{"paper":{"title":"Beyond Performance Disparities: A Three-Level Audit of Representational Harm in CelebA","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Cultural double standards of beauty and aging in CelebA labels, features, and attention create representational harms for women and older men.","cross_cats":["cs.CV"],"primary_cat":"cs.CY","authors_text":"Sieun Park, Yuanmo He","submitted_at":"2026-05-14T18:25:17Z","abstract_excerpt":"Large-scale facial datasets like CelebA are widely used in computer vision, yet the cultural biases embedded in their labels remain underexplored. Fairness research has distinguished representational from allocational harms, but audits of computer vision datasets have mostly examined categorical labels, leaving open how such harms appear in learned features and model attention. This paper examines CelebA at three levels: dataset structure, learned feature weights, and spatial attention, focusing on how gendered double standards of ageing and beauty are encoded in the data and reproduced in mod"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Cultural double standards thus pass from media representation into dataset labels, feature weights, and model attention, producing two representational harms: hyper-scrutiny of women under a narrow evaluative template, and exclusion of older men from the scheme entirely.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The premise that hierarchical clustering of the 39 attributes, SHAP effects, and Grad-CAM attention maps can be directly interpreted as evidence of cultural archetypes and double standards without substantial confounding from dataset collection processes or model architecture choices. This enters in the alignment of clusters with performative femininity and professional masculinity and in the cultural reading of attention shifts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CelebA encodes gendered double standards of ageing and beauty that produce hyper-scrutiny of women and categorical exclusion of older men across labels, feature weights, and spatial attention.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Cultural double standards of beauty and aging in CelebA labels, features, and attention create representational harms for women and older men.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1fd1a3d88cd8568229f9a56e87fbd80bf67bebb3dd1fe209936b96bb97c85507"},"source":{"id":"2605.15312","kind":"arxiv","version":1},"verdict":{"id":"64b8fc5e-0f2f-470f-8333-3815511cb37a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T16:12:35.283828Z","strongest_claim":"Cultural double standards thus pass from media representation into dataset labels, feature weights, and model attention, producing two representational harms: hyper-scrutiny of women under a narrow evaluative template, and exclusion of older men from the scheme entirely.","one_line_summary":"CelebA encodes gendered double standards of ageing and beauty that produce hyper-scrutiny of women and categorical exclusion of older men across labels, feature weights, and spatial attention.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The premise that hierarchical clustering of the 39 attributes, SHAP effects, and Grad-CAM attention maps can be directly interpreted as evidence of cultural archetypes and double standards without substantial confounding from dataset collection processes or model architecture choices. This enters in the alignment of clusters with performative femininity and professional masculinity and in the cultural reading of attention shifts.","pith_extraction_headline":"Cultural double standards of beauty and aging in CelebA labels, features, and attention create representational harms for women and older men."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15312/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T16:31:18.326311Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T16:25:54.610334Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:41:54.218049Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.773652Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"88b1bb1523007bcf7a04a8869c123dc89dce97780935436447a15db1a5eed456"},"references":{"count":94,"sample":[{"doi":"","year":1976,"title":"1976 , publisher =","work_id":"494cd651-35d2-41a5-a7f3-94afdd63500a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1979,"title":"1979 , publisher =","work_id":"8032c08f-2cea-4354-9c15-44123970e87f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1977,"title":"1977 , publisher =","work_id":"114807f1-fba9-4d17-806f-45f5cf711d18","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1991,"title":"1991 , publisher =","work_id":"0f4f28d1-4a5b-41e1-a9f3-38599f32b5b3","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2007,"title":"Hunter , title =","work_id":"22c9198e-fd79-47a4-8a85-2c4af8f539e8","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":94,"snapshot_sha256":"b885ee89aea04a218e765590b83c09bc4e8271cb7a6f549a8d8a117ddd7352cb","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}