{"id":"https://openalex.org/W3161014260","doi":"https://doi.org/10.1109/icassp39728.2021.9413771","title":"A Scale Invariant Measure of Flatness for Deep Network Minima","display_name":"A Scale Invariant Measure of Flatness for Deep Network Minima","publication_year":2021,"publication_date":"2021-05-13","ids":{"openalex":"https://openalex.org/W3161014260","doi":"https://doi.org/10.1109/icassp39728.2021.9413771","mag":"3161014260"},"language":"en","primary_location":{"id":"doi:10.1109/icassp39728.2021.9413771","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp39728.2021.9413771","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5079543741","display_name":"Akshay Rangamani","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Akshay Rangamani","raw_affiliation_strings":["Center for Brains, Minds, and Machines, MIT"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Center for Brains, Minds, and Machines, MIT","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103324660","display_name":"Nam H. Nguyen","orcid":"https://orcid.org/0000-0002-1254-0069"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nam H. Nguyen","raw_affiliation_strings":["IBM Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104175390","display_name":"Abhishek Kumar","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":"Abhishek Kumar","raw_affiliation_strings":["Google Brain"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Brain","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036688203","display_name":"Dzung T. Phan","orcid":"https://orcid.org/0000-0003-1579-7035"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dzung Phan","raw_affiliation_strings":["IBM Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103044811","display_name":"Sang Chin","orcid":"https://orcid.org/0000-0002-1913-4223"},"institutions":[{"id":"https://openalex.org/I111088046","display_name":"Boston University","ror":"https://ror.org/05qwgg493","country_code":"US","type":"education","lineage":["https://openalex.org/I111088046"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sang Peter Chin","raw_affiliation_strings":["Boston University,CS Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Boston University,CS Department","institution_ids":["https://openalex.org/I111088046"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101413113","display_name":"Trac D. Tran","orcid":"https://orcid.org/0000-0002-0421-8416"},"institutions":[{"id":"https://openalex.org/I145311948","display_name":"Johns Hopkins University","ror":"https://ror.org/00za53h95","country_code":"US","type":"education","lineage":["https://openalex.org/I145311948"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Trac D. Tran","raw_affiliation_strings":["Johns Hopkins University,ECE Department"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Johns Hopkins University,ECE Department","institution_ids":["https://openalex.org/I145311948"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.05840482,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"70","issue":null,"first_page":"1680","last_page":"1684"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.9997000098228455,"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":0.9997000098228455,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.9993000030517578,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/maxima-and-minima","display_name":"Maxima and minima","score":0.7392540574073792},{"id":"https://openalex.org/keywords/flatness","display_name":"Flatness (cosmology)","score":0.6798148155212402},{"id":"https://openalex.org/keywords/hessian-matrix","display_name":"Hessian matrix","score":0.5632243156433105},{"id":"https://openalex.org/keywords/invariant-measure","display_name":"Invariant measure","score":0.5458846092224121},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.5357557535171509},{"id":"https://openalex.org/keywords/equivalence","display_name":"Equivalence (formal languages)","score":0.5289868116378784},{"id":"https://openalex.org/keywords/measure","display_name":"Measure (data warehouse)","score":0.48364531993865967},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"Stochastic gradient descent","score":0.4701179265975952},{"id":"https://openalex.org/keywords/quotient","display_name":"Quotient","score":0.4605301022529602},{"id":"https://openalex.org/keywords/parametric-statistics","display_name":"Parametric statistics","score":0.4405122697353363},{"id":"https://openalex.org/keywords/invariant","display_name":"Invariant (physics)","score":0.42981287837028503},{"id":"https://openalex.org/keywords/applied-mathematics","display_name":"Applied mathematics","score":0.4010801315307617},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.3890787959098816},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3562443256378174},{"id":"https://openalex.org/keywords/pure-mathematics","display_name":"Pure mathematics","score":0.30531615018844604},{"id":"https://openalex.org/keywords/mathematical-analysis","display_name":"Mathematical analysis","score":0.28845399618148804},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.2483023703098297},{"id":"https://openalex.org/keywords/statistics","display_name":"Statistics","score":0.1116950511932373},{"id":"https://openalex.org/keywords/data-mining","display_name":"Data mining","score":0.10458740592002869},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.08528289198875427}],"concepts":[{"id":"https://openalex.org/C186633575","wikidata":"https://www.wikidata.org/wiki/Q845060","display_name":"Maxima and minima","level":2,"score":0.7392540574073792},{"id":"https://openalex.org/C2778530986","wikidata":"https://www.wikidata.org/wiki/Q5457948","display_name":"Flatness (cosmology)","level":3,"score":0.6798148155212402},{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.5632243156433105},{"id":"https://openalex.org/C115667082","wikidata":"https://www.wikidata.org/wiki/Q4200518","display_name":"Invariant measure","level":3,"score":0.5458846092224121},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5357557535171509},{"id":"https://openalex.org/C2780069185","wikidata":"https://www.wikidata.org/wiki/Q7977945","display_name":"Equivalence (formal languages)","level":2,"score":0.5289868116378784},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.48364531993865967},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.4701179265975952},{"id":"https://openalex.org/C199422724","wikidata":"https://www.wikidata.org/wiki/Q41118","display_name":"Quotient","level":2,"score":0.4605301022529602},{"id":"https://openalex.org/C117251300","wikidata":"https://www.wikidata.org/wiki/Q1849855","display_name":"Parametric statistics","level":2,"score":0.4405122697353363},{"id":"https://openalex.org/C190470478","wikidata":"https://www.wikidata.org/wiki/Q2370229","display_name":"Invariant (physics)","level":2,"score":0.42981287837028503},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4010801315307617},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3890787959098816},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3562443256378174},{"id":"https://openalex.org/C202444582","wikidata":"https://www.wikidata.org/wiki/Q837863","display_name":"Pure mathematics","level":1,"score":0.30531615018844604},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.28845399618148804},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2483023703098297},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.1116950511932373},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.10458740592002869},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.08528289198875427},{"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/C37914503","wikidata":"https://www.wikidata.org/wiki/Q156495","display_name":"Mathematical physics","level":1,"score":0.0},{"id":"https://openalex.org/C122044880","wikidata":"https://www.wikidata.org/wiki/Q5498822","display_name":"Ergodic theory","level":2,"score":0.0},{"id":"https://openalex.org/C26405456","wikidata":"https://www.wikidata.org/wiki/Q338","display_name":"Cosmology","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp39728.2021.9413771","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp39728.2021.9413771","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":45,"referenced_works":["https://openalex.org/W577198184","https://openalex.org/W1686810756","https://openalex.org/W1804110266","https://openalex.org/W2006903949","https://openalex.org/W2112796928","https://openalex.org/W2130942839","https://openalex.org/W2143163787","https://openalex.org/W2160815625","https://openalex.org/W2163605009","https://openalex.org/W2284050935","https://openalex.org/W2605372163","https://openalex.org/W2750384547","https://openalex.org/W2750854376","https://openalex.org/W2762473905","https://openalex.org/W2786622092","https://openalex.org/W2793416812","https://openalex.org/W2891872827","https://openalex.org/W2902986194","https://openalex.org/W2911853241","https://openalex.org/W2912811302","https://openalex.org/W2950928354","https://openalex.org/W2962835968","https://openalex.org/W2962836826","https://openalex.org/W2963685250","https://openalex.org/W2963755427","https://openalex.org/W2963959597","https://openalex.org/W3118608800","https://openalex.org/W4205293427","https://openalex.org/W4299971819","https://openalex.org/W6616617004","https://openalex.org/W6637373629","https://openalex.org/W6679436768","https://openalex.org/W6680910308","https://openalex.org/W6684191040","https://openalex.org/W6695676441","https://openalex.org/W6726983090","https://openalex.org/W6729595453","https://openalex.org/W6736583452","https://openalex.org/W6743688258","https://openalex.org/W6743799563","https://openalex.org/W6744810887","https://openalex.org/W6748230616","https://openalex.org/W6749538638","https://openalex.org/W6754238258","https://openalex.org/W6758514558"],"related_works":["https://openalex.org/W2611031068","https://openalex.org/W1704347466","https://openalex.org/W4283017538","https://openalex.org/W1996936972","https://openalex.org/W1545275724","https://openalex.org/W2802707792","https://openalex.org/W2067869010","https://openalex.org/W3143650729","https://openalex.org/W3177326532","https://openalex.org/W2042173174"],"abstract_inverted_index":{"It":[0],"has":[1],"been":[2],"empirically":[3,119],"observed":[4],"that":[5,44,71,85,110,129],"the":[6,39,45,62,95],"flatness":[7,32,48,109,131],"of":[8,31,38,47],"minima":[9,150],"obtained":[10],"from":[11],"training":[12],"deep":[13,23,73,148],"networks":[14,24,74],"seems":[15],"to":[16,36,113,118,127,146],"correlate":[17],"with":[18,25,75,134,142,151],"better":[19],"generalization.":[20],"However,":[21],"for":[22,72,108],"positively":[26,76],"homogeneous":[27],"activations,":[28,78],"most":[29],"measures":[30,64],"are":[33],"not":[34],"invariant":[35,112],"rescaling":[37,114],"network":[40,149],"parameters.":[41],"This":[42],"means":[43],"measure":[46,107,132],"can":[49],"be":[50],"made":[51],"as":[52,55,57],"small":[53],"or":[54],"large":[56],"possible":[58],"through":[59],"rescaling,":[60],"rendering":[61],"quantitative":[63],"meaningless.":[65],"In":[66],"this":[67],"paper":[68],"we":[69,103,124],"show":[70],"homogenous":[77],"these":[79,86],"rescalings":[80],"constitute":[81],"equivalence":[82,87],"relations,":[83],"and":[84,115],"relations":[88],"induce":[89],"a":[90,105],"quotient":[91],"manifold":[92],"structure":[93],"in":[94],"parameter":[96],"space.":[97],"Using":[98],"an":[99],"appropriate":[100],"Riemannian":[101],"metric,":[102],"propose":[104],"Hessian-based":[106],"is":[111],"perform":[116,125],"simulations":[117],"verify":[120,128],"our":[121,130],"claim.":[122],"Finally":[123],"experiments":[126],"correlates":[133],"generalization":[135,153],"by":[136],"using":[137],"minibatch":[138],"stochastic":[139],"gradient":[140],"descent":[141],"different":[143,152],"batch":[144],"sizes":[145],"find":[147],"properties.":[154]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}