{"id":"https://openalex.org/W7164857924","doi":"https://doi.org/10.1145/3805622.3810828","title":"Unveiling PEFT Robustness to Noisy Labels in VLMs: A Gradient-Loss Decoupling Perspective","display_name":"Unveiling PEFT Robustness to Noisy Labels in VLMs: A Gradient-Loss Decoupling Perspective","publication_year":2026,"publication_date":"2026-06-15","ids":{"openalex":"https://openalex.org/W7164857924","doi":"https://doi.org/10.1145/3805622.3810828"},"language":null,"primary_location":{"id":"doi:10.1145/3805622.3810828","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805622.3810828","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 International Conference on Multimedia Retrieval","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3805622.3810828","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5072947645","display_name":"T K","orcid":null},"institutions":[{"id":"https://openalex.org/I47720641","display_name":"Huazhong University of Science and Technology","ror":"https://ror.org/00p991c53","country_code":"CN","type":"education","lineage":["https://openalex.org/I47720641"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tengfei Ma","raw_affiliation_strings":["Huazhong University of Science and technology, Wuhan, China"],"raw_orcid":"https://orcid.org/0009-0003-7179-897X","affiliations":[{"raw_affiliation_string":"Huazhong University of Science and technology, Wuhan, China","institution_ids":["https://openalex.org/I47720641"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008492424","display_name":"Weiran Pan","orcid":null},"institutions":[{"id":"https://openalex.org/I47720641","display_name":"Huazhong University of Science and Technology","ror":"https://ror.org/00p991c53","country_code":"CN","type":"education","lineage":["https://openalex.org/I47720641"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Weiran Pan","raw_affiliation_strings":["Huazhong University of Science and technology, Wuhan, China"],"raw_orcid":"https://orcid.org/0009-0000-8524-6226","affiliations":[{"raw_affiliation_string":"Huazhong University of Science and technology, Wuhan, China","institution_ids":["https://openalex.org/I47720641"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100323842","display_name":"Wei Wei","orcid":"https://orcid.org/0000-0003-4488-0102"},"institutions":[{"id":"https://openalex.org/I47720641","display_name":"Huazhong University of Science and Technology","ror":"https://ror.org/00p991c53","country_code":"CN","type":"education","lineage":["https://openalex.org/I47720641"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wei Wei","raw_affiliation_strings":["Huazhong University of Science and technology, Wuhan, China"],"raw_orcid":"https://orcid.org/0000-0003-4488-0102","affiliations":[{"raw_affiliation_string":"Huazhong University of Science and technology, Wuhan, China","institution_ids":["https://openalex.org/I47720641"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.95080897,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"967","last_page":"976"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12535","display_name":"Machine Learning and Data Classification","score":0.6563000082969666,"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":0.6563000082969666,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.09210000187158585,"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":0.051899999380111694,"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/robustness","display_name":"Robustness (evolution)","score":0.8495000004768372},{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.6230000257492065},{"id":"https://openalex.org/keywords/decoupling","display_name":"Decoupling (probability)","score":0.4627000093460083},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.45820000767707825},{"id":"https://openalex.org/keywords/robust-statistics","display_name":"Robust statistics","score":0.373199999332428},{"id":"https://openalex.org/keywords/perspective","display_name":"Perspective (graphical)","score":0.36959999799728394}],"concepts":[{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.8495000004768372},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.671500027179718},{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.6230000257492065},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5228000283241272},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4848000109195709},{"id":"https://openalex.org/C205606062","wikidata":"https://www.wikidata.org/wiki/Q5249645","display_name":"Decoupling (probability)","level":2,"score":0.4627000093460083},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.45820000767707825},{"id":"https://openalex.org/C67226441","wikidata":"https://www.wikidata.org/wiki/Q1665389","display_name":"Robust statistics","level":3,"score":0.373199999332428},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.36959999799728394},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.36149999499320984},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.35519999265670776},{"id":"https://openalex.org/C29265498","wikidata":"https://www.wikidata.org/wiki/Q7047719","display_name":"Noise measurement","level":3,"score":0.3492000102996826},{"id":"https://openalex.org/C2781170535","wikidata":"https://www.wikidata.org/wiki/Q30587856","display_name":"Noisy data","level":2,"score":0.33820000290870667},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.29980000853538513},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2881999909877777},{"id":"https://openalex.org/C167085575","wikidata":"https://www.wikidata.org/wiki/Q6803654","display_name":"Mean squared prediction error","level":2,"score":0.26669999957084656},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2605000138282776}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3805622.3810828","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805622.3810828","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 International Conference on Multimedia Retrieval","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3805622.3810828","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3805622.3810828","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 International Conference on Multimedia Retrieval","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":34,"referenced_works":["https://openalex.org/W1977295328","https://openalex.org/W2047643928","https://openalex.org/W2138011018","https://openalex.org/W2155904486","https://openalex.org/W2533598788","https://openalex.org/W2962762068","https://openalex.org/W2963465031","https://openalex.org/W2964194231","https://openalex.org/W2981873476","https://openalex.org/W3034185248","https://openalex.org/W3198377975","https://openalex.org/W4312305885","https://openalex.org/W4312310776","https://openalex.org/W4312956471","https://openalex.org/W4313175608","https://openalex.org/W4386071547","https://openalex.org/W4386072228","https://openalex.org/W4386075788","https://openalex.org/W4386790226","https://openalex.org/W4390871786","https://openalex.org/W4390873312","https://openalex.org/W4394625869","https://openalex.org/W4399197779","https://openalex.org/W4402502610","https://openalex.org/W4402702914","https://openalex.org/W4402704589","https://openalex.org/W4402727000","https://openalex.org/W4402916156","https://openalex.org/W4404783209","https://openalex.org/W4413145028","https://openalex.org/W4415797400","https://openalex.org/W7103751069","https://openalex.org/W7133212323","https://openalex.org/W7133219196"],"related_works":[],"abstract_inverted_index":{"Parameter-Efficient":[0],"Fine-Tuning":[1],"(PEFT)":[2],"has":[3],"emerged":[4],"as":[5,112],"the":[6,146],"dominant":[7],"paradigm":[8],"for":[9],"adapting":[10],"Vision-Language":[11],"Models":[12],"(VLMs),":[13],"yet":[14],"its":[15],"robustness":[16,48],"to":[17,68,131],"label":[18],"noise":[19,41,117,150],"remains":[20],"largely":[21],"unexplored.":[22],"To":[23],"address":[24],"this":[25,66,102,123],"gap,":[26],"we":[27,71,120],"conduct":[28],"a":[29,46,69,82],"systematic":[30],"evaluation":[31],"of":[32,149],"representative":[33],"PEFT":[34],"methods":[35,50],"across":[36],"eight":[37],"datasets":[38],"under":[39],"diverse":[40],"conditions.":[42],"Our":[43,75],"experiments":[44],"reveal":[45],"consistent":[47],"hierarchy:":[49],"like":[51,142],"Tip-Adapter-F":[52],"and":[53,88],"LoRA":[54],"demonstrate":[55],"high":[56],"resilience,":[57],"whereas":[58],"Linear":[59],"Probing":[60],"suffers":[61],"significant":[62],"degradation.":[63],"We":[64,100],"attribute":[65],"disparity":[67],"mechanism":[70],"term":[72],"Gradient-Loss":[73],"Decoupling.":[74],"analysis":[76],"shows":[77],"that":[78,107,128,135],"robust":[79],"architectures":[80],"maintain":[81],"\u201cloose":[83],"coupling\"":[84],"between":[85],"prediction":[86],"error":[87],"gradient":[89,136],"magnitude,":[90],"effectively":[91],"preventing":[92],"high-loss":[93],"outliers":[94],"from":[95],"generating":[96],"disruptive":[97],"parameter":[98],"updates.":[99],"substantiate":[101],"through":[103,125],"mathematical":[104],"derivation,":[105],"proving":[106],"specific":[108],"architectural":[109],"constraints,":[110],"such":[111],"affinity-based":[113],"gating,":[114],"naturally":[115],"suppress":[116],"gradients.":[118],"Finally,":[119],"empirically":[121],"validate":[122],"framework":[124],"targeted":[126],"interventions":[127],"enforce":[129],"decoupling":[130],"recover":[132],"robustness,":[133],"confirming":[134],"dynamics,":[137],"rather":[138],"than":[139],"simple":[140],"factors":[141],"model":[143],"capacity,":[144],"are":[145],"key":[147],"drivers":[148],"resilience":[151],"in":[152],"VLM":[153],"adaptation.":[154]},"counts_by_year":[],"updated_date":"2026-06-16T07:37:23.134862","created_date":"2026-06-16T00:00:00"}