{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YQLGT5C3QAJWAWP452AKRJSEYT","short_pith_number":"pith:YQLGT5C3","schema_version":"1.0","canonical_sha256":"c41669f45b80136059fcee80a8a644c4f0449ab110cf1bb6e93444ad6b7f7159","source":{"kind":"arxiv","id":"2306.03819","version":4},"attestation_state":"computed","paper":{"title":"LEACE: Perfect linear concept erasure in closed form","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"David Schneider-Joseph, Edward Raff, Nora Belrose, Ryan Cotterell, Shauli Ravfogel, Stella Biderman","submitted_at":"2023-06-06T16:07:24Z","abstract_excerpt":"Concept erasure aims to remove specified features from an embedding. It can improve fairness (e.g. preventing a classifier from using gender or race) and interpretability (e.g. removing a concept to observe changes in model behavior). We introduce LEAst-squares Concept Erasure (LEACE), a closed-form method which provably prevents all linear classifiers from detecting a concept while changing the embedding as little as possible, as measured by a broad class of norms. We apply LEACE to large language models with a novel procedure called \"concept scrubbing,\" which erases target concept informatio"},"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":"2306.03819","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-06T16:07:24Z","cross_cats_sorted":["cs.CL","cs.CY"],"title_canon_sha256":"f3bc83deadbcea282ea1f34eac7bdac3582237e294eb7e42014afe8f9e199986","abstract_canon_sha256":"833b7c0c5bf3ea8d74a772129a56773ff8fe9ad542bb89ff665ef80ee8c3a5b4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:43:33.549566Z","signature_b64":"uNxAuqMTtoPj69casLgODYOnGV9w23gOIuwY8hLMRfnd8HXNCbN0RG21xW8gTljA7ltrnhroaI/CQotdjhsICA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c41669f45b80136059fcee80a8a644c4f0449ab110cf1bb6e93444ad6b7f7159","last_reissued_at":"2026-07-05T10:43:33.548987Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:43:33.548987Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LEACE: Perfect linear concept erasure in closed form","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"David Schneider-Joseph, Edward Raff, Nora Belrose, Ryan Cotterell, Shauli Ravfogel, Stella Biderman","submitted_at":"2023-06-06T16:07:24Z","abstract_excerpt":"Concept erasure aims to remove specified features from an embedding. It can improve fairness (e.g. preventing a classifier from using gender or race) and interpretability (e.g. removing a concept to observe changes in model behavior). We introduce LEAst-squares Concept Erasure (LEACE), a closed-form method which provably prevents all linear classifiers from detecting a concept while changing the embedding as little as possible, as measured by a broad class of norms. We apply LEACE to large language models with a novel procedure called \"concept scrubbing,\" which erases target concept informatio"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.03819","kind":"arxiv","version":4},"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/2306.03819/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":"2306.03819","created_at":"2026-07-05T10:43:33.549067+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.03819v4","created_at":"2026-07-05T10:43:33.549067+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.03819","created_at":"2026-07-05T10:43:33.549067+00:00"},{"alias_kind":"pith_short_12","alias_value":"YQLGT5C3QAJW","created_at":"2026-07-05T10:43:33.549067+00:00"},{"alias_kind":"pith_short_16","alias_value":"YQLGT5C3QAJWAWP4","created_at":"2026-07-05T10:43:33.549067+00:00"},{"alias_kind":"pith_short_8","alias_value":"YQLGT5C3","created_at":"2026-07-05T10:43:33.549067+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.24583","citing_title":"Measuring Alignment-Induced Activation Shifts Correctly: A Template-Controlled Difference-in-Differences Protocol","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2306.12001","citing_title":"An Overview of Catastrophic AI Risks","ref_index":140,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":87,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08837","citing_title":"The Grounding Gap: How LLMs Anchor the Meaning of Abstract Concepts Differently from Humans","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05715","citing_title":"Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01848","citing_title":"Disentangled Anatomy-Disease Diffusion (DADD) for Controllable Ulcerative Colitis Progression Synthesis","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT","json":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT.json","graph_json":"https://pith.science/api/pith-number/YQLGT5C3QAJWAWP452AKRJSEYT/graph.json","events_json":"https://pith.science/api/pith-number/YQLGT5C3QAJWAWP452AKRJSEYT/events.json","paper":"https://pith.science/paper/YQLGT5C3"},"agent_actions":{"view_html":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT","download_json":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT.json","view_paper":"https://pith.science/paper/YQLGT5C3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.03819&json=true","fetch_graph":"https://pith.science/api/pith-number/YQLGT5C3QAJWAWP452AKRJSEYT/graph.json","fetch_events":"https://pith.science/api/pith-number/YQLGT5C3QAJWAWP452AKRJSEYT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT/action/storage_attestation","attest_author":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT/action/author_attestation","sign_citation":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT/action/citation_signature","submit_replication":"https://pith.science/pith/YQLGT5C3QAJWAWP452AKRJSEYT/action/replication_record"}},"created_at":"2026-07-05T10:43:33.549067+00:00","updated_at":"2026-07-05T10:43:33.549067+00:00"}