{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UBVH3FUIFZPIWTF25HBL7QVNTL","short_pith_number":"pith:UBVH3FUI","schema_version":"1.0","canonical_sha256":"a06a7d96882e5e8b4cbae9c2bfc2ad9aebc4b1bcb7ece0d90eb761c38836098c","source":{"kind":"arxiv","id":"2410.22086","version":3},"attestation_state":"computed","paper":{"title":"Unlearning as multi-task optimization: A normalized gradient difference approach with an adaptive learning rate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Anil Ramakrishna, Bhanukiran Vinzamuri, Kai-Wei Chang, Mingyi Hong, Volkan Cevher, Xiaomeng Jin, Zhiqi Bu","submitted_at":"2024-10-29T14:41:44Z","abstract_excerpt":"Machine unlearning has been used to remove unwanted knowledge acquired by large language models (LLMs). In this paper, we examine machine unlearning from an optimization perspective, framing it as a regularized multi-task optimization problem, where one task optimizes a forgetting objective and another optimizes the model performance. In particular, we introduce a normalized gradient difference (NGDiff) algorithm, enabling us to have better control over the trade-off between the objectives, while integrating a new, automatic learning rate scheduler. We provide a theoretical analysis and empiri"},"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":"2410.22086","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-29T14:41:44Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"49278e9acff5b7d5782476c65f6cfe93a48f960e29eefbdeb2110186f8d966e2","abstract_canon_sha256":"c06f4e1aa5372e3c6949e5f79927ed55815ca71a14f0d2efe6d2c553e07f13d7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:58:46.261594Z","signature_b64":"B1Sh9rwe76/3SkgqJxqrHjxWYCpIKTL8Lg32H102dUF8xuka1Hwwp3yTxrq1y/4/5fx2u3lf+7ZoYnNQDRVuAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a06a7d96882e5e8b4cbae9c2bfc2ad9aebc4b1bcb7ece0d90eb761c38836098c","last_reissued_at":"2026-07-05T10:58:46.261116Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:58:46.261116Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Unlearning as multi-task optimization: A normalized gradient difference approach with an adaptive learning rate","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Anil Ramakrishna, Bhanukiran Vinzamuri, Kai-Wei Chang, Mingyi Hong, Volkan Cevher, Xiaomeng Jin, Zhiqi Bu","submitted_at":"2024-10-29T14:41:44Z","abstract_excerpt":"Machine unlearning has been used to remove unwanted knowledge acquired by large language models (LLMs). In this paper, we examine machine unlearning from an optimization perspective, framing it as a regularized multi-task optimization problem, where one task optimizes a forgetting objective and another optimizes the model performance. In particular, we introduce a normalized gradient difference (NGDiff) algorithm, enabling us to have better control over the trade-off between the objectives, while integrating a new, automatic learning rate scheduler. We provide a theoretical analysis and empiri"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.22086","kind":"arxiv","version":3},"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/2410.22086/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":"2410.22086","created_at":"2026-07-05T10:58:46.261176+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.22086v3","created_at":"2026-07-05T10:58:46.261176+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.22086","created_at":"2026-07-05T10:58:46.261176+00:00"},{"alias_kind":"pith_short_12","alias_value":"UBVH3FUIFZPI","created_at":"2026-07-05T10:58:46.261176+00:00"},{"alias_kind":"pith_short_16","alias_value":"UBVH3FUIFZPIWTF2","created_at":"2026-07-05T10:58:46.261176+00:00"},{"alias_kind":"pith_short_8","alias_value":"UBVH3FUI","created_at":"2026-07-05T10:58:46.261176+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.22483","citing_title":"OFMU: Optimization-Driven Framework for Machine Unlearning","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL","json":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL.json","graph_json":"https://pith.science/api/pith-number/UBVH3FUIFZPIWTF25HBL7QVNTL/graph.json","events_json":"https://pith.science/api/pith-number/UBVH3FUIFZPIWTF25HBL7QVNTL/events.json","paper":"https://pith.science/paper/UBVH3FUI"},"agent_actions":{"view_html":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL","download_json":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL.json","view_paper":"https://pith.science/paper/UBVH3FUI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.22086&json=true","fetch_graph":"https://pith.science/api/pith-number/UBVH3FUIFZPIWTF25HBL7QVNTL/graph.json","fetch_events":"https://pith.science/api/pith-number/UBVH3FUIFZPIWTF25HBL7QVNTL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL/action/storage_attestation","attest_author":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL/action/author_attestation","sign_citation":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL/action/citation_signature","submit_replication":"https://pith.science/pith/UBVH3FUIFZPIWTF25HBL7QVNTL/action/replication_record"}},"created_at":"2026-07-05T10:58:46.261176+00:00","updated_at":"2026-07-05T10:58:46.261176+00:00"}