{"id":"https://openalex.org/W7224734271","doi":"https://doi.org/10.48550/arxiv.2610.39717","title":"Implicit Bias of (Stochastic) Gradient Descent on Deep Linear Diagonal Networks","display_name":"Implicit Bias of (Stochastic) Gradient Descent on Deep Linear Diagonal Networks","publication_year":2026,"publication_date":"2026-10-29","ids":{"openalex":"https://openalex.org/W7224734271","doi":"https://doi.org/10.48550/arxiv.2610.39717"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.39717","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39717","pdf_url":"https://arxiv.org/pdf/2610.39717","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.39717","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5024219432","display_name":"Wiebke Bartolomaeus","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4403386549","display_name":"Munich Center for Machine Learning","ror":"https://ror.org/02nfy3535","country_code":"DE","type":"education","lineage":["https://openalex.org/I4403386549","https://openalex.org/I62916508","https://openalex.org/I8204097"]},{"id":"https://openalex.org/I8204097","display_name":"Ludwig-Maximilians-Universit\u00e4t M\u00fcnchen","ror":"https://ror.org/05591te55","country_code":"DE","type":"education","lineage":["https://openalex.org/I8204097"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Bartolomaeus, Wiebke","raw_affiliation_strings":["Ludwig-Maximilians-University Munich , Department of Mathematics , Germany","Munich Center for Machine Learning (MCML)"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ludwig-Maximilians-University Munich , Department of Mathematics , Germany","institution_ids":["https://openalex.org/I8204097"]},{"raw_affiliation_string":"Munich Center for Machine Learning (MCML)","institution_ids":["https://openalex.org/I4403386549"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5005201576","display_name":"Holger Rauhut","orcid":"https://orcid.org/0000-0003-4750-5092","observed_orcids":["https://orcid.org/0000-0003-4750-5092"]},"institutions":[{"id":"https://openalex.org/I4403386549","display_name":"Munich Center for Machine Learning","ror":"https://ror.org/02nfy3535","country_code":"DE","type":"education","lineage":["https://openalex.org/I4403386549","https://openalex.org/I62916508","https://openalex.org/I8204097"]},{"id":"https://openalex.org/I8204097","display_name":"Ludwig-Maximilians-Universit\u00e4t M\u00fcnchen","ror":"https://ror.org/05591te55","country_code":"DE","type":"education","lineage":["https://openalex.org/I8204097"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Rauhut, Holger","raw_affiliation_strings":["Ludwig-Maximilians-University Munich , Department of Mathematics , Germany","Munich Center for Machine Learning (MCML)"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ludwig-Maximilians-University Munich , Department of Mathematics , Germany","institution_ids":["https://openalex.org/I8204097"]},{"raw_affiliation_string":"Munich Center for Machine Learning (MCML)","institution_ids":["https://openalex.org/I4403386549"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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":0.9810000061988831,"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":0.9810000061988831,"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/T10500","display_name":"Sparse and Compressive Sensing Techniques","score":0.013100000098347664,"subfield":{"id":"https://openalex.org/subfields/2206","display_name":"Computational Mechanics"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.0012000000569969416,"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-bias","display_name":"implicit bias","score":0.9990000128746033},{"id":"https://openalex.org/keywords/stochastic-gradient-descent","display_name":"stochastic gradient descent","score":0.925000011920929},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"gradient descent","score":0.7440000176429749},{"id":"https://openalex.org/keywords/overparameterized-neural-networks","display_name":"overparameterized neural networks","score":0.734000027179718},{"id":"https://openalex.org/keywords/sparse-solutions","display_name":"sparse solutions","score":0.7749999761581421},{"id":"https://openalex.org/keywords/weighted-l1-minimization","display_name":"weighted l1 minimization","score":0.9039999842643738},{"id":"https://openalex.org/keywords/convergence-analysis","display_name":"convergence analysis","score":0.8259999752044678},{"id":"https://openalex.org/keywords/gradient-flow","display_name":"gradient flow","score":0.5070000290870667}],"concepts":[{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.7641000151634216},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.6969000101089478},{"id":"https://openalex.org/C130367717","wikidata":"https://www.wikidata.org/wiki/Q189791","display_name":"Diagonal","level":2,"score":0.6855999827384949},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.6254000067710876},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.6240000128746033},{"id":"https://openalex.org/C167879884","wikidata":"https://www.wikidata.org/wiki/Q727568","display_name":"Balanced flow","level":2,"score":0.5634999871253967},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5340999960899353},{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.5120000243186951},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.507099986076355},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.505299985408783},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4490000009536743},{"id":"https://openalex.org/C116149140","wikidata":"https://www.wikidata.org/wiki/Q2070951","display_name":"Descent direction","level":4,"score":0.4415000081062317},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.41670000553131104},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.37770000100135803},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.351500004529953},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3400000035762787},{"id":"https://openalex.org/C10494615","wikidata":"https://www.wikidata.org/wiki/Q17086765","display_name":"Proximal Gradient Methods","level":4,"score":0.33180001378059387},{"id":"https://openalex.org/C26362088","wikidata":"https://www.wikidata.org/wiki/Q17086453","display_name":"Nonlinear conjugate gradient method","level":4,"score":0.3222000002861023},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3156000077724457},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3057999908924103},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.30070000886917114},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.28760001063346863},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.28220000863075256},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2797999978065491},{"id":"https://openalex.org/C57869625","wikidata":"https://www.wikidata.org/wiki/Q1783502","display_name":"Rate of convergence","level":3,"score":0.2768999934196472},{"id":"https://openalex.org/C2776637919","wikidata":"https://www.wikidata.org/wiki/Q624380","display_name":"Descent (aeronautics)","level":2,"score":0.2732999920845032},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C98156149","wikidata":"https://www.wikidata.org/wiki/Q2414143","display_name":"Pressure gradient","level":2,"score":0.26260000467300415},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.25870001316070557},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.2567000091075897}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.39717","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39717","pdf_url":"https://arxiv.org/pdf/2610.39717","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.2610.39717","doi":"https://doi.org/10.48550/arxiv.2610.39717","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.39717","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:2610.39717","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39717","pdf_url":"https://arxiv.org/pdf/2610.39717","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":[{"id":"https://openalex.org/G4355630769","display_name":"Impliziter Bias und Netzwerke niedriger Komplexit\u00e4t","funder_award_id":"464121491","funder_id":"https://openalex.org/F4320320879","funder_display_name":"Deutsche Forschungsgemeinschaft"},{"id":"https://openalex.org/G442688124","display_name":null,"funder_award_id":"SPP2298","funder_id":"https://openalex.org/F4320320879","funder_display_name":"Deutsche Forschungsgemeinschaft"},{"id":"https://openalex.org/G4771048887","display_name":null,"funder_award_id":"SFB 1481","funder_id":"https://openalex.org/F4320320879","funder_display_name":"Deutsche Forschungsgemeinschaft"}],"funders":[{"id":"https://openalex.org/F4320320879","display_name":"Deutsche Forschungsgemeinschaft","ror":"https://ror.org/018mejw64"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224734271.pdf","grobid_xml":"https://content.openalex.org/works/W7224734271.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7133238145","https://openalex.org/W3170832232","https://openalex.org/W7168364122","https://openalex.org/W3042242685","https://openalex.org/W4402396053","https://openalex.org/W4321392927","https://openalex.org/W4406542805","https://openalex.org/W3123819420","https://openalex.org/W4287182101","https://openalex.org/W7134809700"],"abstract_inverted_index":{"Modern":[0],"machine":[1],"learning":[2,64],"models":[3],"are":[4],"often":[5],"overparameterized,":[6],"i.e.,":[7],"have":[8,32],"more":[9,88],"parameters":[10],"than":[11],"training":[12,26,50],"data.":[13],"This":[14,107],"typically":[15],"implies":[16],"that":[17,25,59],"the":[18,36,47,60,86,115,129,134,137,141],"loss":[19],"function":[20],"admits":[21],"multiple":[22],"minimizers":[23],"so":[24],"methods":[27],"(and":[28],"their":[29],"hyperparameter":[30],"choices)":[31],"significant":[33],"influence":[34,135],"on":[35,128,140],"computed":[37],"solution.":[38],"We":[39],"study":[40],"this":[41,83],"so-called":[42],"implicit":[43,142],"bias":[44],"phenomenon":[45],"in":[46],"context":[48],"of":[49,79,136],"deep":[51],"linear":[52],"diagonal":[53],"networks.":[54],"Previous":[55],"work":[56],"has":[57],"shown":[58],"gradient":[61,91,98,105],"flow":[62],"for":[63],"such":[65],"networks":[66],"converges":[67],"to":[68,85,110,120],"(almost)":[69],"sparse":[70],"solutions":[71],"when":[72],"initialized":[73],"small.":[74],"For":[75],"an":[76],"arbitrary":[77],"number":[78],"layers,":[80],"we":[81,124],"extend":[82],"analysis":[84],"practically":[87],"realistic":[89],"(stochastic)":[90,97,103],"descent":[92,99],"iterations.":[93],"Our":[94],"approach":[95],"identifies":[96],"as":[100],"a":[101],"perturbed":[102],"mirror":[104],"descent.":[106],"allows":[108],"us":[109],"establish":[111],"convergence":[112],"and":[113,132],"characterize":[114],"limiting":[116],"solution,":[117],"relating":[118],"it":[119],"weighted":[121],"$\\ell_1$-minimization.":[122],"Moreover,":[123],"provide":[125],"sufficient":[126],"conditions":[127],"step":[130,138],"sizes":[131,139],"quantify":[133],"bias.":[143]},"counts_by_year":[],"updated_date":"2026-11-03T10:50:19.383937","created_date":"2026-10-31T00:00:00"}