{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:TZFIMW5NGAQI5YAOJJ4HUE43G2","short_pith_number":"pith:TZFIMW5N","schema_version":"1.0","canonical_sha256":"9e4a865bad30208ee00e4a787a139b36b1be48c5a3e54aac71ee396acde6206d","source":{"kind":"arxiv","id":"2210.01504","version":2},"attestation_state":"computed","paper":{"title":"Knowledge Unlearning for Mitigating Privacy Risks in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongkeun Yoon, Joel Jang, Lajanugen Logeswaran, Minjoon Seo, Moontae Lee, Sohee Yang, Sungmin Cha","submitted_at":"2022-10-04T10:18:11Z","abstract_excerpt":"Pretrained Language Models (LMs) memorize a vast amount of knowledge during initial pretraining, including information that may violate the privacy of personal lives and identities. Previous work addressing privacy issues for language models has mostly focused on data preprocessing and differential privacy methods, both requiring re-training the underlying LM. We propose knowledge unlearning as an alternative method to reduce privacy risks for LMs post hoc. We show that simply performing gradient ascent on target token sequences is effective at forgetting them with little to no degradation of "},"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":"2210.01504","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-10-04T10:18:11Z","cross_cats_sorted":[],"title_canon_sha256":"c3f7f4a8829d805ed9dd43b098a4b512d651a864b61cbf4401a065c8ec2d4b50","abstract_canon_sha256":"f15ffaecdf5dc1e3bf398346e3ac7d2f5ae6f284d8e741812fb39c085912579e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:26:09.701225Z","signature_b64":"xBuHeuLpOFriput3mr8qGolmklTp3+gRq+zv/vPZ+II16tTQ7VsR0Qyk9j+EbnU0VsA6n4T4G76u/ZaiuiGlBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9e4a865bad30208ee00e4a787a139b36b1be48c5a3e54aac71ee396acde6206d","last_reissued_at":"2026-07-05T05:26:09.700707Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:26:09.700707Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Knowledge Unlearning for Mitigating Privacy Risks in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongkeun Yoon, Joel Jang, Lajanugen Logeswaran, Minjoon Seo, Moontae Lee, Sohee Yang, Sungmin Cha","submitted_at":"2022-10-04T10:18:11Z","abstract_excerpt":"Pretrained Language Models (LMs) memorize a vast amount of knowledge during initial pretraining, including information that may violate the privacy of personal lives and identities. Previous work addressing privacy issues for language models has mostly focused on data preprocessing and differential privacy methods, both requiring re-training the underlying LM. We propose knowledge unlearning as an alternative method to reduce privacy risks for LMs post hoc. We show that simply performing gradient ascent on target token sequences is effective at forgetting them with little to no degradation of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2210.01504","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/2210.01504/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":"2210.01504","created_at":"2026-07-05T05:26:09.700774+00:00"},{"alias_kind":"arxiv_version","alias_value":"2210.01504v2","created_at":"2026-07-05T05:26:09.700774+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2210.01504","created_at":"2026-07-05T05:26:09.700774+00:00"},{"alias_kind":"pith_short_12","alias_value":"TZFIMW5NGAQI","created_at":"2026-07-05T05:26:09.700774+00:00"},{"alias_kind":"pith_short_16","alias_value":"TZFIMW5NGAQI5YAO","created_at":"2026-07-05T05:26:09.700774+00:00"},{"alias_kind":"pith_short_8","alias_value":"TZFIMW5N","created_at":"2026-07-05T05:26:09.700774+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09132","citing_title":"Vision Language Model Helps Private Information De-Identification in Vision Data","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01129","citing_title":"Revisiting Privacy Leakage in Machine Unlearning: Membership Inference Beyond the Forgotten Set","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2601.06163","citing_title":"Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron Masking","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2506.20941","citing_title":"Revisiting the Past: Data Unlearning with Model State History","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2404.05868","citing_title":"Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12469","citing_title":"Sparse Concept Anchoring for Interpretable and Controllable Neural Representations","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12469","citing_title":"Sparse Concept Anchoring for Interpretable and Controllable Neural Representations","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2401.06121","citing_title":"TOFU: A Task of Fictitious Unlearning for LLMs","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14404","citing_title":"Knowledge Beyond Language: Bridging the Gap in Multilingual Machine Unlearning Evaluation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04030","citing_title":"Jellyfish: Zero-Shot Federated Unlearning Scheme with Knowledge Disentanglement","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01129","citing_title":"Revisiting Privacy Leakage in Machine Unlearning: Membership Inference Beyond the Forgotten Set","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00364","citing_title":"Unlearning What Matters: Token-Level Attribution for Precise Language Model Unlearning","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19089","citing_title":"Towards Scalable Lifelong Knowledge Editing with Selective Knowledge Suppression","ref_index":105,"is_internal_anchor":false},{"citing_arxiv_id":"2404.18930","citing_title":"Hallucination of Multimodal Large Language Models: A Survey","ref_index":76,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2","json":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2.json","graph_json":"https://pith.science/api/pith-number/TZFIMW5NGAQI5YAOJJ4HUE43G2/graph.json","events_json":"https://pith.science/api/pith-number/TZFIMW5NGAQI5YAOJJ4HUE43G2/events.json","paper":"https://pith.science/paper/TZFIMW5N"},"agent_actions":{"view_html":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2","download_json":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2.json","view_paper":"https://pith.science/paper/TZFIMW5N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2210.01504&json=true","fetch_graph":"https://pith.science/api/pith-number/TZFIMW5NGAQI5YAOJJ4HUE43G2/graph.json","fetch_events":"https://pith.science/api/pith-number/TZFIMW5NGAQI5YAOJJ4HUE43G2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2/action/storage_attestation","attest_author":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2/action/author_attestation","sign_citation":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2/action/citation_signature","submit_replication":"https://pith.science/pith/TZFIMW5NGAQI5YAOJJ4HUE43G2/action/replication_record"}},"created_at":"2026-07-05T05:26:09.700774+00:00","updated_at":"2026-07-05T05:26:09.700774+00:00"}