funder_award_id":"1955532","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320337368","display_name":"Division of Graduate Education","ror":"https://ror.org/00whkrf32"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4313965085.pdf","grobid_xml":"https://content.openalex.org/works/W4313965085.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4393172386","https://openalex.org/W4414895674","https://openalex.org/W4318621310","https://openalex.org/W4285173593","https://openalex.org/W4281945007","https://openalex.org/W7133225880","https://openalex.org/W7158422383","https://openalex.org/W3008851394","https://openalex.org/W4318541692","https://openalex.org/W4404450868"],"abstract_inverted_index":{"Large-scale":[0],"Transformer":[1,115],"models":[2],"are":[3],"known":[4],"for":[5,24,93,157],"their":[6],"exceptional":[7],"performance":[8],"in":[9,40,50,57],"a":[10,51,58,113],"range":[11],"of":[12,90,103,160],"tasks,":[13],"but":[14,55],"training":[15,32,116],"them":[16],"can":[17],"be":[18],"difficult":[19],"due":[20],"to":[21,30,35],"the":[22,37,83,88,148],"requirement":[23],"communication-intensive":[25],"model":[26,69,94,149,161],"parallelism.":[27,77,95],"One":[28],"way":[29],"improve":[31],"speed":[33],"is":[34,61,150],"compress":[36],"message":[38],"size":[39],"communication.":[41],"Previous":[42],"approaches":[43],"have":[44,66],"primarily":[45],"focused":[46],"on":[47,87],"compressing":[48],"gradients":[49],"data":[52,76],"parallelism":[53,70,162],"setting,":[54],"compression":[56,91,104,163],"model-parallel":[59],"setting":[60],"an":[62],"understudied":[63],"area.":[64],"We":[65,96,118,143],"discovered":[67],"that":[68],"has":[71],"fundamentally":[72],"different":[73,134],"characteristics":[74],"than":[75,124],"In":[78],"this":[79],"work,":[80],"we":[81,154],"present":[82],"first":[84],"empirical":[85],"study":[86],"effectiveness":[89],"methods":[92,121],"implement":[97],"and":[98,109,127,137,140],"evaluate":[99,119],"three":[100],"common":[101],"classes":[102],"algorithms":[105],"-":[106,111],"pruning-based,":[107],"learning-based,":[108],"quantization-based":[110],"using":[112],"popular":[114,129],"framework.":[117],"these":[120],"across":[122],"more":[123],"160":[125],"settings":[126],"8":[128],"datasets,":[130],"taking":[131],"into":[132],"account":[133],"hyperparameters,":[135],"hardware,":[136],"both":[138],"fine-tuning":[139],"pre-training":[141],"stages.":[142],"also":[144],"provide":[145,155],"analysis":[146],"when":[147],"scaled":[151],"up.":[152],"Finally,":[153],"insights":[156],"future":[158],"development":[159],"algorithms.":[164]},"counts_by_year":[{"year":2025,"cited_by_count":1},{"year":2023,"cited_by_count":1}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2023-01-10T00:00:00"}