{"id":"https://openalex.org/W7140216258","doi":"https://doi.org/10.48550/arxiv.2603.20892","title":"Implicit Regularization of Large Neural Networks via Mean-Field Formulation","display_name":"Implicit Regularization of Large Neural Networks via Mean-Field Formulation","publication_year":2026,"publication_date":"2026-03-21","ids":{"openalex":"https://openalex.org/W7140216258","doi":"https://doi.org/10.48550/arxiv.2603.20892"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2603.20892","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2603.20892","pdf_url":"https://arxiv.org/pdf/2603.20892","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":"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":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2603.20892","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5049941278","display_name":"Beatrice Acciaio","orcid":"https://orcid.org/0000-0002-8138-2578","observed_orcids":["https://orcid.org/0000-0002-8138-2578"]},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Acciaio, Beatrice","raw_affiliation_strings":["Department of Mathematics , ETH Z\u00fcrich , Z\u00fcrich , Switzerland,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Mathematics , ETH Z\u00fcrich , Z\u00fcrich , Switzerland,","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010858034","display_name":"Jakob Heiss","orcid":"https://orcid.org/0000-0003-1447-6782","observed_orcids":["https://orcid.org/0000-0003-1447-6782"]},"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":"Heiss, Jakob","raw_affiliation_strings":["Statistics Department , UC Berkeley , Berkeley , USA,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Statistics Department , UC Berkeley , Berkeley , USA,","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067473464","display_name":"Gudmund Pammer","orcid":"https://orcid.org/0000-0003-2494-8739","observed_orcids":["https://orcid.org/0000-0003-2494-8739"]},"institutions":[{"id":"https://openalex.org/I4092182","display_name":"Graz University of Technology","ror":"https://ror.org/00d7xrm67","country_code":"AT","type":"education","lineage":["https://openalex.org/I4092182"]}],"countries":["AT"],"is_corresponding":false,"raw_author_name":"Pammer, Gudmund","raw_affiliation_strings":["Institute for Statistics , TU Graz , Styria , Austria,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute for Statistics , TU Graz , Styria , Austria,","institution_ids":["https://openalex.org/I4092182"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5010314615","display_name":"Qinxin Yan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I20089843","display_name":"Princeton University","ror":"https://ror.org/00hx57361","country_code":"US","type":"education","lineage":["https://openalex.org/I20089843"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yan, Qinxin","raw_affiliation_strings":["Program in Applied and Computational Mathematics , Princeton University , Princeton , USA,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Program in Applied and Computational Mathematics , Princeton University , Princeton , USA,","institution_ids":["https://openalex.org/I20089843"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":4,"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":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.7813000082969666,"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":0.7813000082969666,"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/T10229","display_name":"Geometric Analysis and Curvature Flows","score":0.06650000065565109,"subfield":{"id":"https://openalex.org/subfields/2604","display_name":"Applied Mathematics"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.03689999878406525,"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/implicit-regularization","display_name":"implicit regularization","score":1.0},{"id":"https://openalex.org/keywords/early-stopping","display_name":"early stopping","score":0.9319999814033508},{"id":"https://openalex.org/keywords/overparametrized-neural-networks","display_name":"overparametrized neural networks","score":0.9750000238418579},{"id":"https://openalex.org/keywords/mean-field-limit","display_name":"mean-field limit","score":0.9399999976158142},{"id":"https://openalex.org/keywords/hamilton-jacobi-bellman-equation","display_name":"Hamilton-Jacobi-Bellman equation","score":0.9789999723434448},{"id":"https://openalex.org/keywords/dynamic-programming-principle","display_name":"dynamic programming principle","score":0.9789999723434448},{"id":"https://openalex.org/keywords/2-wasserstein-distance","display_name":"2-Wasserstein distance","score":0.9240000247955322},{"id":"https://openalex.org/keywords/nonasymptotic-bounds","display_name":"non-asymptotic bounds","score":0.9700000286102295}],"concepts":[{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.742900013923645},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5501000285148621},{"id":"https://openalex.org/C5465570","wikidata":"https://www.wikidata.org/wiki/Q5326898","display_name":"Early stopping","level":3,"score":0.5200999975204468},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.49549999833106995},{"id":"https://openalex.org/C141718189","wikidata":"https://www.wikidata.org/wiki/Q7309628","display_name":"Regularization perspectives on support vector machines","level":4,"score":0.474700003862381},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.4657000005245209},{"id":"https://openalex.org/C167879884","wikidata":"https://www.wikidata.org/wiki/Q727568","display_name":"Balanced flow","level":2,"score":0.44339999556541443},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.43549999594688416},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.4348999857902527},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4320000112056732},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.42669999599456787},{"id":"https://openalex.org/C91575142","wikidata":"https://www.wikidata.org/wiki/Q1971426","display_name":"Optimal control","level":2,"score":0.3968999981880188},{"id":"https://openalex.org/C21031990","wikidata":"https://www.wikidata.org/wiki/Q355020","display_name":"Probability measure","level":2,"score":0.3887999951839447},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.385699987411499},{"id":"https://openalex.org/C79248915","wikidata":"https://www.wikidata.org/wiki/Q17086776","display_name":"Proximal gradient methods for learning","level":5,"score":0.3224000036716461},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3043999969959259},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.3027999997138977},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3027999997138977},{"id":"https://openalex.org/C27872270","wikidata":"https://www.wikidata.org/wiki/Q4839810","display_name":"Backus\u2013Gilbert method","level":5,"score":0.3003000020980835},{"id":"https://openalex.org/C198043062","wikidata":"https://www.wikidata.org/wiki/Q180953","display_name":"Metric space","level":2,"score":0.26969999074935913},{"id":"https://openalex.org/C2777634741","wikidata":"https://www.wikidata.org/wiki/Q768993","display_name":"Wasserstein metric","level":2,"score":0.2687000036239624},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.2623000144958496},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.25780001282691956},{"id":"https://openalex.org/C37404715","wikidata":"https://www.wikidata.org/wiki/Q380679","display_name":"Dynamic programming","level":2,"score":0.25760000944137573},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2603.20892","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2603.20892","pdf_url":"https://arxiv.org/pdf/2603.20892","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":"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":"text"},{"id":"doi:10.48550/arxiv.2603.20892","doi":"https://doi.org/10.48550/arxiv.2603.20892","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.20892","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2603.20892","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2603.20892","pdf_url":"https://arxiv.org/pdf/2603.20892","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":"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":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140216258.pdf","grobid_xml":"https://content.openalex.org/works/W7140216258.grobid-xml"},"referenced_works_count":26,"referenced_works":["https://openalex.org/W1988115241","https://openalex.org/W2007048878","https://openalex.org/W2078272725","https://openalex.org/W2103496339","https://openalex.org/W2766965791","https://openalex.org/W2809090039","https://openalex.org/W2810862998","https://openalex.org/W2889237180","https://openalex.org/W2914984858","https://openalex.org/W2943463079","https://openalex.org/W2963095610","https://openalex.org/W2963164374","https://openalex.org/W2963518130","https://openalex.org/W2978704341","https://openalex.org/W2986469991","https://openalex.org/W2998122242","https://openalex.org/W4255839052","https://openalex.org/W4286717651","https://openalex.org/W4307073093","https://openalex.org/W4392471145","https://openalex.org/W4407683974","https://openalex.org/W4412112932","https://openalex.org/W4415254072","https://openalex.org/W7117998796","https://openalex.org/W7133206325","https://openalex.org/W7154583106"],"related_works":["https://openalex.org/W1593822055","https://openalex.org/W2919767206","https://openalex.org/W2940107683","https://openalex.org/W2943463079","https://openalex.org/W4413393862","https://openalex.org/W7024724156","https://openalex.org/W4405029675","https://openalex.org/W6908998498","https://openalex.org/W2964915616","https://openalex.org/W3046476838"],"abstract_inverted_index":{"We":[0,38,105],"propose":[1],"a":[2,29,60,70,82,98],"mathematical":[3],"framework":[4],"to":[5,28,68],"explain":[6],"implicit":[7],"regularization":[8,99,114],"from":[9],"early":[10,103],"stopping":[11,118],"during":[12],"the":[13,20,23,33,51,86,90,112,117],"training":[14],"of":[15,35,85,89],"overparametrized":[16],"neural":[17],"networks.":[18],"In":[19],"mean-field":[21,83],"limit,":[22],"parameter":[24],"distribution":[25],"evolves":[26],"according":[27],"gradient":[30],"flow":[31],"on":[32,73,116],"space":[34],"probability":[36,74],"measures.":[37,75],"show":[39],"that":[40],"these":[41],"dynamics":[42],"admit":[43],"an":[44],"equivalent":[45],"McKean-Vlasov":[46],"stochastic":[47],"control":[48,57,87],"formulation":[49],"through":[50],"corresponding":[52],"Hamilton-Jacobi-Bellman":[53],"(HJB)":[54],"equation.":[55],"The":[56],"viewpoint":[58],"yields":[59],"Dynamic":[61],"Programming":[62],"Principle":[63],"(DPP),":[64],"which":[65],"we":[66],"use":[67],"define":[69],"new":[71],"metric":[72,77],"This":[76],"can":[78],"be":[79],"viewed":[80],"as":[81,97],"generalization":[84],"representation":[88],"Wasserstein-2":[91],"distance,":[92],"and":[93],"it":[94],"naturally":[95],"appears":[96],"term":[100],"selected":[101],"by":[102],"stopping.":[104],"further":[106],"obtain":[107],"non-asymptotic":[108],"bounds":[109],"describing":[110],"how":[111],"induced":[113],"depends":[115],"time.":[119]},"counts_by_year":[],"updated_date":"2026-11-13T13:42:25.865203","created_date":"2026-03-25T00:00:00"}