{"id":"https://openalex.org/W4289446290","doi":"https://doi.org/10.48550/arxiv.2207.14484","title":"Adaptive Gradient Methods at the Edge of Stability","display_name":"Adaptive Gradient Methods at the Edge of Stability","publication_year":2022,"publication_date":"2022-07-29","ids":{"openalex":"https://openalex.org/W4289446290","doi":"https://doi.org/10.48550/arxiv.2207.14484"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2207.14484","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2207.14484","pdf_url":"https://arxiv.org/pdf/2207.14484","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","license_id":"https://openalex.org/licenses/cc-by","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/2207.14484","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Cohen, Jeremy M.","orcid":null,"observed_orcids":[]},"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":"Cohen, Jeremy M.","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051865121","display_name":"Behrooz Ghorbani","orcid":null,"observed_orcids":[]},"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":"Ghorbani, Behrooz","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Krishnan, Shankar","orcid":null,"observed_orcids":[]},"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":"Krishnan, Shankar","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007875487","display_name":"Naman Agarwal","orcid":"https://orcid.org/0000-0003-0320-0238","observed_orcids":["https://orcid.org/0000-0003-0320-0238"]},"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":"Agarwal, Naman","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5038526086","display_name":"Sourabh Medapati","orcid":null,"observed_orcids":[]},"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":"Medapati, Sourabh","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082238359","display_name":"Michal Badura","orcid":null,"observed_orcids":[]},"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":"Badura, Michal","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037613447","display_name":"Daniel Suo","orcid":null,"observed_orcids":[]},"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":"Suo, Daniel","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073584072","display_name":"David E. Cardoze","orcid":null,"observed_orcids":[]},"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":"Cardoze, David","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016611202","display_name":"Zachary Nado","orcid":null,"observed_orcids":[]},"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":"Nado, Zachary","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047062711","display_name":"George E. Dahl","orcid":"https://orcid.org/0000-0002-0083-844X","observed_orcids":["https://orcid.org/0000-0002-0083-844X"]},"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":"Dahl, George E.","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5102853126","display_name":"Justin Gilmer","orcid":"https://orcid.org/0009-0003-4813-7874","observed_orcids":["https://orcid.org/0009-0003-4813-7874"]},"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":"Gilmer, Justin","raw_affiliation_strings":["Google Research , Brain Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I1291425158"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":102,"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":0.9994999766349792,"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":0.9994999766349792,"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":9.999999747378752e-05,"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/T12535","display_name":"Machine Learning and Data Classification","score":9.999999747378752e-05,"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/adaptive-gradient-methods","display_name":"adaptive gradient methods","score":0.9679999947547913},{"id":"https://openalex.org/keywords/adam-optimizer","display_name":"Adam optimizer","score":0.9710000157356262},{"id":"https://openalex.org/keywords/edge-of-stability","display_name":"edge of stability","score":0.9419999718666077},{"id":"https://openalex.org/keywords/training-dynamics","display_name":"training dynamics","score":0.8330000042915344},{"id":"https://openalex.org/keywords/largest-eigenvalue","display_name":"largest eigenvalue","score":0.8560000061988831},{"id":"https://openalex.org/keywords/batch-size-scaling","display_name":"batch size scaling","score":0.8069999814033508},{"id":"https://openalex.org/keywords/loss-landscape-curvature","display_name":"loss landscape curvature","score":0.718999981880188},{"id":"https://openalex.org/keywords/stability-threshold","display_name":"stability threshold","score":0.43700000643730164}],"concepts":[{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.8363397121429443},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.7369097471237183},{"id":"https://openalex.org/C167431342","wikidata":"https://www.wikidata.org/wiki/Q1754327","display_name":"Preconditioner","level":3,"score":0.6689313650131226},{"id":"https://openalex.org/C158693339","wikidata":"https://www.wikidata.org/wiki/Q190524","display_name":"Eigenvalues and eigenvectors","level":2,"score":0.5822917819023132},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5545399785041809},{"id":"https://openalex.org/C195065555","wikidata":"https://www.wikidata.org/wiki/Q214881","display_name":"Curvature","level":2,"score":0.5455493330955505},{"id":"https://openalex.org/C125014702","wikidata":"https://www.wikidata.org/wiki/Q4680749","display_name":"Adaptive learning","level":2,"score":0.5175844430923462},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.4970991909503937},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.46696048974990845},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3708813786506653},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36531561613082886},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.3612516522407532},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.34843313694000244},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3475390672683716},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3226364254951477},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.18140831589698792},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.16539552807807922},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.13436108827590942},{"id":"https://openalex.org/C159694833","wikidata":"https://www.wikidata.org/wiki/Q2321565","display_name":"Iterative method","level":2,"score":0.12329694628715515},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.09771937131881714},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2207.14484","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2207.14484","pdf_url":"https://arxiv.org/pdf/2207.14484","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","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2207.14484","doi":"https://doi.org/10.48550/arxiv.2207.14484","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2207.14484","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","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2207.14484","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2207.14484","pdf_url":"https://arxiv.org/pdf/2207.14484","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","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4289446290.pdf","grobid_xml":"https://content.openalex.org/works/W4289446290.grobid-xml"},"referenced_works_count":27,"referenced_works":["https://openalex.org/W2127265926","https://openalex.org/W2518108298","https://openalex.org/W2776855315","https://openalex.org/W2796108585","https://openalex.org/W2886463271","https://openalex.org/W2886837499","https://openalex.org/W2908510526","https://openalex.org/W2945236893","https://openalex.org/W2962819303","https://openalex.org/W2964121744","https://openalex.org/W2964137095","https://openalex.org/W2975507180","https://openalex.org/W2995625976","https://openalex.org/W2996642285","https://openalex.org/W3008829306","https://openalex.org/W3009686669","https://openalex.org/W3034491857","https://openalex.org/W3034995656","https://openalex.org/W3086677109","https://openalex.org/W3096312061","https://openalex.org/W3103334733","https://openalex.org/W3168489096","https://openalex.org/W3207956290","https://openalex.org/W4226393973","https://openalex.org/W4283382874","https://openalex.org/W4287750641","https://openalex.org/W4293072421"],"related_works":["https://openalex.org/W7204101404","https://openalex.org/W3132026801","https://openalex.org/W7220386575","https://openalex.org/W4300980664","https://openalex.org/W4405956499","https://openalex.org/W7172102869","https://openalex.org/W2912323147","https://openalex.org/W4308670011","https://openalex.org/W7221809689","https://openalex.org/W4396600484"],"abstract_inverted_index":{"Very":[0],"little":[1],"is":[2,81],"known":[3],"about":[4],"the":[5,24,30,45,49,60,91,102,124,130,139,146,156,168],"training":[6],"dynamics":[7],"of":[8,26,48,63,105,120,138,172],"adaptive":[9,98,142,173],"gradient":[10,65,143,174],"methods":[11,99,122,144,175],"like":[12],"Adam":[13,69],"in":[14,29,110,114,176],"deep":[15,177],"learning.":[16,178],"In":[17],"this":[18,78,111],"paper,":[19],"we":[20,38],"shed":[21],"light":[22],"on":[23],"behavior":[25,109],"these":[27],"algorithms":[28,128],"full-batch":[31,43],"and":[32,74],"sufficiently":[33],"large":[34],"batch":[35,92],"settings.":[36],"Specifically,":[37],"empirically":[39],"demonstrate":[40],"that":[41,119],"during":[42,86],"training,":[44,88],"maximum":[46],"eigenvalue":[47],"preconditioned":[50],"Hessian":[51],"typically":[52],"equilibrates":[53],"at":[54,101,123,129,145],"a":[55,64,115,165],"certain":[56],"numerical":[57],"value":[58],"--":[59],"stability":[61,79],"threshold":[62,80],"descent":[66],"algorithm.":[67],"For":[68],"with":[70],"step":[71],"size":[72,93],"$\u03b7$":[73],"$\u03b2_1":[75],"=":[76],"0.9$,":[77],"$38/\u03b7$.":[82],"Similar":[83],"effects":[84],"occur":[85],"minibatch":[87],"especially":[89],"as":[90,164],"grows.":[94],"Yet,":[95],"even":[96],"though":[97],"train":[100],"``Adaptive":[103],"Edge":[104],"Stability''":[106],"(AEoS),":[107],"their":[108],"regime":[112],"differs":[113],"significant":[116],"way":[117],"from":[118,134],"non-adaptive":[121,127],"EoS.":[125],"Whereas":[126],"EoS":[131],"are":[132],"blocked":[133],"entering":[135],"high-curvature":[136,152],"regions":[137],"loss":[140],"landscape,":[141],"AEoS":[147],"can":[148,162],"keep":[149],"advancing":[150],"into":[151],"regions,":[153],"while":[154],"adapting":[155],"preconditioner":[157],"to":[158],"compensate.":[159],"Our":[160],"findings":[161],"serve":[163],"foundation":[166],"for":[167],"community's":[169],"future":[170],"understanding":[171]},"counts_by_year":[{"year":2026,"cited_by_count":42},{"year":2025,"cited_by_count":17},{"year":2024,"cited_by_count":16},{"year":2023,"cited_by_count":23},{"year":2022,"cited_by_count":4}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2022-08-02T00:00:00"}