{"id":"https://openalex.org/W3216654142","doi":null,"title":"Implicit Data-Driven Regularization in Deep Neural Networks under SGD","display_name":"Implicit Data-Driven Regularization in Deep Neural Networks under SGD","publication_year":2021,"publication_date":"2021-11-26","ids":{"openalex":"https://openalex.org/W3216654142","mag":"3216654142"},"language":"en","primary_location":{"id":"mag:3216654142","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/2111.13331","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2111.13331","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078135437","display_name":"Xuran Meng","orcid":"https://orcid.org/0000-0001-7265-1521","observed_orcids":["https://orcid.org/0000-0001-7265-1521"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuran Meng","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0001-7265-1521","affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101512985","display_name":"Jianfeng Yao","orcid":"https://orcid.org/0000-0003-3147-2775","observed_orcids":["https://orcid.org/0000-0003-3147-2775"]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Jianfeng Yao","raw_affiliation_strings":["UNIVERSITY OF HONG KONG"],"raw_orcid":"https://orcid.org/0000-0003-3147-2775","affiliations":[{"raw_affiliation_string":"UNIVERSITY OF HONG KONG","institution_ids":["https://openalex.org/I889458895"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.8090999722480774,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.8090999722480774,"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/T10320","display_name":"Neural Networks and Applications","score":0.12780000269412994,"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/T11716","display_name":"Random Matrices and Applications","score":0.010099999606609344,"subfield":{"id":"https://openalex.org/subfields/2613","display_name":"Statistics and Probability"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/data-driven-regularization","display_name":"data-driven regularization","score":0.8149999976158142},{"id":"https://openalex.org/keywords/implicit-regularization","display_name":"implicit regularization","score":0.652999997138977},{"id":"https://openalex.org/keywords/random-matrix-theory","display_name":"random matrix theory","score":0.781000018119812},{"id":"https://openalex.org/keywords/spectral-analysis","display_name":"spectral analysis","score":0.6240000128746033},{"id":"https://openalex.org/keywords/marcenko-pastur-distribution","display_name":"Marcenko-Pastur distribution","score":0.9120000004768372},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"stochastic gradient descent","score":0.8149999976158142},{"id":"https://openalex.org/keywords/early-stopping","display_name":"early stopping","score":0.9509999752044678},{"id":"https://openalex.org/keywords/generalization","display_name":"generalization","score":0.8199999928474426},{"id":"https://openalex.org/keywords/deep-neural-network","display_name":"deep neural network","score":0.8460000157356262}],"concepts":[{"id":"https://openalex.org/C190502265","wikidata":"https://www.wikidata.org/wiki/Q17069496","display_name":"MNIST database","level":3,"score":0.6694095730781555},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.666697084903717},{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.6185104250907898},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.6108847260475159},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5673795342445374},{"id":"https://openalex.org/C163716315","wikidata":"https://www.wikidata.org/wiki/Q901177","display_name":"Gaussian","level":2,"score":0.4859010577201843},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.4690297842025757},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.43855294585227966},{"id":"https://openalex.org/C4839761","wikidata":"https://www.wikidata.org/wiki/Q212111","display_name":"Spectral line","level":2,"score":0.4321420192718506},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.42566120624542236},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.42418861389160156},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4045407772064209},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.28882116079330444},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.11333954334259033},{"id":"https://openalex.org/C1276947","wikidata":"https://www.wikidata.org/wiki/Q333","display_name":"Astronomy","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}],"mesh":[],"locations_count":1,"locations":[{"id":"mag:3216654142","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/2111.13331","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null}],"best_oa_location":{"id":"mag:3216654142","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/2111.13331","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":22,"referenced_works":["https://openalex.org/W1573149518","https://openalex.org/W2065424480","https://openalex.org/W2112796928","https://openalex.org/W2163605009","https://openalex.org/W2622263826","https://openalex.org/W2752366553","https://openalex.org/W2752851182","https://openalex.org/W2766151966","https://openalex.org/W2766678531","https://openalex.org/W2914462410","https://openalex.org/W2919115771","https://openalex.org/W2947654433","https://openalex.org/W2963959597","https://openalex.org/W3006197067","https://openalex.org/W3035064176","https://openalex.org/W3065974826","https://openalex.org/W3095618168","https://openalex.org/W3116687755","https://openalex.org/W3138582970","https://openalex.org/W3170871273","https://openalex.org/W3176544167","https://openalex.org/W3191734561"],"related_works":["https://openalex.org/W2907762094","https://openalex.org/W3191734561","https://openalex.org/W4226320345","https://openalex.org/W4386937939","https://openalex.org/W2940107683","https://openalex.org/W4392271447","https://openalex.org/W2618017917","https://openalex.org/W7162436653","https://openalex.org/W7109197596","https://openalex.org/W2970176896"],"abstract_inverted_index":{"Much":[0],"research":[1],"effort":[2],"has":[3],"been":[4],"devoted":[5],"to":[6,21,47,72,113,135],"explaining":[7],"the":[8,48,74,114,119,124,130,136,172,183,186,190],"success":[9],"of":[10,26,76,116,126,132,192],"deep":[11,34],"learning.":[12],"Random":[13],"Matrix":[14],"Theory":[15],"(RMT)":[16],"provides":[17],"an":[18,167],"emerging":[19],"way":[20],"this":[22,54],"end:":[23],"spectral":[24,163],"analysis":[25],"large":[27],"random":[28],"matrices":[29,41,44,62],"involved":[30],"in":[31,63,118,145],"a":[32,162,204],"trained":[33],"neural":[35],"network":[36],"(DNN)":[37],"such":[38],"as":[39],"weight":[40,61],"or":[42],"Hessian":[43],"with":[45,96],"respect":[46],"stochastic":[49],"gradient":[50],"descent":[51],"algorithm.":[52],"In":[53],"paper,":[55],"we":[56,160],"conduct":[57],"extensive":[58],"experiments":[59],"on":[60,129],"different":[64],"modules,":[65],"e.g.,":[66],"layers,":[67],"networks":[68],"and":[69,101,152,157,165,189],"data":[70,133,151,154,180],"sets,":[71],"analyze":[73],"evolution":[75],"their":[77],"spectra.":[78],"We":[79,121],"find":[80],"that":[81,123],"these":[82,107],"spectra":[83,109,187],"can":[84],"be":[85],"classified":[86],"into":[87],"three":[88],"main":[89],"types:":[90],"Mar\\v{c}enko-Pastur":[91,94],"spectrum":[92,95,104],"(MP),":[93],"few":[97],"bleeding":[98],"outliers":[99],"(MPB),":[100],"Heavy":[102],"tailed":[103],"(HT).":[105],"Moreover,":[106],"discovered":[108],"are":[110,143],"directly":[111],"connected":[112],"degree":[115,125],"regularization":[117,127],"DNN.":[120],"argue":[122],"depends":[128],"quality":[131],"fed":[134],"DNN,":[137],"namely":[138],"Data-Driven":[139],"Regularization.":[140],"These":[141],"findings":[142],"validated":[144],"several":[146],"NNs,":[147],"using":[148,182],"Gaussian":[149],"synthetic":[150],"real":[153],"sets":[155],"(MNIST":[156],"CIFAR10).":[158],"Finally,":[159],"propose":[161],"criterion":[164],"construct":[166],"early":[168,195],"stopping":[169],"procedure":[170],"when":[171],"NN":[173],"is":[174],"found":[175],"highly":[176],"regularized":[177],"without":[178],"test":[179],"by":[181],"connection":[184],"between":[185],"types":[188],"degrees":[191],"regularization.":[193],"Such":[194],"stopped":[196],"DNNs":[197],"avoid":[198],"unnecessary":[199],"extra":[200],"training":[201],"while":[202],"preserving":[203],"much":[205],"comparable":[206],"generalization":[207],"ability.":[208]},"counts_by_year":[],"updated_date":"2026-11-05T12:25:35.769452","created_date":"2021-12-06T00:00:00"}