{"id":"https://openalex.org/W7135201902","doi":"https://doi.org/10.48550/arxiv.2603.11793","title":"Locating Demographic Bias at the Attention-Head Level in CLIP's Vision Encoder","display_name":"Locating Demographic Bias at the Attention-Head Level in CLIP's Vision Encoder","publication_year":2026,"publication_date":"2026-03-12","ids":{"openalex":"https://openalex.org/W7135201902","doi":"https://doi.org/10.48550/arxiv.2603.11793"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.11793","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.11793","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.11793","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5128922922","display_name":"Alaa Yasser","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yasser, Alaa","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119466132","display_name":"Kittipat Phunjanna","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Phunjanna, Kittipat","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128940587","display_name":"Marcos Escudero Vi\u00f1olo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vi\u00f1olo, Marcos Escudero","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010674128","display_name":"Catarina Barata","orcid":"https://orcid.org/0000-0002-6478-8007"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barata, Catarina","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5104384396","display_name":"Jenny Benois-Pineau","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Benois-Pineau, Jenny","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.7531999945640564,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.7531999945640564,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.08749999850988388,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.01600000075995922,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/interpretability","display_name":"Interpretability","score":0.7106999754905701},{"id":"https://openalex.org/keywords/discriminative-model","display_name":"Discriminative model","score":0.551800012588501},{"id":"https://openalex.org/keywords/encoder","display_name":"Encoder","score":0.46639999747276306},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.4332999885082245},{"id":"https://openalex.org/keywords/gender-bias","display_name":"Gender bias","score":0.41530001163482666},{"id":"https://openalex.org/keywords/audit","display_name":"Audit","score":0.37040001153945923},{"id":"https://openalex.org/keywords/probabilistic-logic","display_name":"Probabilistic logic","score":0.35839998722076416}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.7106999754905701},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6302000284194946},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.551800012588501},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5333999991416931},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.46639999747276306},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4332999885082245},{"id":"https://openalex.org/C2983427547","wikidata":"https://www.wikidata.org/wiki/Q93200","display_name":"Gender bias","level":2,"score":0.41530001163482666},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.4113999903202057},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.37040001153945923},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3675000071525574},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.35839998722076416},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.33180001378059387},{"id":"https://openalex.org/C5339829","wikidata":"https://www.wikidata.org/wiki/Q1425977","display_name":"Machine vision","level":2,"score":0.3138999938964844},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2969000041484833},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.29109999537467957},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.28769999742507935},{"id":"https://openalex.org/C142853389","wikidata":"https://www.wikidata.org/wiki/Q744778","display_name":"Association (psychology)","level":2,"score":0.2827000021934509},{"id":"https://openalex.org/C148417208","wikidata":"https://www.wikidata.org/wiki/Q4825882","display_name":"Authentication (law)","level":2,"score":0.25600001215934753},{"id":"https://openalex.org/C739882","wikidata":"https://www.wikidata.org/wiki/Q3560506","display_name":"Anomaly detection","level":2,"score":0.2549999952316284},{"id":"https://openalex.org/C2776809875","wikidata":"https://www.wikidata.org/wiki/Q1375963","display_name":"Converse","level":2,"score":0.2529999911785126}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.11793","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.11793","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,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.11793","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.11793","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,"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"},"sustainable_development_goals":[{"score":0.5248388051986694,"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Standard":[0],"fairness":[1,25],"audits":[2],"of":[3,47,72,130,199],"foundation":[4],"models":[5],"quantify":[6],"that":[7,27,110,142,167,185,196],"a":[8,23,55,105],"model":[9],"is":[10,113,170,189],"biased,":[11],"but":[12,158],"not":[13],"where":[14],"inside":[15],"the":[16,18,45,64,73,84,116,123,128,131,134,147,152,197],"network":[17],"bias":[19,43,94,169,176,187],"resides.":[20],"We":[21],"propose":[22],"mechanistic":[24],"audit":[26],"combines":[28],"projected":[29],"residual-stream":[30],"decomposition,":[31],"zero-shot":[32],"Concept":[33],"Activation":[34],"Vectors,":[35],"and":[36,79,138,162,195],"bias-augmented":[37],"TextSpan":[38],"analysis":[39,140],"to":[40,63,115,127],"locate":[41],"demographic":[42],"at":[44],"level":[46],"individual":[48],"attention":[49],"heads":[50,89],"in":[51,122,133,177],"vision":[52,193],"transformers.":[53],"As":[54],"feasibility":[56],"case":[57],"study,":[58],"we":[59],"apply":[60],"this":[61,111,178],"pipeline":[62,85,154],"CLIP":[65,209],"ViT-L-14":[66],"encoder":[67],"on":[68],"42":[69],"profession":[70],"classes":[71],"FACET":[74],"benchmark,":[75],"auditing":[76],"both":[77],"gender":[78,175],"age":[80,168],"bias.":[81],"For":[82,150],"gender,":[83],"identifies":[86,155],"four":[87],"terminal-layer":[88],"whose":[90],"ablation":[91,159],"reduces":[92],"global":[93],"(Cramer's":[95],"V:":[96],"0.381":[97],"->":[98],"0.362)":[99],"while":[100],"marginally":[101],"improving":[102],"accuracy":[103],"(+0.42%);":[104],"layer-matched":[106],"random":[107],"control":[108],"confirms":[109],"effect":[112],"specific":[114],"identified":[117],"heads.":[118],"A":[119],"single":[120],"head":[121],"final":[124],"layer":[125],"contributes":[126],"majority":[129],"reduction":[132],"most":[135],"stereotyped":[136],"classes,":[137],"class-level":[139],"shows":[141],"corrected":[143],"predictions":[144],"shift":[145],"toward":[146],"correct":[148],"occupation.":[149],"age,":[151],"same":[153],"candidate":[156],"heads,":[157],"produces":[160],"weaker":[161],"less":[163],"consistent":[164],"effects,":[165],"suggesting":[166],"encoded":[171],"more":[172],"diffusely":[173],"than":[174],"model.":[179],"These":[180],"results":[181],"provide":[182],"preliminary":[183],"evidence":[184],"head-level":[186],"localisation":[188],"feasible":[190],"for":[191],"discriminative":[192],"encoders":[194],"degree":[198],"localisability":[200],"may":[201],"vary":[202],"across":[203],"protected":[204],"attributes.":[205],"keywords:":[206],"Bias":[207],".":[208,210,213,216],"Mechanistic":[211],"Interpretability":[212],"Vision":[214],"Transformer":[215],"Fairness":[217]},"counts_by_year":[],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2026-03-14T00:00:00"}