{"id":"https://openalex.org/W7214766825","doi":"https://doi.org/10.48550/arxiv.2609.31983","title":"Understanding the Subspace Stabilization of the Hessian and Gradient Covariance Matrix","display_name":"Understanding the Subspace Stabilization of the Hessian and Gradient Covariance Matrix","publication_year":2026,"publication_date":"2026-09-25","ids":{"openalex":"https://openalex.org/W7214766825","doi":"https://doi.org/10.48550/arxiv.2609.31983"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.31983","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31983","pdf_url":"https://arxiv.org/pdf/2609.31983","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/2609.31983","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5072289737","display_name":"Fangshuo Liao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I74775410","display_name":"Rice University","ror":"https://ror.org/008zs3103","country_code":"US","type":"education","lineage":["https://openalex.org/I74775410"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liao, Fangshuo","raw_affiliation_strings":["Rice University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rice University","institution_ids":["https://openalex.org/I74775410"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5024280658","display_name":"Anastasios Kyrillidis","orcid":"https://orcid.org/0000-0003-1954-8730","observed_orcids":["https://orcid.org/0000-0003-1954-8730"]},"institutions":[{"id":"https://openalex.org/I74775410","display_name":"Rice University","ror":"https://ror.org/008zs3103","country_code":"US","type":"education","lineage":["https://openalex.org/I74775410"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kyrillidis, Anastasios","raw_affiliation_strings":["Rice University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rice University","institution_ids":["https://openalex.org/I74775410"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I74775410"],"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":0.9897000193595886,"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":0.9897000193595886,"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":0.00570000009611249,"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/T10792","display_name":"Matrix Theory and Algorithms","score":0.0005000000237487257,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/hessian-matrix","display_name":"Hessian matrix","score":0.7379999756813049},{"id":"https://openalex.org/keywords/second-order-information","display_name":"second-order information","score":0.699999988079071},{"id":"https://openalex.org/keywords/neural-network-training","display_name":"neural network training","score":0.9470000267028809},{"id":"https://openalex.org/keywords/outlier-eigenvalues","display_name":"outlier eigenvalues","score":0.6919999718666077},{"id":"https://openalex.org/keywords/gradient-flow","display_name":"gradient flow","score":0.7179999947547913}],"concepts":[{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.8888999819755554},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.7785999774932861},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.6585999727249146},{"id":"https://openalex.org/C73729460","wikidata":"https://www.wikidata.org/wiki/Q17023339","display_name":"Hessian equation","level":4,"score":0.5652999877929688},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.5322999954223633},{"id":"https://openalex.org/C178650346","wikidata":"https://www.wikidata.org/wiki/Q201984","display_name":"Covariance","level":2,"score":0.5236999988555908},{"id":"https://openalex.org/C12362212","wikidata":"https://www.wikidata.org/wiki/Q728435","display_name":"Linear subspace","level":2,"score":0.4788999855518341},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.47769999504089355},{"id":"https://openalex.org/C185142706","wikidata":"https://www.wikidata.org/wiki/Q1134404","display_name":"Covariance matrix","level":2,"score":0.4429999887943268},{"id":"https://openalex.org/C169756996","wikidata":"https://www.wikidata.org/wiki/Q194919","display_name":"Eigendecomposition of a matrix","level":3,"score":0.4287000000476837},{"id":"https://openalex.org/C158693339","wikidata":"https://www.wikidata.org/wiki/Q190524","display_name":"Eigenvalues and eigenvectors","level":2,"score":0.4219000041484833},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4165000021457672},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.38589999079704285},{"id":"https://openalex.org/C167879884","wikidata":"https://www.wikidata.org/wiki/Q727568","display_name":"Balanced flow","level":2,"score":0.35350000858306885},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.33799999952316284},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.28839999437332153},{"id":"https://openalex.org/C38349280","wikidata":"https://www.wikidata.org/wiki/Q1434290","display_name":"Flow (mathematics)","level":2,"score":0.28690001368522644},{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.2824999988079071},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.25769999623298645}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.31983","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31983","pdf_url":"https://arxiv.org/pdf/2609.31983","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.2609.31983","doi":"https://doi.org/10.48550/arxiv.2609.31983","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.31983","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:2609.31983","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31983","pdf_url":"https://arxiv.org/pdf/2609.31983","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/W7214766825.pdf","grobid_xml":"https://content.openalex.org/works/W7214766825.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4399394287","https://openalex.org/W2914462410","https://openalex.org/W4415344577","https://openalex.org/W2904243021","https://openalex.org/W7223554795","https://openalex.org/W4417300196","https://openalex.org/W7162638367","https://openalex.org/W4405768271","https://openalex.org/W6948129847","https://openalex.org/W7166712898"],"abstract_inverted_index":{"The":[0],"phenomenon":[1,185],"of":[2,7,19,23,34,46,62,87,90,107,121,136,140,159,175,186,191],"the":[3,8,20,31,35,40,43,47,59,74,88,94,99,112,122,137,141,144,149,156,160,166,169,172,176,184,187,192],"top":[4,32,44,109,188],"subspace":[5,33,75,82,110,139,189],"stabilization":[6,53,83,190],"Hessian":[9,36,142,177,193],"matrix":[10,102],"is":[11,85,163],"an":[12,128,181],"surprising":[13],"and":[14,50,77,118,143,171],"critical":[15],"aspect":[16],"in":[17,58],"study":[18],"second-order":[21],"information":[22],"neural":[24],"network":[25],"training.":[26,63],"Prior":[27],"work":[28],"argues":[29],"that":[30,84,98,131,155],"stabilizes":[37],"by":[38],"measuring":[39],"overlap":[41],"between":[42,168],"subspaces":[45],"step-wise":[48],"Hessian,":[49],"explains":[51],"this":[52,65],"with":[54],"diminishing":[55],"parameter":[56,91],"change":[57],"late":[60],"phase":[61],"In":[64,93,148],"paper,":[66],"we":[67,96,126,153],"define":[68],"a":[69,104,116,133],"new":[70],"instability":[71],"metric":[72],"for":[73],"evolution,":[76],"use":[78],"it":[79],"to":[80,111,165,183],"detect":[81],"independent":[86],"magnitude":[89],"change.":[92],"meantime,":[95],"observe":[97],"gradient":[100,123,145,150],"covariance":[101,124,146],"has":[103],"similar":[105],"property":[106],"its":[108],"Hessian.":[113],"By":[114],"using":[115],"between-class":[117],"within-class":[119],"decomposition":[120],"matrix,":[125,178],"identify":[127],"explicit":[129],"form":[130],"gives":[132],"near-perfect":[134],"approximation":[135,162],"top-$(C-1)$":[138],"matrix.":[147,194],"flow":[151],"set-up,":[152],"show":[154],"slow":[157],"evolution":[158],"idenfied":[161],"due":[164],"separation":[167],"outlier":[170],"bulk":[173],"eigenvalues":[174],"thus":[179],"providing":[180],"explanation":[182]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-30T00:00:00"}