{"id":"https://openalex.org/W2566079294","doi":"https://doi.org/10.1145/3446776","title":"Understanding deep learning (still) requires rethinking generalization","display_name":"Understanding deep learning (still) requires rethinking generalization","publication_year":2021,"publication_date":"2021-02-22","ids":{"openalex":"https://openalex.org/W2566079294","doi":"https://doi.org/10.1145/3446776","mag":"2566079294"},"language":"en","primary_location":{"id":"doi:10.1145/3446776","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3446776","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3446776","source":{"id":"https://openalex.org/S103482838","display_name":"Communications of the ACM","issn_l":"0001-0782","issn":["0001-0782","1557-7317"],"is_oa":false,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Communications of the ACM","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3446776","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5016054994","display_name":"Chiyuan Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Chiyuan Zhang","raw_affiliation_strings":["Google Brain, Mountain View, CA"],"affiliations":[{"raw_affiliation_string":"Google Brain, Mountain View, CA","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017529415","display_name":"Samy Bengio","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Samy Bengio","raw_affiliation_strings":["Google Brain, Mountain View, CA"],"affiliations":[{"raw_affiliation_string":"Google Brain, Mountain View, CA","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039915143","display_name":"Moritz Hardt","orcid":"https://orcid.org/0009-0000-7694-3038"},"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":"Moritz Hardt","raw_affiliation_strings":["University of California, Berkeley, CA"],"affiliations":[{"raw_affiliation_string":"University of California, Berkeley, CA","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012870568","display_name":"Benjamin Recht","orcid":"https://orcid.org/0000-0002-0293-593X"},"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":"Benjamin Recht","raw_affiliation_strings":["University of California, Berkeley, CA"],"affiliations":[{"raw_affiliation_string":"University of California, Berkeley, CA","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5003562101","display_name":"Oriol Vinyals","orcid":"https://orcid.org/0000-0001-7848-7283"},"institutions":[{"id":"https://openalex.org/I4210090411","display_name":"DeepMind (United Kingdom)","ror":"https://ror.org/00971b260","country_code":"GB","type":"company","lineage":["https://openalex.org/I4210090411","https://openalex.org/I4210128969"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Oriol Vinyals","raw_affiliation_strings":["DeepMind, London N1C 4AG, U.K"],"affiliations":[{"raw_affiliation_string":"DeepMind, London N1C 4AG, U.K","institution_ids":["https://openalex.org/I4210090411"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5016054994"],"corresponding_institution_ids":["https://openalex.org/I1291425158"],"apc_list":null,"apc_paid":null,"fwci":213.7634,"has_fulltext":true,"cited_by_count":2223,"citation_normalized_percentile":{"value":0.99991609,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":90,"max":100},"biblio":{"volume":"64","issue":"3","first_page":"107","last_page":"115"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9998000264167786,"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"}},"topics":[{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9998000264167786,"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/T12814","display_name":"Gaussian Processes and Bayesian Inference","score":0.9984999895095825,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","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/regularization","display_name":"Regularization (linguistics)","score":0.7545468807220459},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7414992451667786},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.6969637274742126},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.6522877216339111},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.5938319563865662},{"id":"https://openalex.org/keywords/deep-neural-networks","display_name":"Deep neural networks","score":0.5848677754402161},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.5814019441604614},{"id":"https://openalex.org/keywords/early-stopping","display_name":"Early stopping","score":0.5390787124633789},{"id":"https://openalex.org/keywords/convolutional-neural-network","display_name":"Convolutional neural network","score":0.49897265434265137},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.4719128906726837},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"Stochastic gradient descent","score":0.4570103585720062},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.42039021849632263},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.22549289464950562}],"concepts":[{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.7545468807220459},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7414992451667786},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6969637274742126},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6522877216339111},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5938319563865662},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.5848677754402161},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.5814019441604614},{"id":"https://openalex.org/C5465570","wikidata":"https://www.wikidata.org/wiki/Q5326898","display_name":"Early stopping","level":3,"score":0.5390787124633789},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.49897265434265137},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4719128906726837},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.4570103585720062},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.42039021849632263},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.22549289464950562},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3446776","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3446776","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3446776","source":{"id":"https://openalex.org/S103482838","display_name":"Communications of the ACM","issn_l":"0001-0782","issn":["0001-0782","1557-7317"],"is_oa":false,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Communications of the ACM","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1145/3446776","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3446776","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3446776","source":{"id":"https://openalex.org/S103482838","display_name":"Communications of the ACM","issn_l":"0001-0782","issn":["0001-0782","1557-7317"],"is_oa":false,"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":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Communications of the ACM","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320309327","display_name":"Google","ror":"https://ror.org/00njsd438"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W2566079294.pdf","grobid_xml":"https://content.openalex.org/works/W2566079294.grobid-xml"},"referenced_works_count":53,"referenced_works":["https://openalex.org/W27434444","https://openalex.org/W180242331","https://openalex.org/W1487202328","https://openalex.org/W1540155273","https://openalex.org/W1726806267","https://openalex.org/W1899249567","https://openalex.org/W1981276685","https://openalex.org/W2006240266","https://openalex.org/W2034978228","https://openalex.org/W2082233060","https://openalex.org/W2095705004","https://openalex.org/W2099579348","https://openalex.org/W2103496339","https://openalex.org/W2127265926","https://openalex.org/W2139338362","https://openalex.org/W2141473882","https://openalex.org/W2143859313","https://openalex.org/W2148603752","https://openalex.org/W2166198634","https://openalex.org/W2513671774","https://openalex.org/W2604117713","https://openalex.org/W2604272474","https://openalex.org/W2682189153","https://openalex.org/W2709553318","https://openalex.org/W2732724430","https://openalex.org/W2741952635","https://openalex.org/W2766196653","https://openalex.org/W2767382501","https://openalex.org/W2807842867","https://openalex.org/W2889599215","https://openalex.org/W2911742574","https://openalex.org/W2947451636","https://openalex.org/W2951013084","https://openalex.org/W2962742960","https://openalex.org/W2962990163","https://openalex.org/W2963069632","https://openalex.org/W2963236897","https://openalex.org/W2963518130","https://openalex.org/W2963664410","https://openalex.org/W2963739978","https://openalex.org/W2963742538","https://openalex.org/W2963759780","https://openalex.org/W2963794891","https://openalex.org/W2964056216","https://openalex.org/W3000680563","https://openalex.org/W3119586787","https://openalex.org/W3137695714","https://openalex.org/W4285719527","https://openalex.org/W4299785371","https://openalex.org/W6608518175","https://openalex.org/W6659529383","https://openalex.org/W6740483536","https://openalex.org/W6745276634"],"related_works":["https://openalex.org/W4287755480","https://openalex.org/W2785875001","https://openalex.org/W4283332751","https://openalex.org/W4220659530","https://openalex.org/W1535041666","https://openalex.org/W4206677161","https://openalex.org/W2417932737","https://openalex.org/W3013844214","https://openalex.org/W2141403321","https://openalex.org/W4367060723"],"abstract_inverted_index":{"Despite":[0],"their":[1],"massive":[2],"size,":[3],"successful":[4],"deep":[5],"artificial":[6],"neural":[7,56,123],"networks":[8,57,69,124],"can":[9],"exhibit":[10],"a":[11,80,115,165],"remarkably":[12],"small":[13,23],"gap":[14],"between":[15],"training":[16,85],"and":[17,95],"test":[18],"performance.":[19],"Conventional":[20],"wisdom":[21],"attributes":[22],"generalization":[24],"error":[25],"either":[26],"to":[27,34,52],"properties":[28],"of":[29,83,136,141,181],"the":[30,35,84,101,134,139,169,175,178],"model":[31],"family":[32],"or":[33],"regularization":[36,94],"techniques":[37],"used":[38],"during":[39],"training.":[40],"Through":[41],"extensive":[42],"systematic":[43],"experiments,":[44],"we":[45,99],"show":[46],"how":[47],"these":[48,111],"traditional":[49,158],"approaches":[50],"fail":[51],"explain":[53],"why":[54],"large":[55],"generalize":[58],"well":[59],"in":[60,148,174],"practice.":[61,149],"Specifically,":[62],"our":[63,152],"experiments":[64],"establish":[65],"that":[66,119],"state-of-the-art":[67],"convolutional":[68],"for":[70],"image":[71],"classification":[72],"trained":[73],"with":[74,114,157,164],"stochastic":[75],"gradient":[76],"methods":[77],"easily":[78],"fit":[79],"random":[81,107],"labeling":[82],"data.":[86],"This":[87],"phenomenon":[88],"is":[89],"qualitatively":[90],"unaffected":[91],"by":[92,104,155],"explicit":[93],"occurs":[96],"even":[97],"if":[98],"replace":[100],"true":[102],"images":[103],"completely":[105],"unstructured":[106],"noise.":[108],"We":[109,150,160],"corroborate":[110],"experimental":[112,153],"findings":[113,154],"theoretical":[116],"construction":[117],"showing":[118],"simple":[120],"depth":[121],"two":[122],"already":[125],"have":[126],"perfect":[127],"finite":[128],"sample":[129],"expressivity":[130],"as":[131,133,144],"soon":[132],"number":[135,140],"parameters":[137],"exceeds":[138],"data":[142],"points":[143],"it":[145],"usually":[146],"does":[147],"interpret":[151],"comparison":[156],"models.":[159],"supplement":[161],"this":[162,182],"republication":[163],"new":[166],"section":[167],"at":[168],"end":[170],"summarizing":[171],"recent":[172],"progresses":[173],"field":[176],"since":[177],"original":[179],"version":[180],"paper.":[183]},"counts_by_year":[{"year":2026,"cited_by_count":101},{"year":2025,"cited_by_count":552},{"year":2024,"cited_by_count":589},{"year":2023,"cited_by_count":509},{"year":2022,"cited_by_count":317},{"year":2021,"cited_by_count":110},{"year":2020,"cited_by_count":14},{"year":2019,"cited_by_count":16},{"year":2018,"cited_by_count":7},{"year":2017,"cited_by_count":7},{"year":2016,"cited_by_count":1}],"updated_date":"2026-04-14T08:04:32.555800","created_date":"2025-10-10T00:00:00"}