{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3DPCXC5S7NY4UTTM4R6KFWCNKG","short_pith_number":"pith:3DPCXC5S","schema_version":"1.0","canonical_sha256":"d8de2b8bb2fb71ca4e6ce47ca2d84d51bc495f88a0c7447c2b99518d7d0dd418","source":{"kind":"arxiv","id":"2410.13146","version":2},"attestation_state":"computed","paper":{"title":"debiaSAE: Benchmarking and Mitigating Vision-Language Model Bias","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Danielle Bitterman, Jackson Pond, John Osborne, Kuleen Sasse, Shan Chen","submitted_at":"2024-10-17T02:03:27Z","abstract_excerpt":"As Vision Language Models (VLMs) gain widespread use, their fairness remains under-explored. In this paper, we analyze demographic biases across five models and six datasets. We find that portrait datasets like UTKFace and CelebA are the best tools for bias detection, finding gaps in performance and fairness for both LLaVa and CLIP models. Scene-based datasets like PATA and VLStereoSet fail to be useful benchmarks for bias due to their text prompts allowing the model to guess the answer without a picture. As for pronoun-based datasets like VisoGender, we receive mixed signals as only some subs"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2410.13146","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-17T02:03:27Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"69b2c320435c650afdc503122261059cbc618d247a92957b3a3e27a5bb7024d5","abstract_canon_sha256":"adcd9197a793f464ca213ebfa1314a36964733330b965b22264846a67888d4d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:41:16.063164Z","signature_b64":"Ndk+mXfHVLTMy1pPXZ80bdu45ZP9/vI0WYBYvNknCvdHw2EFxoq+i7odtCsQ3+jqNQvtoOXQbb8dm/3sZekaCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d8de2b8bb2fb71ca4e6ce47ca2d84d51bc495f88a0c7447c2b99518d7d0dd418","last_reissued_at":"2026-07-05T10:41:16.062698Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:41:16.062698Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"debiaSAE: Benchmarking and Mitigating Vision-Language Model Bias","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Danielle Bitterman, Jackson Pond, John Osborne, Kuleen Sasse, Shan Chen","submitted_at":"2024-10-17T02:03:27Z","abstract_excerpt":"As Vision Language Models (VLMs) gain widespread use, their fairness remains under-explored. In this paper, we analyze demographic biases across five models and six datasets. We find that portrait datasets like UTKFace and CelebA are the best tools for bias detection, finding gaps in performance and fairness for both LLaVa and CLIP models. Scene-based datasets like PATA and VLStereoSet fail to be useful benchmarks for bias due to their text prompts allowing the model to guess the answer without a picture. As for pronoun-based datasets like VisoGender, we receive mixed signals as only some subs"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.13146","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2410.13146/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2410.13146","created_at":"2026-07-05T10:41:16.062755+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.13146v2","created_at":"2026-07-05T10:41:16.062755+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.13146","created_at":"2026-07-05T10:41:16.062755+00:00"},{"alias_kind":"pith_short_12","alias_value":"3DPCXC5S7NY4","created_at":"2026-07-05T10:41:16.062755+00:00"},{"alias_kind":"pith_short_16","alias_value":"3DPCXC5S7NY4UTTM","created_at":"2026-07-05T10:41:16.062755+00:00"},{"alias_kind":"pith_short_8","alias_value":"3DPCXC5S","created_at":"2026-07-05T10:41:16.062755+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.04568","citing_title":"Feedback-Driven Vision-Language Alignment with Minimal Human Supervision","ref_index":75,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG","json":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG.json","graph_json":"https://pith.science/api/pith-number/3DPCXC5S7NY4UTTM4R6KFWCNKG/graph.json","events_json":"https://pith.science/api/pith-number/3DPCXC5S7NY4UTTM4R6KFWCNKG/events.json","paper":"https://pith.science/paper/3DPCXC5S"},"agent_actions":{"view_html":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG","download_json":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG.json","view_paper":"https://pith.science/paper/3DPCXC5S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.13146&json=true","fetch_graph":"https://pith.science/api/pith-number/3DPCXC5S7NY4UTTM4R6KFWCNKG/graph.json","fetch_events":"https://pith.science/api/pith-number/3DPCXC5S7NY4UTTM4R6KFWCNKG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG/action/storage_attestation","attest_author":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG/action/author_attestation","sign_citation":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG/action/citation_signature","submit_replication":"https://pith.science/pith/3DPCXC5S7NY4UTTM4R6KFWCNKG/action/replication_record"}},"created_at":"2026-07-05T10:41:16.062755+00:00","updated_at":"2026-07-05T10:41:16.062755+00:00"}