{"id":"https://openalex.org/W7070174441","doi":null,"title":"Optimizing checkpointing techniques for machine learning frameworks","display_name":"Optimizing checkpointing techniques for machine learning frameworks","publication_year":2023,"publication_date":"2023-05-15","ids":{"openalex":"https://openalex.org/W7070174441"},"language":"en","primary_location":{"id":"pmh:oai:upcommons.upc.edu:2117/390645","doi":null,"is_oa":true,"landing_page_url":"https://hdl.handle.net/2117/390645","pdf_url":"https://upcommons.upc.edu/bitstreams/219e560d-cdae-4eba-9026-7793516fcedd/download","source":{"id":"https://openalex.org/S4377196262","display_name":"UPCommons institutional repository (Universitat Polit\u00e8cnica de Catalunya)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I9617848","host_organization_name":"Universitat Polit\u00e8cnica de Catalunya","host_organization_lineage":["https://openalex.org/I9617848"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Master thesis"},"volume":null,"type":"dissertation","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://upcommons.upc.edu/bitstreams/219e560d-cdae-4eba-9026-7793516fcedd/download","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Perell\u00f3 Bacardit, Marc","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Perell\u00f3 Bacardit, Marc","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":0,"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":true,"primary_topic":{"id":"https://openalex.org/T10772","display_name":"Distributed systems and fault tolerance","score":0.8169000148773193,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10772","display_name":"Distributed systems and fault tolerance","score":0.8169000148773193,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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":0.07760000228881836,"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/T10101","display_name":"Cloud Computing and Resource Management","score":0.032999999821186066,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/distributed-checkpointing","display_name":"distributed checkpointing","score":0.6669999957084656},{"id":"https://openalex.org/keywords/deep-learning-training","display_name":"deep learning training","score":0.5410000085830688},{"id":"https://openalex.org/keywords/parallel-file-systems","display_name":"parallel file systems","score":0.5230000019073486},{"id":"https://openalex.org/keywords/i-o-bottleneck","display_name":"I/O bottleneck","score":0.8870000243186951},{"id":"https://openalex.org/keywords/local-storage","display_name":"local storage","score":0.6060000061988831},{"id":"https://openalex.org/keywords/pytorch","display_name":"PyTorch","score":0.8519999980926514},{"id":"https://openalex.org/keywords/speedup-analysis","display_name":"speedup analysis","score":0.46000000834465027},{"id":"https://openalex.org/keywords/python-module","display_name":"Python module","score":0.4580000042915344}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8690999746322632},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6572999954223633},{"id":"https://openalex.org/C42812","wikidata":"https://www.wikidata.org/wiki/Q1082910","display_name":"Partition (number theory)","level":2,"score":0.5303999781608582},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4496000111103058},{"id":"https://openalex.org/C519991488","wikidata":"https://www.wikidata.org/wiki/Q28865","display_name":"Python (programming language)","level":2,"score":0.4260999858379364},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.37549999356269836},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.32850000262260437},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3260999917984009},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3249000012874603},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.30000001192092896},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.29019999504089355},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.2849000096321106},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.2635999917984009},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2567000091075897}],"mesh":[],"locations_count":1,"locations":[{"id":"pmh:oai:upcommons.upc.edu:2117/390645","doi":null,"is_oa":true,"landing_page_url":"https://hdl.handle.net/2117/390645","pdf_url":"https://upcommons.upc.edu/bitstreams/219e560d-cdae-4eba-9026-7793516fcedd/download","source":{"id":"https://openalex.org/S4377196262","display_name":"UPCommons institutional repository (Universitat Polit\u00e8cnica de Catalunya)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I9617848","host_organization_name":"Universitat Polit\u00e8cnica de Catalunya","host_organization_lineage":["https://openalex.org/I9617848"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Master thesis"}],"best_oa_location":{"id":"pmh:oai:upcommons.upc.edu:2117/390645","doi":null,"is_oa":true,"landing_page_url":"https://hdl.handle.net/2117/390645","pdf_url":"https://upcommons.upc.edu/bitstreams/219e560d-cdae-4eba-9026-7793516fcedd/download","source":{"id":"https://openalex.org/S4377196262","display_name":"UPCommons institutional repository (Universitat Polit\u00e8cnica de Catalunya)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I9617848","host_organization_name":"Universitat Polit\u00e8cnica de Catalunya","host_organization_lineage":["https://openalex.org/I9617848"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Master thesis"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7070174441.pdf","grobid_xml":"https://content.openalex.org/works/W7070174441.grobid-xml"},"referenced_works_count":19,"referenced_works":["https://openalex.org/W1981432246","https://openalex.org/W2592929672","https://openalex.org/W2612445135","https://openalex.org/W2745999522","https://openalex.org/W2767346922","https://openalex.org/W2769569970","https://openalex.org/W2787998955","https://openalex.org/W2884159461","https://openalex.org/W2896457183","https://openalex.org/W2911892981","https://openalex.org/W2949650786","https://openalex.org/W2951947775","https://openalex.org/W2980282514","https://openalex.org/W2981749440","https://openalex.org/W3021124033","https://openalex.org/W3030163527","https://openalex.org/W3043802286","https://openalex.org/W3107864711","https://openalex.org/W3187255235"],"related_works":["https://openalex.org/W3107864711","https://openalex.org/W4214854835","https://openalex.org/W3021124033","https://openalex.org/W7125957433","https://openalex.org/W4403007766","https://openalex.org/W4406014579","https://openalex.org/W4404171822","https://openalex.org/W7129068819","https://openalex.org/W4400141586","https://openalex.org/W4416203804"],"abstract_inverted_index":{"While":[0,23],"most":[1,108],"deep":[2,99],"learning":[3,100],"frameworks":[4],"provide":[5],"mechanisms":[6],"to":[7,86,93,113,126,147,163,187,203,210,227,232,237,254,262,267,288,315,323,392,409],"checkpoint":[8,57,68,129,139,152,292,300,319],"models,":[9],"their":[10,233],"implementation":[11,25],"is":[12,82,352,368,390],"naive":[13],"and":[14,70,259,270,304,308,362,367,376,398,430],"based":[15,433],"on":[16,354,403,434],"the":[17,41,64,88,143,165,174,184,195,199,246,346,355,358,363,395,399,435],"assumption":[18],"of":[19,63,167,194,198,249,279,285,302,340,349,365,401,438],"single":[20],"machine":[21],"training.":[22,42],"this":[24,115],"can":[26],"be":[27,211,228],"adequate":[28],"at":[29,49],"a":[30,94,120,137,149,168,179,276,289,299,316,350,373,414],"small":[31],"scale,":[32],"it":[33,135,389],"becomes":[34],"progressively":[35],"inefficient":[36],"when":[37,370,383],"further":[38],"scaling":[39],"out":[40],"Since":[43],"large":[44,50,428,436],"models":[45],"are":[46],"often":[47],"trained":[48],"scale":[51],"in":[52,78,84,235,407,441],"HPC":[53,76,109,151],"environments,":[54,77],"we":[55,117,206,281,297,328,416,418],"need":[56],"procedures":[58],"which":[59,79,154],"make":[60],"efficient":[61],"use":[62],"available":[65],"resources.":[66],"Several":[67],"techniques":[69],"libraries":[71],"have":[72],"been":[73],"designed":[74],"for":[75,181],"local":[80,156],"storage":[81],"leveraged":[83],"order":[85,112,236,322,408],"alleviate":[87],"I/O":[89],"bottleneck":[90],"from":[91],"writing":[92],"parallel":[95,257],"file":[96],"system.":[97],"However,":[98],"training":[101],"has":[102],"very":[103,427],"different":[104,338,442],"data":[105,130,146,335,386,422],"requirements":[106],"than":[107],"applications.":[110],"In":[111,321],"solve":[114],"problem,":[116],"develop":[118],"DeepPart,":[119],"python":[121],"module":[122],"that":[123,245,345],"provides":[124],"optimizations":[125],"distribute":[127,164],"shared":[128],"across":[131,170,387],"processes,":[132,388],"effectively":[133],"transforming":[134],"into":[136],"HPC-like":[138],"procedure.":[140],"DeepPart":[141],"sends":[142],"resulting":[144],"distributed":[145,310],"FTI,":[148],"multi-level":[150],"library":[153],"leverages":[155],"storage.":[157],"We":[158,177,224,243],"implement":[159],"an":[160,255,330,378],"heuristic":[161],"algorithm":[162,220],"elements":[166,209],"collection":[169,201,241],"processes":[171,217,272],"while":[172],"minimizing":[173],"computational":[175,247],"cost.":[176,412],"devise":[178],"method":[180],"automatically":[182],"choosing":[183],"best":[185],"sub-collection":[186],"partition,":[188],"referred":[189],"as":[190],"partition":[191,239],"candidate,":[192],"independently":[193],"specific":[196],"structure":[197],"main":[200],"passed":[202],"checkpoint.":[204],"Additionally,":[205],"allow":[207,225],"individual":[208],"partitioned":[212],"between":[213,357],"two":[214],"or":[215],"more":[216],"if":[218],"our":[219,250,421],"detects":[221],"size":[222,278,361,397],"imbalance.":[223],"sub-collections":[226],"recursively":[229],"partitioned,":[230],"proportionally":[231,406],"size,":[234],"efficiently":[238],"non-trivial":[240],"structures.":[242],"show":[244,344,419],"cost":[248,348],"approach":[251,424],"behaves":[252],"similarly":[253],"embarrassingly":[256],"workload,":[258],"achieves":[260],"close":[261],"ideal":[263],"speed-ups":[264],"with":[265,309,337,426],"up":[266],"16":[268],"nodes":[269],"4":[271],"per":[273],"node.":[274],"With":[275],"model":[277,325,334,351,360,385,429,440],"20GB,":[280],"observe":[282],"overall":[283],"gains":[284],"5.6x":[286],"compared":[287,314],"standard":[290,317],"PyTorch":[291,318],"implementation.":[293],"Using":[294,413],"BERT-LARGE":[295],"model,":[296],"obtain":[298],"speed-up":[301],"2.1x":[303],"2.7x":[305],"without":[306],"compression":[307,312],"lossless":[311],"respectively":[313],"approach.":[320],"understand":[324],"serialization":[326,347,411],"cost,":[327],"perform":[329],"analysis":[331],"using":[332],"several":[333],"sizes":[336],"number":[339,364,400],"tensors.":[341],"Our":[342],"findings":[343],"dependant":[353],"relation":[356],"total":[359,396],"tensors,":[366],"optimal":[369],"being":[371],"above":[372],"lower":[374],"threshold":[375],"below":[377],"upper":[379],"threshold.":[380],"As":[381],"such,":[382],"distributing":[384],"important":[391],"reduce":[393],"both":[394],"tensors":[402],"each":[404],"process":[405],"minimize":[410],"simulator":[415],"designed,":[417],"how":[420],"distribution":[423],"scales":[425],"optimizer":[431],"structures":[432],"variant":[437],"BERT":[439],"configurations.":[443]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-10T00:00:00"}