{"id":"https://openalex.org/W3112412645","doi":null,"title":"Finding Critical and Gradient-Flat Points of Deep Neural Network Loss Functions","display_name":"Finding Critical and Gradient-Flat Points of Deep Neural Network Loss Functions","publication_year":2020,"publication_date":"2020-01-01","ids":{"openalex":"https://openalex.org/W3112412645","mag":"3112412645"},"language":"en","primary_location":{"id":"pmh:oai:digitalcommons.imsa.edu:alumni_dissertations-1025","doi":null,"is_oa":false,"landing_page_url":"https://digitalcommons.imsa.edu/alumni_dissertations/26","pdf_url":null,"source":{"id":"https://openalex.org/S4377196676","display_name":"DigitalCommons-IMSA (Illinois Mathematics and Science Academy)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I4210116845","host_organization_name":"Illinois Mathematics and Science Academy","host_organization_lineage":["https://openalex.org/I4210116845"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Doctoral Dissertations","raw_type":"text"},"volume":null,"type":"dissertation","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5084043552","display_name":"Charles G. Frye","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Charles G. Frye","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5084043552"],"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":0.9794999957084656,"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":0.9794999957084656,"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":0.01510000042617321,"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/T10963","display_name":"Advanced Optimization Algorithms Research","score":0.0015999999595806003,"subfield":{"id":"https://openalex.org/subfields/2612","display_name":"Numerical Analysis"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/critical-points","display_name":"critical points","score":0.9459999799728394},{"id":"https://openalex.org/keywords/nonconvex-optimization","display_name":"non-convex optimization","score":0.8289999961853027},{"id":"https://openalex.org/keywords/local-curvature","display_name":"local curvature","score":0.6710000038146973},{"id":"https://openalex.org/keywords/random-matrix-theory","display_name":"random matrix theory","score":0.6359999775886536}],"concepts":[{"id":"https://openalex.org/C186633575","wikidata":"https://www.wikidata.org/wiki/Q845060","display_name":"Maxima and minima","level":2,"score":0.8496189117431641},{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.7812427282333374},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.6705130338668823},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5603106617927551},{"id":"https://openalex.org/C195065555","wikidata":"https://www.wikidata.org/wiki/Q214881","display_name":"Curvature","level":2,"score":0.5034922957420349},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4494740664958954},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.4367135465145111},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.4319068491458893},{"id":"https://openalex.org/C196298200","wikidata":"https://www.wikidata.org/wiki/Q577705","display_name":"Critical point (mathematics)","level":2,"score":0.41944289207458496},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4187823534011841},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.36707374453544617},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3408862054347992},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.27342432737350464},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.21560126543045044},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.1829315721988678},{"id":"https://openalex.org/C78458016","wikidata":"https://www.wikidata.org/wiki/Q840400","display_name":"Evolutionary biology","level":1,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:digitalcommons.imsa.edu:alumni_dissertations-1025","doi":null,"is_oa":false,"landing_page_url":"https://digitalcommons.imsa.edu/alumni_dissertations/26","pdf_url":null,"source":{"id":"https://openalex.org/S4377196676","display_name":"DigitalCommons-IMSA (Illinois Mathematics and Science Academy)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I4210116845","host_organization_name":"Illinois Mathematics and Science Academy","host_organization_lineage":["https://openalex.org/I4210116845"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Doctoral Dissertations","raw_type":"text"},{"id":"pmh:oai:escholarship.org/ark:/13030/qt4fw6x5b3","doi":null,"is_oa":false,"landing_page_url":"https://escholarship.org/uc/item/4fw6x5b3","pdf_url":null,"source":{"id":"https://openalex.org/S4306400115","display_name":"eScholarship (California Digital Library)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I2801248553","host_organization_name":"California Digital Library","host_organization_lineage":["https://openalex.org/I2801248553"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"etd"},{"id":"mag:3112412645","doi":null,"is_oa":false,"landing_page_url":"https://escholarship.org/content/qt4fw6x5b3/qt4fw6x5b3.pdf?t=qhtt5i","pdf_url":null,"source":null,"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":null}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":54,"referenced_works":["https://openalex.org/W602904462","https://openalex.org/W1529395041","https://openalex.org/W1567512734","https://openalex.org/W1666670079","https://openalex.org/W1697075315","https://openalex.org/W1969811075","https://openalex.org/W1974511160","https://openalex.org/W1982097255","https://openalex.org/W2006903949","https://openalex.org/W2012960907","https://openalex.org/W2030521309","https://openalex.org/W2041792275","https://openalex.org/W2042860072","https://openalex.org/W2043938310","https://openalex.org/W2058194880","https://openalex.org/W2073370371","https://openalex.org/W2078626246","https://openalex.org/W2079135578","https://openalex.org/W2101234009","https://openalex.org/W2101762657","https://openalex.org/W2101961546","https://openalex.org/W2105724942","https://openalex.org/W2131703182","https://openalex.org/W2137234026","https://openalex.org/W2145889472","https://openalex.org/W2146502635","https://openalex.org/W2152703617","https://openalex.org/W2155894447","https://openalex.org/W2187365590","https://openalex.org/W2296319761","https://openalex.org/W2328019091","https://openalex.org/W2604901683","https://openalex.org/W2618381130","https://openalex.org/W2626325961","https://openalex.org/W2767286248","https://openalex.org/W2769394111","https://openalex.org/W2894604724","https://openalex.org/W2896083421","https://openalex.org/W2899748887","https://openalex.org/W2912713668","https://openalex.org/W2913473169","https://openalex.org/W2915116423","https://openalex.org/W2919651864","https://openalex.org/W2963326510","https://openalex.org/W2963588390","https://openalex.org/W2964072429","https://openalex.org/W2964121744","https://openalex.org/W2970971581","https://openalex.org/W2986941256","https://openalex.org/W2987473824","https://openalex.org/W2994927236","https://openalex.org/W3005768598","https://openalex.org/W3012856280","https://openalex.org/W4212915593"],"related_works":["https://openalex.org/W3012856280","https://openalex.org/W2765428107","https://openalex.org/W2963651774","https://openalex.org/W7008742394","https://openalex.org/W2787132992","https://openalex.org/W2608609325","https://openalex.org/W4416445141","https://openalex.org/W2996284974","https://openalex.org/W2998438489","https://openalex.org/W4287065277"],"abstract_inverted_index":{"Despite":[0],"the":[1,4,19,74,81,92,141,174,177,186,195,201],"fact":[2],"that":[3,91,103,124,133,140,215],"loss":[5,82,93,187,227],"functions":[6,94],"of":[7,49,60,63,80,136,148,159,176,180,197,222],"deep":[8],"neural":[9,28,98,149,225],"networks":[10,29,99],"are":[11,86,104,218],"highly":[12],"non-convex,gradient-based":[13],"optimization":[14],"algorithms":[15,48],"converge":[16],"to":[17,31,46,96,190,204],"approximately":[18,216],"same":[20],"performancefrom":[21],"many":[22],"random":[23],"initial":[24],"points.":[25],"This":[26,162],"makes":[27],"easy":[30],"train,":[32],"which,":[33],"combined":[34],"with":[35,52],"their":[36,160],"high":[37,50],"representational":[38],"capacity":[39],"and":[40,42],"implicit":[41],"explicit":[43],"regularizationstrategies,":[44],"leads":[45],"machine-learned":[47],"quality":[51],"reasonable":[53],"computational":[54],"cost":[55],"in":[56,173,194],"a":[57,155,219],"wide":[58],"variety":[59],"domains.One":[61],"thread":[62],"work":[64],"has":[65,122],"focused":[66],"on":[67],"explaining":[68],"this":[69,129,137],"phenomenon":[70],"by":[71,112,166],"numerically":[72],"characterizing":[73],"local":[75],"curvature":[76],"at":[77],"critical":[78,146,209],"points":[79,147],"function,":[83],"where":[84],"gradients":[85],"zero.Such":[87],"studies":[88],"have":[89,100],"reported":[90],"used":[95,143],"train":[97],"nolocal":[101],"minima":[102,126],"much":[105],"worse":[106],"than":[107],"global":[108],"minima,":[109],"backed":[110],"up":[111],"arguments":[113],"fromrandom":[114],"matrix":[115,179],"theory.":[116],"More":[117],"recent":[118],"theoretical":[119],"work,":[120],"however,":[121],"suggested":[123],"badlocal":[125],"do":[127],"exist.In":[128],"dissertation,":[130],"we":[131],"show":[132],"one":[134],"cause":[135],"gap":[138],"is":[139,164,172],"methods":[142],"tonumerically":[144],"find":[145],"network":[150,226],"losses":[151],"suffer,":[152],"ironically,":[153],"from":[154],"bad":[156],"localminimum":[157],"problem":[158,163],"own.":[161],"caused":[165],"gradient-flat":[167],"points,":[168,185],"wherethe":[169],"gradient":[170],"vector":[171],"kernel":[175],"Hessian":[178],"second":[181,191],"partial":[182],"derivatives.At":[183],"these":[184],"function":[188],"becomes,":[189],"order,":[192],"linear":[193],"direction":[196],"thegradient,":[198],"which":[199],"violates":[200],"assumptions":[202],"necessary":[203],"guarantee":[205],"convergence":[206],"for":[207],"secondorder":[208],"point-finding":[210],"methods.":[211],"We":[212],"present":[213],"evidence":[214],"gradient-flatpoints":[217],"common":[220],"feature":[221],"several":[223],"prototypical":[224],"functions.":[228]},"counts_by_year":[],"updated_date":"2026-11-11T12:16:26.678958","created_date":"2020-12-21T00:00:00"}