{"id":"https://openalex.org/W7152708613","doi":"https://doi.org/10.48550/arxiv.2604.06366","title":"Stochastic Gradient Descent in the Saddle-to-Saddle Regime of Deep Linear Networks","display_name":"Stochastic Gradient Descent in the Saddle-to-Saddle Regime of Deep Linear Networks","publication_year":2026,"publication_date":"2026-04-07","ids":{"openalex":"https://openalex.org/W7152708613","doi":"https://doi.org/10.48550/arxiv.2604.06366"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.06366","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06366","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.06366","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133308946","display_name":"Guillaume Corlouer","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Corlouer, Guillaume","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133309943","display_name":"Avi Semler","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Semler, Avi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121043476","display_name":"Alexander Strang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Strang, Alexander","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133247731","display_name":"Alexander Gietelink Oldenziel","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Oldenziel, Alexander Gietelink","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5133308946"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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":0.3630000054836273,"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":0.3630000054836273,"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":0.2287999987602234,"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"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.07039999961853027,"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/stochastic-gradient-descent","display_name":"Stochastic gradient descent","score":0.6837000250816345},{"id":"https://openalex.org/keywords/stationary-distribution","display_name":"Stationary distribution","score":0.6671000123023987},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.6444000005722046},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.5508999824523926},{"id":"https://openalex.org/keywords/langevin-dynamics","display_name":"Langevin dynamics","score":0.45680001378059387},{"id":"https://openalex.org/keywords/distribution","display_name":"Distribution (mathematics)","score":0.41510000824928284},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.4092000126838684},{"id":"https://openalex.org/keywords/dynamics","display_name":"Dynamics (music)","score":0.3882000148296356},{"id":"https://openalex.org/keywords/stochastic-process","display_name":"Stochastic process","score":0.38100001215934753}],"concepts":[{"id":"https://openalex.org/C206688291","wikidata":"https://www.wikidata.org/wiki/Q7617819","display_name":"Stochastic gradient descent","level":3,"score":0.6837000250816345},{"id":"https://openalex.org/C98951983","wikidata":"https://www.wikidata.org/wiki/Q7604341","display_name":"Stationary distribution","level":3,"score":0.6671000123023987},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.6444000005722046},{"id":"https://openalex.org/C121864883","wikidata":"https://www.wikidata.org/wiki/Q677916","display_name":"Statistical physics","level":1,"score":0.593500018119812},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.5508999824523926},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.517799973487854},{"id":"https://openalex.org/C2780004032","wikidata":"https://www.wikidata.org/wiki/Q6485978","display_name":"Langevin dynamics","level":2,"score":0.45680001378059387},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.41510000824928284},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.4092000126838684},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.40299999713897705},{"id":"https://openalex.org/C145912823","wikidata":"https://www.wikidata.org/wiki/Q113558","display_name":"Dynamics (music)","level":2,"score":0.3882000148296356},{"id":"https://openalex.org/C8272713","wikidata":"https://www.wikidata.org/wiki/Q176737","display_name":"Stochastic process","level":2,"score":0.38100001215934753},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.3668999969959259},{"id":"https://openalex.org/C51955184","wikidata":"https://www.wikidata.org/wiki/Q1545585","display_name":"Stochastic differential equation","level":2,"score":0.34929999709129333},{"id":"https://openalex.org/C86582703","wikidata":"https://www.wikidata.org/wiki/Q7617824","display_name":"Stochastic neural network","level":4,"score":0.3465000092983246},{"id":"https://openalex.org/C35304006","wikidata":"https://www.wikidata.org/wiki/Q5962","display_name":"Boltzmann constant","level":2,"score":0.3294000029563904},{"id":"https://openalex.org/C69357855","wikidata":"https://www.wikidata.org/wiki/Q163214","display_name":"Diffusion","level":2,"score":0.321399986743927},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3068000078201294},{"id":"https://openalex.org/C193999330","wikidata":"https://www.wikidata.org/wiki/Q834200","display_name":"Boltzmann distribution","level":2,"score":0.30489999055862427},{"id":"https://openalex.org/C112633086","wikidata":"https://www.wikidata.org/wiki/Q381287","display_name":"White noise","level":2,"score":0.302700012922287},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.2912999987602234},{"id":"https://openalex.org/C192576344","wikidata":"https://www.wikidata.org/wiki/Q194706","display_name":"Boltzmann machine","level":3,"score":0.29019999504089355},{"id":"https://openalex.org/C50318809","wikidata":"https://www.wikidata.org/wiki/Q2145081","display_name":"Replicator equation","level":3,"score":0.2736000120639801},{"id":"https://openalex.org/C127491075","wikidata":"https://www.wikidata.org/wiki/Q7617825","display_name":"Stochastic modelling","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C112401455","wikidata":"https://www.wikidata.org/wiki/Q178036","display_name":"Brownian motion","level":2,"score":0.26030001044273376},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.2578999996185303},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.25619998574256897}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.06366","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06366","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.06366","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06366","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Deep":[0],"linear":[1],"networks":[2],"(DLNs)":[3],"are":[4],"used":[5],"as":[6,63],"an":[7,81],"analytically":[8],"tractable":[9],"model":[10,59],"of":[11,15,32,48,52,74,84,90,118,126,140,147,182],"the":[12,30,46,55,60,72,85,99,106,115,124,137,145,160,180,190],"training":[13,51,61],"dynamics":[14,47,62,66,86],"deep":[16],"neural":[17],"networks.":[18],"While":[19],"gradient":[20,34,141],"descent":[21,35],"in":[22,54,123,144],"DLNs":[23,53],"is":[24],"known":[25],"to":[26],"exhibit":[27],"saddle-to-saddle":[28,56,191],"dynamics,":[29],"impact":[31],"stochastic":[33,64,93],"(SGD)":[36],"noise":[37,149,176],"on":[38],"this":[39],"regime":[40],"remains":[41],"poorly":[42],"understood.":[43],"We":[44,58,112],"investigate":[45],"SGD":[49,119,175],"during":[50],"regime.":[57],"Langevin":[65],"with":[67,136],"anisotropic,":[68],"state-dependent":[69],"noise.":[70],"Under":[71],"assumption":[73],"aligned":[75,167],"and":[76],"balanced":[77,169],"weights,":[78],"we":[79,156],"derive":[80,114],"exact":[82],"decomposition":[83],"into":[87],"a":[88,103,152],"system":[89],"one-dimensional":[91],"per-mode":[92],"differential":[94],"equations.":[95],"This":[96],"establishes":[97],"that":[98,159,174],"maximal":[100],"diffusion":[101],"along":[102,132],"mode":[104],"precedes":[105],"corresponding":[107],"feature":[108,183],"being":[109],"completely":[110],"learned.":[111],"also":[113],"stationary":[116,138],"distribution":[117,131,139],"for":[120],"each":[121],"mode:":[122],"absence":[125],"label":[127,148],"noise,":[128],"its":[129],"marginal":[130],"specific":[133],"features":[134],"coincides":[135],"flow,":[142],"while":[143],"presence":[146],"it":[150],"approximates":[151],"Boltzmann":[153],"distribution.":[154],"Finally,":[155],"confirm":[157],"experimentally":[158],"theoretical":[161],"results":[162,172],"hold":[163],"qualitatively":[164],"even":[165],"without":[166],"or":[168],"weights.":[170],"These":[171],"establish":[173],"encodes":[177],"information":[178],"about":[179],"progression":[181],"learning":[184],"but":[185],"does":[186],"not":[187],"fundamentally":[188],"alter":[189],"dynamics.":[192]},"counts_by_year":[],"updated_date":"2026-04-10T06:07:51.998497","created_date":"2026-04-10T00:00:00"}