{"id":"https://openalex.org/W4387171542","doi":"https://doi.org/10.3233/faia230623","title":"Adam Accumulation to Reduce Memory Footprints of Both Activations and Gradients for Large-Scale DNN Training","display_name":"Adam Accumulation to Reduce Memory Footprints of Both Activations and Gradients for Large-Scale DNN Training","publication_year":2023,"publication_date":"2023-09-28","ids":{"openalex":"https://openalex.org/W4387171542","doi":"https://doi.org/10.3233/faia230623"},"language":"en","primary_location":{"id":"doi:10.3233/faia230623","is_oa":true,"landing_page_url":"http://dx.doi.org/10.3233/faia230623","pdf_url":"https://ebooks.iospress.nl/pdf/doi/10.3233/FAIA230623","source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"},"type":"book-chapter","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://ebooks.iospress.nl/pdf/doi/10.3233/FAIA230623","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5046020427","display_name":"Yijia Zhang","orcid":"https://orcid.org/0000-0002-5843-4675"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yijia Zhang","raw_affiliation_strings":["Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102518831","display_name":"Yibo Han","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yibo Han","raw_affiliation_strings":["Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101401413","display_name":"Shijie Cao","orcid":"https://orcid.org/0000-0002-5646-2947"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]},{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shijie Cao","raw_affiliation_strings":["Microsoft Research Asia","Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research Asia","institution_ids":["https://openalex.org/I4210113369"]},{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5015946486","display_name":"Guohao Dai","orcid":"https://orcid.org/0000-0003-0849-3252"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Guohao Dai","raw_affiliation_strings":["Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075076538","display_name":"Youshan Miao","orcid":"https://orcid.org/0000-0002-2395-9965"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Youshan Miao","raw_affiliation_strings":["Microsoft Research Asia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research Asia","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101534161","display_name":"Ting Cao","orcid":"https://orcid.org/0000-0002-9107-013X"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ting Cao","raw_affiliation_strings":["Microsoft Research Asia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research Asia","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100346602","display_name":"Fan Yang","orcid":"https://orcid.org/0000-0002-0378-060X"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fan Yang","raw_affiliation_strings":["Microsoft Research Asia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research Asia","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5105346323","display_name":"Ningyi Xu","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Ningyi Xu","raw_affiliation_strings":["Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5105346323"],"corresponding_institution_ids":["https://openalex.org/I183067930"],"apc_list":null,"apc_paid":null,"fwci":1.1319,"has_fulltext":true,"cited_by_count":3,"citation_normalized_percentile":{"value":0.84453409,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":96,"max":97},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9986000061035156,"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":0.9986000061035156,"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":0.998199999332428,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9979000091552734,"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/memory-footprint","display_name":"Memory footprint","score":0.8178113698959351},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.7492164373397827},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6407214403152466},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.4534192681312561},{"id":"https://openalex.org/keywords/reduction","display_name":"Reduction (mathematics)","score":0.424323707818985},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.3604927957057953},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.16811075806617737},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.12889066338539124},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.07599565386772156}],"concepts":[{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.8178113698959351},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.7492164373397827},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6407214403152466},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.4534192681312561},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.424323707818985},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3604927957057953},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.16811075806617737},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.12889066338539124},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.07599565386772156},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","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/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.3233/faia230623","is_oa":true,"landing_page_url":"http://dx.doi.org/10.3233/faia230623","pdf_url":"https://ebooks.iospress.nl/pdf/doi/10.3233/FAIA230623","source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"}],"best_oa_location":{"id":"doi:10.3233/faia230623","is_oa":true,"landing_page_url":"http://dx.doi.org/10.3233/faia230623","pdf_url":"https://ebooks.iospress.nl/pdf/doi/10.3233/FAIA230623","source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4387171542.pdf","grobid_xml":"https://content.openalex.org/works/W4387171542.grobid-xml"},"referenced_works_count":34,"referenced_works":["https://openalex.org/W1522301498","https://openalex.org/W1836465849","https://openalex.org/W2108598243","https://openalex.org/W2148825261","https://openalex.org/W2194775991","https://openalex.org/W2338908902","https://openalex.org/W2763421725","https://openalex.org/W2797328513","https://openalex.org/W2799054028","https://openalex.org/W2896457183","https://openalex.org/W2901299405","https://openalex.org/W2907886210","https://openalex.org/W2923014074","https://openalex.org/W2941620283","https://openalex.org/W2946948417","https://openalex.org/W2949413532","https://openalex.org/W2955425717","https://openalex.org/W2965373594","https://openalex.org/W2971216715","https://openalex.org/W2991040477","https://openalex.org/W3006131567","https://openalex.org/W3025935268","https://openalex.org/W3030163527","https://openalex.org/W3094502228","https://openalex.org/W3121562065","https://openalex.org/W3129831491","https://openalex.org/W3156643189","https://openalex.org/W3174394143","https://openalex.org/W4221167110","https://openalex.org/W4239072543","https://openalex.org/W4288365403","https://openalex.org/W4292779060","https://openalex.org/W4293569541","https://openalex.org/W4299287062"],"related_works":["https://openalex.org/W1657880117","https://openalex.org/W2595172197","https://openalex.org/W2127970246","https://openalex.org/W2084856301","https://openalex.org/W1001352512","https://openalex.org/W4382618745","https://openalex.org/W2885125400","https://openalex.org/W1989889224","https://openalex.org/W2748922771","https://openalex.org/W1987128138"],"abstract_inverted_index":{"Running":[0],"out":[1],"of":[2],"GPU":[3],"memory":[4,18,35,43,125,145,165],"has":[5,22],"become":[6],"a":[7,47,61],"main":[8],"bottleneck":[9],"for":[10,66,148],"large-scale":[11],"DNN":[12],"training.":[13],"How":[14],"to":[15,38,46,123,128,151],"reduce":[16],"the":[17,110],"footprint":[19],"during":[20],"training":[21,137],"received":[23],"intensive":[24],"research":[25],"attention.":[26],"We":[27,103],"find":[28],"that":[29,95],"previous":[30],"gradient":[31,42,78,129],"accumulation":[32,64,130],"reduces":[33],"activation":[34,76],"but":[36],"fails":[37],"be":[39,98],"compatible":[40],"with":[41,131,144,163],"reduction":[44,126,146],"due":[45],"contradiction":[48],"between":[49],"preserving":[50],"gradients":[51,84,96],"and":[52,77,88,105,158],"releasing":[53],"gradients.":[54],"To":[55],"address":[56],"this":[57],"issue,":[58],"we":[59],"propose":[60],"novel":[62],"optimizer":[63,86,90,149],"method":[65],"Adam,":[67],"named":[68],"Adam":[69],"Accumulation":[70],"(AdamA),":[71],"which":[72],"enables":[73],"reducing":[74],"both":[75],"memory.":[79],"Specifically,":[80],"AdamA":[81,108,120,140],"directly":[82],"integrates":[83],"into":[85],"states":[87,91,150],"accumulates":[89],"over":[92,156],"micro-batches,":[93],"so":[94],"can":[97,141],"released":[99],"immediately":[100],"after":[101],"use.":[102],"mathematically":[104],"experimentally":[106],"demonstrate":[107],"yields":[109],"same":[111],"convergence":[112],"properties":[113],"as":[114],"Adam.":[115],"Evaluated":[116],"on":[117,161],"transformer-based":[118],"models,":[119],"achieves":[121],"up":[122],"23%":[124],"compared":[127],"less":[132],"than":[133],"2%":[134],"degradation":[135],"in":[136],"throughput.":[138],"Notably,":[139],"work":[142],"together":[143],"methods":[147],"fit":[152],"1.26\u00d7~3.14\u00d7":[153],"larger":[154],"models":[155],"PyTorch":[157],"DeepSpeed":[159],"baseline":[160],"GPUs":[162],"different":[164],"capacities.":[166]},"counts_by_year":[{"year":2025,"cited_by_count":3}],"updated_date":"2026-05-21T06:26:12.895304","created_date":"2025-10-10T00:00:00"}