{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6IUIISFQZ4OXWV7WKJD2GHQYOX","short_pith_number":"pith:6IUIISFQ","schema_version":"1.0","canonical_sha256":"f2288448b0cf1d7b57f65247a31e1875c79eff1b78e27d2574183cacc39273ae","source":{"kind":"arxiv","id":"2507.19962","version":1},"attestation_state":"computed","paper":{"title":"KLAAD: Refining Attention Mechanisms to Reduce Societal Bias in Generative Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongyoung Lee, Jaejin Lee, Seorin Kim","submitted_at":"2025-07-26T14:24:19Z","abstract_excerpt":"Large language models (LLMs) often exhibit societal biases in their outputs, prompting ethical concerns regarding fairness and harm. In this work, we propose KLAAD (KL-Attention Alignment Debiasing), an attention-based debiasing framework that implicitly aligns attention distributions between stereotypical and anti-stereotypical sentence pairs without directly modifying model weights. KLAAD introduces a composite training objective combining Cross-Entropy, KL divergence, and Triplet losses, guiding the model to consistently attend across biased and unbiased contexts while preserving fluency an"},"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":"2507.19962","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-26T14:24:19Z","cross_cats_sorted":[],"title_canon_sha256":"cfd438e26b9db169272e4df225af4e6afe7c4a96d2bd041cf1ba2d6c33cab7b9","abstract_canon_sha256":"2988ec31e19614b7a2bb82780a4dd07261ee9a186deba6799ea206ef7692fb0e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:45.611724Z","signature_b64":"+FoiXT3kMVJzEVfV/Z3Hk1TOGeMPs3YmEMsFrtOczHrurYAkO2tPn/APApKPjTGofGk7mME3aSMGv1/H8DeACQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f2288448b0cf1d7b57f65247a31e1875c79eff1b78e27d2574183cacc39273ae","last_reissued_at":"2026-07-05T11:43:45.611157Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:45.611157Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"KLAAD: Refining Attention Mechanisms to Reduce Societal Bias in Generative Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongyoung Lee, Jaejin Lee, Seorin Kim","submitted_at":"2025-07-26T14:24:19Z","abstract_excerpt":"Large language models (LLMs) often exhibit societal biases in their outputs, prompting ethical concerns regarding fairness and harm. In this work, we propose KLAAD (KL-Attention Alignment Debiasing), an attention-based debiasing framework that implicitly aligns attention distributions between stereotypical and anti-stereotypical sentence pairs without directly modifying model weights. KLAAD introduces a composite training objective combining Cross-Entropy, KL divergence, and Triplet losses, guiding the model to consistently attend across biased and unbiased contexts while preserving fluency an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.19962","kind":"arxiv","version":1},"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/2507.19962/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":"2507.19962","created_at":"2026-07-05T11:43:45.611228+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.19962v1","created_at":"2026-07-05T11:43:45.611228+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.19962","created_at":"2026-07-05T11:43:45.611228+00:00"},{"alias_kind":"pith_short_12","alias_value":"6IUIISFQZ4OX","created_at":"2026-07-05T11:43:45.611228+00:00"},{"alias_kind":"pith_short_16","alias_value":"6IUIISFQZ4OXWV7W","created_at":"2026-07-05T11:43:45.611228+00:00"},{"alias_kind":"pith_short_8","alias_value":"6IUIISFQ","created_at":"2026-07-05T11:43:45.611228+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX","json":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX.json","graph_json":"https://pith.science/api/pith-number/6IUIISFQZ4OXWV7WKJD2GHQYOX/graph.json","events_json":"https://pith.science/api/pith-number/6IUIISFQZ4OXWV7WKJD2GHQYOX/events.json","paper":"https://pith.science/paper/6IUIISFQ"},"agent_actions":{"view_html":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX","download_json":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX.json","view_paper":"https://pith.science/paper/6IUIISFQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.19962&json=true","fetch_graph":"https://pith.science/api/pith-number/6IUIISFQZ4OXWV7WKJD2GHQYOX/graph.json","fetch_events":"https://pith.science/api/pith-number/6IUIISFQZ4OXWV7WKJD2GHQYOX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX/action/storage_attestation","attest_author":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX/action/author_attestation","sign_citation":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX/action/citation_signature","submit_replication":"https://pith.science/pith/6IUIISFQZ4OXWV7WKJD2GHQYOX/action/replication_record"}},"created_at":"2026-07-05T11:43:45.611228+00:00","updated_at":"2026-07-05T11:43:45.611228+00:00"}