{"id":"https://openalex.org/W7236192542","doi":"https://doi.org/10.48550/arxiv.2611.22449","title":"How Many Training Labels Are Wrong? Auditing a Fitted Model through Gradient Allocation","display_name":"How Many Training Labels Are Wrong? Auditing a Fitted Model through Gradient Allocation","alternate_titles":[],"publication_year":2026,"publication_date":"2026-11-17","ids":{"openalex":"https://openalex.org/W7236192542","doi":"https://doi.org/10.48550/arxiv.2611.22449"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2611.22449","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.22449","pdf_url":"https://arxiv.org/pdf/2611.22449","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"},"parent_works":[],"lineage":["https://openalex.org/W7236192542"],"child_works_count":0,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2611.22449","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5162749350","display_name":"Yuchen Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129604602","display_name":"The University of Sydney","ror":"https://ror.org/0384j8v12","country_code":"AU","type":"education","lineage":["https://openalex.org/I129604602"]}],"countries":["AU"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Li, Yuchen","raw_affiliation_strings":["The University of Sydney"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Sydney","institution_ids":["https://openalex.org/I129604602"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153564694","display_name":"Zongqi Fan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129604602","display_name":"The University of Sydney","ror":"https://ror.org/0384j8v12","country_code":"AU","type":"education","lineage":["https://openalex.org/I129604602"]}],"countries":["AU"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Fan, Zongqi","raw_affiliation_strings":["The University of Sydney"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Sydney","institution_ids":["https://openalex.org/I129604602"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5162774812","display_name":"Xintong Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129604602","display_name":"The University of Sydney","ror":"https://ror.org/0384j8v12","country_code":"AU","type":"education","lineage":["https://openalex.org/I129604602"]}],"countries":["AU"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Li, Xintong","raw_affiliation_strings":["The University of Sydney"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Sydney","institution_ids":["https://openalex.org/I129604602"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027558961","display_name":"Nguyen Hoang Khoi Tran","orcid":"https://orcid.org/0000-0001-7323-9213","observed_orcids":["https://orcid.org/0000-0001-7323-9213"]},"institutions":[],"countries":[],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Tran, Nguyen H.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5129708493","display_name":"Ken-Tye Yong","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129604602","display_name":"The University of Sydney","ror":"https://ror.org/0384j8v12","country_code":"AU","type":"education","lineage":["https://openalex.org/I129604602"]}],"countries":["AU"],"is_corresponding":false,"contributor_roles":[],"raw_author_name":"Yong, Ken-Tye","raw_affiliation_strings":["The University of Sydney"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Sydney","institution_ids":["https://openalex.org/I129604602"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"reviewed_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_pulled":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12535","display_name":"Machine Learning and Data Classification","score":0.9955999851226807,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.9955999851226807,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.0026000000070780516,"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/T10743","display_name":"Software Testing and Debugging Techniques","score":0.00019999999494757503,"subfield":{"id":"https://openalex.org/subfields/1712","display_name":"Software"},"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/noisy-labels","display_name":"label noise","score":0.7239999771118164},{"id":"https://openalex.org/keywords/model-auditing","display_name":"model auditing","score":0.6370000243186951},{"id":"https://openalex.org/keywords/self-influence","display_name":"self-influence","score":0.5759999752044678},{"id":"https://openalex.org/keywords/vision-transformer","display_name":"vision transformer","score":0.41200000047683716},{"id":"https://openalex.org/keywords/label-noise-robustness","display_name":"label noise robustness","score":0.34700000286102295}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6060000061988831},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.5821999907493591},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5583000183105469},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.5141000151634216},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49939998984336853},{"id":"https://openalex.org/C191795146","wikidata":"https://www.wikidata.org/wiki/Q3878446","display_name":"Norm (philosophy)","level":2,"score":0.4952999949455261},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.4560000002384186},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.4083999991416931},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.40619999170303345},{"id":"https://openalex.org/C554936623","wikidata":"https://www.wikidata.org/wiki/Q199657","display_name":"Reading (process)","level":2,"score":0.4056999981403351},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.33090001344680786},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.3278999924659729},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.32429999113082886},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3199000060558319},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3118000030517578},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3057999908924103},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.28519999980926514},{"id":"https://openalex.org/C100675267","wikidata":"https://www.wikidata.org/wiki/Q1371624","display_name":"Background noise","level":2,"score":0.27950000762939453},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.2678000032901764},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2669000029563904}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2611.22449","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.22449","pdf_url":"https://arxiv.org/pdf/2611.22449","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.2611.22449","doi":"https://doi.org/10.48550/arxiv.2611.22449","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.22449","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2611.22449","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.22449","pdf_url":"https://arxiv.org/pdf/2611.22449","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":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7236192542.pdf","grobid_xml":"https://content.openalex.org/works/W7236192542.grobid-xml"},"referenced_works_count":15,"referenced_works":["https://openalex.org/W1921293667","https://openalex.org/W2682189153","https://openalex.org/W2905631704","https://openalex.org/W2969769592","https://openalex.org/W2988175183","https://openalex.org/W2996108195","https://openalex.org/W3004146535","https://openalex.org/W3006856964","https://openalex.org/W3088647504","https://openalex.org/W3102583815","https://openalex.org/W3119786062","https://openalex.org/W3175937066","https://openalex.org/W3177900337","https://openalex.org/W4312443924","https://openalex.org/W7216531531"],"related_works":["https://openalex.org/W7220427833","https://openalex.org/W7165927710","https://openalex.org/W7214726705","https://openalex.org/W7223692667","https://openalex.org/W2608085836","https://openalex.org/W2396424716","https://openalex.org/W4386798110","https://openalex.org/W3146670357","https://openalex.org/W7221307151","https://openalex.org/W7235826944"],"abstract_inverted_index":{"How":[0],"many":[1],"labels":[2,181,191,271,302],"in":[3,87],"a":[4,16,51,72,88,99,195,217,264,319],"training":[5],"set":[6,18,31,159],"are":[7,272],"wrong,":[8],"and":[9,34,57,102,134,152,160,201,207,251,310],"which":[10,295],"ones?":[11],"The":[12,147],"usual":[13],"answers":[14],"read":[15,64],"held-out":[17,24,30,158],"or":[19,227],"the":[20,29,40,48,58,65,68,83,93,106,135,141,144,165,174,188,202,204,208,222,243,248,256,269,274,282,313],"model's":[21],"outputs.":[22],"A":[23],"reading":[25,37,142,275],"depends":[26],"on":[27,187,238,286,332],"how":[28],"was":[32],"drawn,":[33],"an":[35,183,211],"output-based":[36],"fails":[38],"once":[39],"model":[41,166],"has":[42,149,167],"fitted":[43,52,168,218],"its":[44,80,96,120,125,131,171],"wrong":[45,53,172,178,261,270,301],"labels,":[46,173],"because":[47],"loss":[49],"of":[50,71,95,109,137,143,170,185,192,210,221,236,247,255,281,324],"label":[54],"is":[55,114,140],"small":[56],"prediction":[59],"agrees":[60],"with":[61,79,105,182,240],"it.":[62],"We":[63,91],"answer":[66],"from":[67,179,225,288],"per-example":[69],"gradients":[70],"single":[73],"trained":[74,266],"model.":[75],"Each":[76],"example,":[77],"together":[78],"label,":[81,133],"asks":[82],"network":[84],"to":[85,199,213,231,290,329],"change":[86],"particular":[89],"way.":[90],"normalise":[92],"magnitudes":[94],"gradient":[97,100,205],"into":[98],"allocation":[101,108,126,175],"compare":[103],"it":[104,118,258],"mean":[107],"every":[110],"class.":[111],"An":[112,322],"example":[113,212],"flagged":[115,138],"when":[116],"relabelling":[117],"as":[119,304],"nearest":[121],"other":[122],"class":[123],"makes":[124],"fit":[127],"better":[128],"than":[129],"under":[130],"own":[132],"share":[136],"examples":[139,257,326],"noise":[145,223,284,314],"rate.":[146],"rule":[148],"no":[150,154,157,161],"threshold":[151],"needs":[153],"clean":[155],"data,":[156],"further":[162,298],"training.":[163],"After":[164],"99-100%":[169],"still":[176],"separates":[177,300],"right":[180],"AUC":[184],"0.94-0.96":[186],"noisiest":[189],"human":[190],"CIFAR-10N,":[193],"where":[194],"feature-space":[196],"margin":[197],"falls":[198],"0.79-0.84":[200],"loss,":[203],"norm":[206],"self-influence":[209],"0.73-0.76.":[214],"On":[215,263],"such":[216],"model,":[219,267],"readings":[220],"rate":[224,285,315],"predictions":[226],"feature":[228],"margins":[229],"fall":[230],"0-6%":[232],"at":[233,252,318],"true":[234,249,283],"rates":[235],"8-58%":[237],"CIFAR-10":[239,287],"ViT-S,":[241],"while":[242],"audit":[244,323],"reads":[245,312],"54-96%":[246],"rate,":[250],"least":[253],"93%":[254],"counts":[259],"carry":[260],"labels.":[262],"briefly":[265],"before":[268],"fitted,":[273],"lies":[276],"within":[277],"1.7":[278],"percentage":[279],"points":[280],"0%":[289],"58%":[291],"noise.":[292],"Confident":[293],"learning,":[294],"trains":[296],"four":[297],"models,":[299],"about":[303],"well":[305],"(AUC":[306],"0.925-0.998":[307],"against":[308],"0.900-0.995)":[309],"mostly":[311],"more":[316],"closely,":[317],"lower":[320],"precision.":[321],"5,000":[325],"takes":[327],"14":[328],"27":[330],"seconds":[331],"one":[333],"GPU.":[334]},"cited_by_api_url":"https://api.openalex.org/works?filter=cites:W7236192542","counts_by_year":[],"updated_date":"2026-11-22T12:46:00.618207","created_date":"2026-11-19T00:00:00"}
