{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DUUMFKS3MXJP5RFCTPHS2DGPFB","short_pith_number":"pith:DUUMFKS3","schema_version":"1.0","canonical_sha256":"1d28c2aa5b65d2fec4a29bcf2d0ccf284a2b377cd9b0156afe7e5eb7ff22656b","source":{"kind":"arxiv","id":"2505.24428","version":1},"attestation_state":"computed","paper":{"title":"Model Unlearning via Sparse Autoencoder Subspace Guided Projections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Benyou Wang, Difan Zou, Xu Wang, Yan Hu, Zihao Li","submitted_at":"2025-05-30T10:07:52Z","abstract_excerpt":"Large language models (LLMs) store vast amounts of information, making them powerful yet raising privacy and safety concerns when selective knowledge removal is required. Existing unlearning strategies, ranging from gradient-based fine-tuning and model editing to sparse autoencoder (SAE) steering, either lack interpretability or fail to provide a robust defense against adversarial prompts. We propose SAE-Guided Subspace Projection Unlearning (SSPU), a novel framework that leverages SAE features to drive targeted updates in the model's parameter space, enabling precise, interpretable, and robus"},"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":"2505.24428","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-30T10:07:52Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e7c3bc6d1acf0a5f8de4f9ac48ec4995b17e18638e037301dc4a98f76c2ef66e","abstract_canon_sha256":"cdfb1d4bacb891d0760250c6834fbfe6c4d422f9bedf3a0b27b9b061f7ce4a91"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:49.555015Z","signature_b64":"R6bumi397xxUENFd1lrRYAAmHCSZ+vWNz7MDzNYZN4ssM0XS/1eozeipDRDO3fMuQ/RxsgN/OYY/minEYBIkAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1d28c2aa5b65d2fec4a29bcf2d0ccf284a2b377cd9b0156afe7e5eb7ff22656b","last_reissued_at":"2026-07-05T11:12:49.554514Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:49.554514Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Model Unlearning via Sparse Autoencoder Subspace Guided Projections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Benyou Wang, Difan Zou, Xu Wang, Yan Hu, Zihao Li","submitted_at":"2025-05-30T10:07:52Z","abstract_excerpt":"Large language models (LLMs) store vast amounts of information, making them powerful yet raising privacy and safety concerns when selective knowledge removal is required. Existing unlearning strategies, ranging from gradient-based fine-tuning and model editing to sparse autoencoder (SAE) steering, either lack interpretability or fail to provide a robust defense against adversarial prompts. We propose SAE-Guided Subspace Projection Unlearning (SSPU), a novel framework that leverages SAE features to drive targeted updates in the model's parameter space, enabling precise, interpretable, and robus"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.24428","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/2505.24428/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":"2505.24428","created_at":"2026-07-05T11:12:49.554580+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.24428v1","created_at":"2026-07-05T11:12:49.554580+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.24428","created_at":"2026-07-05T11:12:49.554580+00:00"},{"alias_kind":"pith_short_12","alias_value":"DUUMFKS3MXJP","created_at":"2026-07-05T11:12:49.554580+00:00"},{"alias_kind":"pith_short_16","alias_value":"DUUMFKS3MXJP5RFC","created_at":"2026-07-05T11:12:49.554580+00:00"},{"alias_kind":"pith_short_8","alias_value":"DUUMFKS3","created_at":"2026-07-05T11:12:49.554580+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/DUUMFKS3MXJP5RFCTPHS2DGPFB","json":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB.json","graph_json":"https://pith.science/api/pith-number/DUUMFKS3MXJP5RFCTPHS2DGPFB/graph.json","events_json":"https://pith.science/api/pith-number/DUUMFKS3MXJP5RFCTPHS2DGPFB/events.json","paper":"https://pith.science/paper/DUUMFKS3"},"agent_actions":{"view_html":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB","download_json":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB.json","view_paper":"https://pith.science/paper/DUUMFKS3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.24428&json=true","fetch_graph":"https://pith.science/api/pith-number/DUUMFKS3MXJP5RFCTPHS2DGPFB/graph.json","fetch_events":"https://pith.science/api/pith-number/DUUMFKS3MXJP5RFCTPHS2DGPFB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB/action/storage_attestation","attest_author":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB/action/author_attestation","sign_citation":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB/action/citation_signature","submit_replication":"https://pith.science/pith/DUUMFKS3MXJP5RFCTPHS2DGPFB/action/replication_record"}},"created_at":"2026-07-05T11:12:49.554580+00:00","updated_at":"2026-07-05T11:12:49.554580+00:00"}