{"id":"https://openalex.org/W7214746260","doi":"https://doi.org/10.48550/arxiv.2609.32270","title":"Overfitting of Spectral Gradient Descent: How Matrix Geometry shapes Generalization and Implicit Bias","display_name":"Overfitting of Spectral Gradient Descent: How Matrix Geometry shapes Generalization and Implicit Bias","publication_year":2026,"publication_date":"2026-09-26","ids":{"openalex":"https://openalex.org/W7214746260","doi":"https://doi.org/10.48550/arxiv.2609.32270"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.32270","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32270","pdf_url":"https://arxiv.org/pdf/2609.32270","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":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/2609.32270","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5006029732","display_name":"Guillaume Braun","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210110652","display_name":"RIKEN","ror":"https://ror.org/01sjwvz98","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652"]},{"id":"https://openalex.org/I4210126580","display_name":"RIKEN Center for Advanced Intelligence Project","ror":"https://ror.org/03ckxwf91","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652","https://openalex.org/I4210126580"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Braun, Guillaume","raw_affiliation_strings":["RIKEN AIP"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"RIKEN AIP","institution_ids":["https://openalex.org/I4210110652","https://openalex.org/I4210126580"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153536454","display_name":"Ichiro Hashimoto","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I185261750","display_name":"University of Toronto","ror":"https://ror.org/03dbr7087","country_code":"CA","type":"education","lineage":["https://openalex.org/I185261750"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Hashimoto, Ichiro","raw_affiliation_strings":["University of Toronto"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Toronto","institution_ids":["https://openalex.org/I185261750"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5066308746","display_name":"Masaaki Imaizumi","orcid":"https://orcid.org/0000-0001-6186-613X","observed_orcids":["https://orcid.org/0000-0001-6186-613X"]},"institutions":[{"id":"https://openalex.org/I22299242","display_name":"Kyoto University","ror":"https://ror.org/02kpeqv85","country_code":"JP","type":"education","lineage":["https://openalex.org/I22299242"]},{"id":"https://openalex.org/I4210110652","display_name":"RIKEN","ror":"https://ror.org/01sjwvz98","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652"]},{"id":"https://openalex.org/I4210126580","display_name":"RIKEN Center for Advanced Intelligence Project","ror":"https://ror.org/03ckxwf91","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652","https://openalex.org/I4210126580"]},{"id":"https://openalex.org/I74801974","display_name":"The University of Tokyo","ror":"https://ror.org/057zh3y96","country_code":"JP","type":"education","lineage":["https://openalex.org/I74801974"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Imaizumi, Masaaki","raw_affiliation_strings":["University of Tokyo , RIKEN AIP , Kyoto University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Tokyo , RIKEN AIP , Kyoto University","institution_ids":["https://openalex.org/I22299242","https://openalex.org/I4210110652","https://openalex.org/I4210126580","https://openalex.org/I74801974"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":5,"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":0.989300012588501,"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":0.989300012588501,"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":0.0024999999441206455,"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":0.001500000013038516,"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/overfitting","display_name":"overfitting","score":0.8510000109672546},{"id":"https://openalex.org/keywords/generalization","display_name":"generalization","score":0.8849999904632568},{"id":"https://openalex.org/keywords/implicit-bias","display_name":"implicit bias","score":0.9340000152587891},{"id":"https://openalex.org/keywords/overparameterized-neural-networks","display_name":"overparameterized neural networks","score":0.6940000057220459},{"id":"https://openalex.org/keywords/matrix-classification","display_name":"matrix classification","score":0.734000027179718},{"id":"https://openalex.org/keywords/shortcut-learning","display_name":"shortcut learning","score":0.7229999899864197},{"id":"https://openalex.org/keywords/singular-directions","display_name":"singular directions","score":0.6610000133514404},{"id":"https://openalex.org/keywords/interpolation","display_name":"interpolation","score":0.46700000762939453},{"id":"https://openalex.org/keywords/memorization","display_name":"memorization","score":0.46399998664855957}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.7760000228881836},{"id":"https://openalex.org/C17137986","wikidata":"https://www.wikidata.org/wiki/Q215067","display_name":"Orthogonality","level":2,"score":0.7109000086784363},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.661300003528595},{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.6032000184059143},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5472999811172485},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.53329998254776},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.5270000100135803},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.5098000168800354},{"id":"https://openalex.org/C109282560","wikidata":"https://www.wikidata.org/wiki/Q4166054","display_name":"Singular value","level":3,"score":0.4481000006198883},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.3935999870300293},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.39259999990463257},{"id":"https://openalex.org/C21556879","wikidata":"https://www.wikidata.org/wiki/Q43219517","display_name":"Moore\u2013Penrose pseudoinverse","level":3,"score":0.39070001244544983},{"id":"https://openalex.org/C51364350","wikidata":"https://www.wikidata.org/wiki/Q3123529","display_name":"Spectral geometry","level":2,"score":0.37529999017715454},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.3594000041484833},{"id":"https://openalex.org/C109546454","wikidata":"https://www.wikidata.org/wiki/Q3798604","display_name":"Information geometry","level":4,"score":0.35899999737739563},{"id":"https://openalex.org/C64812099","wikidata":"https://www.wikidata.org/wiki/Q176604","display_name":"Random matrix","level":3,"score":0.35839998722076416},{"id":"https://openalex.org/C152822103","wikidata":"https://www.wikidata.org/wiki/Q7575207","display_name":"Spectral shape analysis","level":3,"score":0.34549999237060547},{"id":"https://openalex.org/C156778621","wikidata":"https://www.wikidata.org/wiki/Q1365748","display_name":"Spectrum (functional analysis)","level":2,"score":0.3416000008583069},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.31679999828338623},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.31520000100135803},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.3122999966144562},{"id":"https://openalex.org/C2780980858","wikidata":"https://www.wikidata.org/wiki/Q110022","display_name":"Dual (grammatical number)","level":2,"score":0.310699999332428},{"id":"https://openalex.org/C136272165","wikidata":"https://www.wikidata.org/wiki/Q4048889","display_name":"Singular spectrum analysis","level":3,"score":0.29580000042915344},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.2953000068664551},{"id":"https://openalex.org/C45347329","wikidata":"https://www.wikidata.org/wiki/Q5166604","display_name":"Convolution (computer science)","level":3,"score":0.2870999872684479},{"id":"https://openalex.org/C207467116","wikidata":"https://www.wikidata.org/wiki/Q4385666","display_name":"Inverse","level":2,"score":0.2806999981403351},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2759999930858612},{"id":"https://openalex.org/C102519508","wikidata":"https://www.wikidata.org/wiki/Q6520159","display_name":"Fourier transform","level":2,"score":0.2700999975204468},{"id":"https://openalex.org/C2778029271","wikidata":"https://www.wikidata.org/wiki/Q5421931","display_name":"Extension (predicate logic)","level":2,"score":0.26420000195503235},{"id":"https://openalex.org/C92207270","wikidata":"https://www.wikidata.org/wiki/Q939253","display_name":"Matrix norm","level":3,"score":0.2515999972820282}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.32270","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32270","pdf_url":"https://arxiv.org/pdf/2609.32270","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":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.32270","doi":"https://doi.org/10.48550/arxiv.2609.32270","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.32270","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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.32270","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.32270","pdf_url":"https://arxiv.org/pdf/2609.32270","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":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/W7214746260.pdf","grobid_xml":"https://content.openalex.org/works/W7214746260.grobid-xml"},"referenced_works_count":3,"referenced_works":["https://openalex.org/W1560153690","https://openalex.org/W1724170917","https://openalex.org/W3157298807"],"related_works":["https://openalex.org/W7101885942","https://openalex.org/W3175418406","https://openalex.org/W7166712237","https://openalex.org/W4413182473","https://openalex.org/W4386484876","https://openalex.org/W6929358514","https://openalex.org/W4406418082","https://openalex.org/W7197048738","https://openalex.org/W3113636031","https://openalex.org/W7168791831"],"abstract_inverted_index":{"We":[0,40],"study":[1],"the":[2,63,82,89,92,113,119,136],"generalization":[3,65],"of":[4,66,132,139],"spectral":[5,120],"gradient":[6],"descent":[7],"(SpecGD)":[8],"in":[9],"overparameterized":[10],"matrix":[11],"classification":[12],"with":[13,23,49,135,164],"corrupted":[14],"labels.":[15],"Each":[16],"input":[17],"combines":[18],"a":[19,24,31,46,102,108,129,147,162],"shared":[20],"low-rank":[21],"signal":[22,90,106],"rank-one":[25],"sample-specific":[26],"perturbation,":[27],"referred":[28],"to":[29,161],"as":[30],"shortcut,":[32],"that":[33,146],"enables":[34],"memorization":[35],"but":[36,104],"does":[37,124],"not":[38,125],"generalize.":[39],"contrast":[41],"collapsed":[42,70],"shortcuts,":[43,51],"which":[44,52,118],"share":[45],"singular":[47,55],"direction,":[48,95],"dispersed":[50,76,83],"occupy":[53],"distinct":[54],"directions.":[56],"Changing":[57],"only":[58],"this":[59],"geometry":[60],"can":[61,72,78,151],"reverse":[62],"relative":[64],"GD":[67],"and":[68,154],"SpecGD:":[69],"shortcuts":[71,77],"favor":[73,79],"SpecGD,":[74,117],"while":[75,96,157],"GD.":[80],"In":[81],"regime,":[84],"exact":[85],"shortcut":[86],"orthogonality":[87],"eliminates":[88],"from":[91],"late-stage":[93],"SpecGD":[94,149],"vanishing":[97],"random":[98],"correlations":[99],"collectively":[100],"generate":[101],"small":[103],"generalization-relevant":[105],"through":[107],"second-order":[109],"effect.":[110],"To":[111],"identify":[112],"direction":[114,163],"selected":[115],"by":[116],"max-margin":[121],"problem":[122],"alone":[123],"determine,":[126],"we":[127,144],"combine":[128],"refined":[130],"analysis":[131],"its":[133],"dual":[134],"exponentiated-gradient":[137],"dynamics":[138],"normalized":[140],"loss":[141],"weights.":[142],"Finally,":[143],"show":[145],"single":[148],"step":[150],"already":[152],"interpolate":[153],"generalize":[155],"well,":[156],"continued":[158],"training":[159],"converges":[160],"substantially":[165],"worse":[166],"generalization.":[167]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}