{"id":"https://openalex.org/W4401977362","doi":"https://doi.org/10.1145/3689338","title":"DELTA: Memory-Efficient Training via Dynamic Fine-Grained Recomputation and Swapping","display_name":"DELTA: Memory-Efficient Training via Dynamic Fine-Grained Recomputation and Swapping","publication_year":2024,"publication_date":"2024-08-20","ids":{"openalex":"https://openalex.org/W4401977362","doi":"https://doi.org/10.1145/3689338"},"language":"en","primary_location":{"id":"doi:10.1145/3689338","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3689338","pdf_url":null,"source":{"id":"https://openalex.org/S26056741","display_name":"ACM Transactions on Architecture and Code Optimization","issn_l":"1544-3566","issn":["1544-3566","1544-3973"],"is_oa":true,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Architecture and Code Optimization","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1145/3689338","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102970145","display_name":"Yu Tang","orcid":"https://orcid.org/0000-0002-8595-1547"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Yu Tang","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100393695","display_name":"Qiao Li","orcid":"https://orcid.org/0000-0002-4579-4268"},"institutions":[{"id":"https://openalex.org/I191208505","display_name":"Xiamen University","ror":"https://ror.org/00mcjh785","country_code":"CN","type":"education","lineage":["https://openalex.org/I191208505"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qiao Li","raw_affiliation_strings":["Xiamen University, Xiamen, China"],"affiliations":[{"raw_affiliation_string":"Xiamen University, Xiamen, China","institution_ids":["https://openalex.org/I191208505"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029674340","display_name":"Lujia Yin","orcid":"https://orcid.org/0009-0005-1494-2853"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Lujia Yin","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100440903","display_name":"Dongsheng Li","orcid":"https://orcid.org/0000-0001-9743-2034"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Dongsheng Li","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100395351","display_name":"Yiming Zhang","orcid":"https://orcid.org/0000-0001-6450-8485"},"institutions":[{"id":"https://openalex.org/I191208505","display_name":"Xiamen University","ror":"https://ror.org/00mcjh785","country_code":"CN","type":"education","lineage":["https://openalex.org/I191208505"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yiming Zhang","raw_affiliation_strings":["Xiamen University, Xiamen, China"],"affiliations":[{"raw_affiliation_string":"Xiamen University, Xiamen, China","institution_ids":["https://openalex.org/I191208505"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077573056","display_name":"C.X. Wang","orcid":"https://orcid.org/0009-0006-8122-498X"},"institutions":[{"id":"https://openalex.org/I4210128910","display_name":"Group Sense (China)","ror":"https://ror.org/036wd5777","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128910"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chenyu Wang","raw_affiliation_strings":["Sensetime, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"Sensetime, Shanghai, China","institution_ids":["https://openalex.org/I4210128910"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102773472","display_name":"Xingcheng Zhang","orcid":"https://orcid.org/0009-0006-8525-0608"},"institutions":[{"id":"https://openalex.org/I4210100255","display_name":"Beijing Academy of Artificial Intelligence","ror":"https://ror.org/016a74861","country_code":"CN","type":"other","lineage":["https://openalex.org/I4210100255"]},{"id":"https://openalex.org/I4391012619","display_name":"Shanghai Artificial Intelligence Laboratory","ror":"https://ror.org/03wkvpx79","country_code":null,"type":"facility","lineage":["https://openalex.org/I4391012619"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xingcheng Zhang","raw_affiliation_strings":["Shanghai Artificial Intelligence Laboratory, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"Shanghai Artificial Intelligence Laboratory, Shanghai, China","institution_ids":["https://openalex.org/I4210100255","https://openalex.org/I4391012619"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043045076","display_name":"Linbo Qiao","orcid":"https://orcid.org/0000-0002-8285-2738"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Linbo Qiao","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050642628","display_name":"Zhaoning Zhang","orcid":"https://orcid.org/0000-0001-7518-1385"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhaoning Zhang","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100625354","display_name":"Kai L\u00fc","orcid":"https://orcid.org/0000-0002-6378-7002"},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Kai Lu","raw_affiliation_strings":["National University of Defense Technology, Changsha, China"],"affiliations":[{"raw_affiliation_string":"National University of Defense Technology, Changsha, China","institution_ids":["https://openalex.org/I170215575"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5102970145"],"corresponding_institution_ids":["https://openalex.org/I170215575"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.13158667,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"21","issue":"4","first_page":"1","last_page":"25"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9994000196456909,"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":0.9994000196456909,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9965000152587891,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.9962999820709229,"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":0.7386361360549927},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.5046094655990601},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.46282750368118286},{"id":"https://openalex.org/keywords/delta","display_name":"Delta","score":0.4284262955188751},{"id":"https://openalex.org/keywords/computational-science","display_name":"Computational science","score":0.3569110035896301},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.11122313141822815}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7386361360549927},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5046094655990601},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.46282750368118286},{"id":"https://openalex.org/C5072461","wikidata":"https://www.wikidata.org/wiki/Q49506","display_name":"Delta","level":2,"score":0.4284262955188751},{"id":"https://openalex.org/C459310","wikidata":"https://www.wikidata.org/wiki/Q117801","display_name":"Computational science","level":1,"score":0.3569110035896301},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.11122313141822815},{"id":"https://openalex.org/C1276947","wikidata":"https://www.wikidata.org/wiki/Q333","display_name":"Astronomy","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:10.1145/3689338","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3689338","pdf_url":null,"source":{"id":"https://openalex.org/S26056741","display_name":"ACM Transactions on Architecture and Code Optimization","issn_l":"1544-3566","issn":["1544-3566","1544-3973"],"is_oa":true,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Architecture and Code Optimization","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1145/3689338","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3689338","pdf_url":null,"source":{"id":"https://openalex.org/S26056741","display_name":"ACM Transactions on Architecture and Code Optimization","issn_l":"1544-3566","issn":["1544-3566","1544-3973"],"is_oa":true,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Architecture and Code Optimization","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G3606055544","display_name":null,"funder_award_id":"62025208","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":51,"referenced_works":["https://openalex.org/W1548885290","https://openalex.org/W1583837637","https://openalex.org/W1776713878","https://openalex.org/W2009443800","https://openalex.org/W2072803231","https://openalex.org/W2074629033","https://openalex.org/W2089047615","https://openalex.org/W2148150086","https://openalex.org/W2163605009","https://openalex.org/W2194775991","https://openalex.org/W2338908902","https://openalex.org/W2489529491","https://openalex.org/W2531523563","https://openalex.org/W2602633191","https://openalex.org/W2612690371","https://openalex.org/W2751343396","https://openalex.org/W2790501674","https://openalex.org/W2963073398","https://openalex.org/W2963748441","https://openalex.org/W2964350391","https://openalex.org/W2969388332","https://openalex.org/W2970587001","https://openalex.org/W2970971581","https://openalex.org/W2973727699","https://openalex.org/W2979044977","https://openalex.org/W3006131567","https://openalex.org/W3012479151","https://openalex.org/W3012514909","https://openalex.org/W3035078899","https://openalex.org/W3037585619","https://openalex.org/W3037639655","https://openalex.org/W3081168214","https://openalex.org/W3102511045","https://openalex.org/W3124474054","https://openalex.org/W3138516171","https://openalex.org/W3202489111","https://openalex.org/W3204105967","https://openalex.org/W3205803342","https://openalex.org/W4213019189","https://openalex.org/W4221100183","https://openalex.org/W4292779060","https://openalex.org/W4295312788","https://openalex.org/W4301183483","https://openalex.org/W4301361180","https://openalex.org/W4306873623","https://openalex.org/W4320495408","https://openalex.org/W4327694855","https://openalex.org/W4387933214","https://openalex.org/W6703652217","https://openalex.org/W6767541031","https://openalex.org/W6778883912"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4221046490","https://openalex.org/W4297616267","https://openalex.org/W2502651140","https://openalex.org/W3195406774","https://openalex.org/W4297615481","https://openalex.org/W4297616317","https://openalex.org/W3111070561"],"abstract_inverted_index":{"To":[0],"accommodate":[1],"the":[2,84,109,116,133,155,166,205],"increasingly":[3],"large-scale":[4,60,227],"models":[5],"within":[6],"limited-capacity":[7],"GPU":[8,172,231],"memory,":[9],"various":[10,195],"coarse-grained":[11],"techniques,":[12],"such":[13],"as":[14,90],"recomputation":[15,53,142],"and":[16,52,68,200,209,217,223],"swapping,":[17],"have":[18,28],"been":[19],"proposed":[20],"to":[21,44,71,107,115,138,176,187],"optimize":[22],"memory":[23,36,66,73,128,173,182],"usage.":[24],"However,":[25],"these":[26],"methods":[27],"encountered":[29],"limitations,":[30],"either":[31],"in":[32,170,225],"terms":[33],"of":[34,86,118,168,207],"inefficient":[35],"reduction":[37],"or":[38,143,160],"diminished":[39],"training":[40,62,78,135,151,206,229],"performance.":[41],"In":[42],"response":[43],"this,":[45],"our":[46,119],"article":[47],"introduces":[48],"dynamic":[49,127],"tensor":[50],"offloading":[51],"(DELTA),":[54],"an":[55,91,101,149],"innovative":[56],"approach":[57],"for":[58,157,194],"memory-efficient":[59],"model":[61,134,228],"that":[63,130],"combines":[64],"fine-grained":[65],"optimization":[67,89],"prefetching":[69,124],"technology":[70,125],"reduce":[72],"usage":[74],"while":[75,189],"maintaining":[76,190],"high":[77],"throughput":[79],"concurrently.":[80],"Initially,":[81],"we":[82,99,113],"formulate":[83],"problem":[85,110],"memory-throughput":[87],"joint":[88],"easy-solving":[92],"0/1":[93],"Knapsack":[94],"problem.":[95],"Leveraging":[96],"this":[97],"formalization,":[98],"use":[100],"improving":[102],"polynomial":[103],"complexity":[104],"heuristic":[105],"algorithm":[106],"address":[108],"effectively.":[111],"Furthermore,":[112],"introduce,":[114],"best":[117],"knowledge,":[120],"a":[121],"novel":[122],"bidirectional":[123],"into":[126],"management":[129],"significantly":[131],"accelerates":[132],"when":[136],"compared":[137],"relying":[139],"solely":[140],"on":[141,230],"swapping.":[144],"Finally,":[145],"DELTA":[146,169,179,203],"offers":[147],"users":[148],"automated":[150],"execution":[152],"library,":[153],"eliminating":[154],"need":[156],"manual":[158],"configuration":[159],"specialized":[161],"expertise.":[162],"Experimental":[163],"results":[164],"demonstrate":[165],"effectiveness":[167],"reducing":[171],"consumption.":[174],"Compared":[175],"state-of-the-art":[177],"methods,":[178],"achieves":[180],"substantial":[181],"savings":[183],"ranging":[184],"from":[185],"40%":[186],"72%,":[188],"comparable":[191],"convergence":[192],"performance":[193],"models,":[196],"including":[197],"ResNet-50,":[198],"ResNet-101,":[199],"BERT-Large.":[201],"Notably,":[202],"enables":[204],"GPT2-Large":[208],"GPT2-XL":[210],"with":[211],"batch":[212],"sizes":[213],"increased":[214],"by":[215],"5.5\u00d7":[216],"6\u00d7,":[218],"respectively,":[219],"showcasing":[220],"its":[221],"versatility":[222],"practicality":[224],"enabling":[226],"hardware.":[232]},"counts_by_year":[],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2025-10-10T00:00:00"}