{"id":"https://openalex.org/W7221518656","doi":"https://doi.org/10.48550/arxiv.2610.13123","title":"Why Heavy-Tailed Weight Spectra Predict Model Quality","display_name":"Why Heavy-Tailed Weight Spectra Predict Model Quality","publication_year":2026,"publication_date":"2026-10-09","ids":{"openalex":"https://openalex.org/W7221518656","doi":"https://doi.org/10.48550/arxiv.2610.13123"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.13123","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.13123","pdf_url":"https://arxiv.org/pdf/2610.13123","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/2610.13123","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5061246468","display_name":"Joseph Wilson","orcid":"https://orcid.org/0000-0003-2111-507X","observed_orcids":["https://orcid.org/0000-0003-2111-507X"]},"institutions":[{"id":"https://openalex.org/I165143802","display_name":"The University of Queensland","ror":"https://ror.org/00rqy9422","country_code":"AU","type":"education","lineage":["https://openalex.org/I165143802"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Wilson, Joseph","raw_affiliation_strings":["School of Mathematics and Physics University of Queensland"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Mathematics and Physics University of Queensland","institution_ids":["https://openalex.org/I165143802"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056317705","display_name":"Chris van der Heide","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165779595","display_name":"The University of Melbourne","ror":"https://ror.org/01ej9dk98","country_code":"AU","type":"education","lineage":["https://openalex.org/I165779595"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"van der Heide, Chris","raw_affiliation_strings":["Dept. of Electrical and Electronic Engineering University of Melbourne"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Electrical and Electronic Engineering University of Melbourne","institution_ids":["https://openalex.org/I165779595"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030249020","display_name":"Liam Hodgkinson","orcid":"https://orcid.org/0000-0002-4595-0347","observed_orcids":["https://orcid.org/0000-0002-4595-0347"]},"institutions":[{"id":"https://openalex.org/I165779595","display_name":"The University of Melbourne","ror":"https://ror.org/01ej9dk98","country_code":"AU","type":"education","lineage":["https://openalex.org/I165779595"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Hodgkinson, Liam","raw_affiliation_strings":["School of Mathematics and Statistics University of Melbourne"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Mathematics and Statistics University of Melbourne","institution_ids":["https://openalex.org/I165779595"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5158502036","display_name":"Zhichao Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1297971548","display_name":"International Computer Science Institute","ror":"https://ror.org/01ewh7m12","country_code":"US","type":"nonprofit","lineage":["https://openalex.org/I1297971548"]},{"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":"Wang, Zhichao","raw_affiliation_strings":["ICSI and Department of Statistics University of California , Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ICSI and Department of Statistics University of California , Berkeley","institution_ids":["https://openalex.org/I1297971548","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056884940","display_name":"Fred Roosta","orcid":"https://orcid.org/0000-0002-6920-7072","observed_orcids":["https://orcid.org/0000-0002-6920-7072"]},"institutions":[{"id":"https://openalex.org/I165143802","display_name":"The University of Queensland","ror":"https://ror.org/00rqy9422","country_code":"AU","type":"education","lineage":["https://openalex.org/I165143802"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Roosta, Fred","raw_affiliation_strings":["CIRES and School of Mathematics and Physics University of Queensland"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CIRES and School of Mathematics and Physics University of Queensland","institution_ids":["https://openalex.org/I165143802"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5033006662","display_name":"Michael W. Mahoney","orcid":"https://orcid.org/0000-0001-7920-4652","observed_orcids":["https://orcid.org/0000-0001-7920-4652"]},"institutions":[{"id":"https://openalex.org/I148283060","display_name":"Lawrence Berkeley National Laboratory","ror":"https://ror.org/02jbv0t02","country_code":"US","type":"facility","lineage":["https://openalex.org/I1330989302","https://openalex.org/I148283060","https://openalex.org/I39565521"]},{"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":"Mahoney, Michael W.","raw_affiliation_strings":["LBNL and Department of Statistics University of California , Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LBNL and Department of Statistics University of California , Berkeley","institution_ids":["https://openalex.org/I148283060","https://openalex.org/I95457486"]}]}],"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":0.33640000224113464,"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":0.33640000224113464,"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":0.27000001072883606,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.06989999860525131,"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/marginal-likelihood","display_name":"marginal likelihood","score":0.6980000138282776},{"id":"https://openalex.org/keywords/deep-neural-network","display_name":"deep neural network","score":0.6399999856948853},{"id":"https://openalex.org/keywords/overfitting","display_name":"overfitting","score":0.7239999771118164},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9309999942779541}],"concepts":[{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.5828999876976013},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5511999726295471},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.4952999949455261},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.482699990272522},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.47040000557899475},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.453900009393692},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4480000138282776},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.428600013256073},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.42809998989105225},{"id":"https://openalex.org/C95923904","wikidata":"https://www.wikidata.org/wiki/Q6760420","display_name":"Marginal likelihood","level":3,"score":0.40560001134872437},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4027999937534332},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3991999924182892},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.39899998903274536},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3668000102043152},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3578000068664551},{"id":"https://openalex.org/C45804977","wikidata":"https://www.wikidata.org/wiki/Q7239673","display_name":"Predictive modelling","level":2,"score":0.35589998960494995},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.33059999346733093},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3181000053882599},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.30320000648498535},{"id":"https://openalex.org/C18653775","wikidata":"https://www.wikidata.org/wiki/Q1333358","display_name":"Joint probability distribution","level":2,"score":0.3012999892234802},{"id":"https://openalex.org/C165216359","wikidata":"https://www.wikidata.org/wiki/Q670653","display_name":"Marginal distribution","level":3,"score":0.2888999879360199},{"id":"https://openalex.org/C89106044","wikidata":"https://www.wikidata.org/wiki/Q45284","display_name":"Likelihood function","level":3,"score":0.28679999709129333},{"id":"https://openalex.org/C49781872","wikidata":"https://www.wikidata.org/wiki/Q1045555","display_name":"Maximum likelihood","level":2,"score":0.2827000021934509},{"id":"https://openalex.org/C182081679","wikidata":"https://www.wikidata.org/wiki/Q1275153","display_name":"Expectation\u2013maximization algorithm","level":3,"score":0.2712000012397766},{"id":"https://openalex.org/C61224824","wikidata":"https://www.wikidata.org/wiki/Q2260434","display_name":"Mixture model","level":2,"score":0.2703999876976013},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.2648000121116638},{"id":"https://openalex.org/C111110010","wikidata":"https://www.wikidata.org/wiki/Q2627315","display_name":"Convex combination","level":4,"score":0.2621000111103058},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2572999894618988},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.25679999589920044},{"id":"https://openalex.org/C9483764","wikidata":"https://www.wikidata.org/wiki/Q585740","display_name":"Likelihood-ratio test","level":2,"score":0.25130000710487366},{"id":"https://openalex.org/C83546350","wikidata":"https://www.wikidata.org/wiki/Q1139051","display_name":"Regression","level":2,"score":0.25099998712539673}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.13123","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.13123","pdf_url":"https://arxiv.org/pdf/2610.13123","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.2610.13123","doi":"https://doi.org/10.48550/arxiv.2610.13123","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.13123","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:2610.13123","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.13123","pdf_url":"https://arxiv.org/pdf/2610.13123","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/G2488694901","display_name":"Mathematical Sciences Research Institute (MSRI)","funder_award_id":"1928930","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G329468232","display_name":null,"funder_award_id":"DMS-1928930","funder_id":"https://openalex.org/F4320337380","funder_display_name":"Division of Mathematical Sciences"},{"id":"https://openalex.org/G6019427330","display_name":null,"funder_award_id":"1928930","funder_id":"https://openalex.org/F4320337380","funder_display_name":"Division of Mathematical Sciences"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320306084","display_name":"U.S. Department of Energy","ror":"https://ror.org/01bj3aw27"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"},{"id":"https://openalex.org/F4320337345","display_name":"Office of Naval Research","ror":"https://ror.org/00rk2pe57"},{"id":"https://openalex.org/F4320337380","display_name":"Division of Mathematical Sciences","ror":"https://ror.org/051fftw81"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7221518656.pdf","grobid_xml":"https://content.openalex.org/works/W7221518656.grobid-xml"},"referenced_works_count":11,"referenced_works":["https://openalex.org/W1613885383","https://openalex.org/W2766346350","https://openalex.org/W3001279689","https://openalex.org/W4289600961","https://openalex.org/W4384643890","https://openalex.org/W4385562468","https://openalex.org/W4399510454","https://openalex.org/W4403618979","https://openalex.org/W4407310283","https://openalex.org/W6892060006","https://openalex.org/W7119523234"],"related_works":["https://openalex.org/W4415136440","https://openalex.org/W2912497910","https://openalex.org/W4226320345","https://openalex.org/W7148179057","https://openalex.org/W4417078370","https://openalex.org/W2907762094","https://openalex.org/W4386937939","https://openalex.org/W3006197067","https://openalex.org/W4399511860","https://openalex.org/W7135398029"],"abstract_inverted_index":{"A":[0,51],"solid":[1],"understanding":[2],"of":[3,7,27,56,81],"the":[4,118,127,148,158,166],"predictive":[5,80,154],"success":[6],"deep":[8],"neural":[9],"networks":[10],"(DNNs)":[11],"remains":[12],"elusive.":[13],"For":[14],"example,":[15],"while":[16],"many":[17],"data-dependent":[18],"metrics":[19,30,86,101],"exist":[20],"that":[21,69,139,164,174,190],"seek":[22],"to":[23,34,43,97,116],"predict":[24],"model":[25,57,82,144],"quality":[26,58],"DNNs,":[28],"these":[29,100],"may":[31,40],"be":[32,41,78],"expensive":[33],"compute":[35,44],"for":[36,121,162,194],"large":[37],"datasets":[38,47],"and/or":[39],"impossible":[42],"when":[45],"training":[46],"are":[48],"not":[49],"released.":[50],"practical":[52],"and":[53,135,184],"effective":[54],"predictor":[55],"involves":[59],"analyzing":[60],"DNN":[61],"weight":[62,75],"matrices,":[63],"as":[64],"prior":[65],"work":[66],"has":[67],"established":[68],"heavy-tailed":[70,150],"spectral":[71],"distributions":[72],"in":[73,153,180],"trained":[74],"matrices":[76],"can":[77],"strongly":[79],"quality.":[83,145],"However,":[84],"such":[85],"lack":[87],"a":[88,105,113,122,181],"rigorous":[89],"statistical":[90,106],"justification.":[91],"In":[92],"this":[93,140,175],"work,":[94],"we":[95,111,137,156,185],"aim":[96],"understand":[98,147],"why":[99],"perform":[102],"well,":[103],"from":[104],"perspective.":[107],"To":[108,146],"do":[109],"so,":[110],"employ":[112],"probabilistic":[114],"framework":[115],"derive":[117,157,186],"marginal":[119,141],"likelihood":[120,142,161],"DNN,":[123],"evaluated":[124],"using":[125],"only":[126],"final":[128],"layer":[129],"weights.":[130],"Across":[131],"large-scale":[132,195],"convolutional":[133],"DNNs":[134],"LLMs,":[136],"find":[138],"predicts":[143,192],"role":[149],"spectra":[151,163],"play":[152],"performance,":[155],"limiting":[159,176],"log-marginal":[160],"follow":[165],"(heavy-tailed)":[167],"High-Temperature":[168],"Marchenko-Pastur":[169],"(HTMP)":[170],"distribution.":[171],"We":[172],"show":[173],"quantity":[177],"is":[178],"convex":[179],"heaviness":[182,189],"parameter,":[183],"an":[187],"optimal":[188],"empirically":[191],"over-fitting":[193],"DNNs.":[196]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-13T00:00:00"}