{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:2SOGRFIV2DF6XNLRFT22CR5G7E","short_pith_number":"pith:2SOGRFIV","canonical_record":{"source":{"id":"2508.06467","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T17:15:32Z","cross_cats_sorted":[],"title_canon_sha256":"f610b6e9720dacd766a9b2ec2b154f841ec2986d80b5911cb49cd8392a31063d","abstract_canon_sha256":"a4afd24b36c8bffdaa96d51c04eec0e5eb4120366de1b507fb99ebbfa86cefd9"},"schema_version":"1.0"},"canonical_sha256":"d49c689515d0cbebb5712cf5a147a6f93dfca853c78a0d004da556ef237561ad","source":{"kind":"arxiv","id":"2508.06467","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2508.06467","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"arxiv_version","alias_value":"2508.06467v1","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.06467","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_12","alias_value":"2SOGRFIV2DF6","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_16","alias_value":"2SOGRFIV2DF6XNLR","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_8","alias_value":"2SOGRFIV","created_at":"2026-07-05T11:50:56Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:2SOGRFIV2DF6XNLRFT22CR5G7E","target":"record","payload":{"canonical_record":{"source":{"id":"2508.06467","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T17:15:32Z","cross_cats_sorted":[],"title_canon_sha256":"f610b6e9720dacd766a9b2ec2b154f841ec2986d80b5911cb49cd8392a31063d","abstract_canon_sha256":"a4afd24b36c8bffdaa96d51c04eec0e5eb4120366de1b507fb99ebbfa86cefd9"},"schema_version":"1.0"},"canonical_sha256":"d49c689515d0cbebb5712cf5a147a6f93dfca853c78a0d004da556ef237561ad","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:56.833040Z","signature_b64":"4JKEddHMC0TNEUiPHcPBItkrL8/5VmINSrgHrtAXMU+kDa+HNVtly0LmkphwLbJ23imApBNJQo6pjLuyuzJkBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d49c689515d0cbebb5712cf5a147a6f93dfca853c78a0d004da556ef237561ad","last_reissued_at":"2026-07-05T11:50:56.832562Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:56.832562Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2508.06467","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:50:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"xmOYhFrGuap3rS9+dZ5fGyaT/bokxZqWNjM0y6ca/YZGNAY8d9i3RnbeAaRKeAe5WsVCf+UAWNyU7OKCOjIgCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-12T08:39:20.122393Z"},"content_sha256":"8abfa62dda9e0641c2b1c4cc7a9258bc2c3b46be545967b9cc857b5587ef635d","schema_version":"1.0","event_id":"sha256:8abfa62dda9e0641c2b1c4cc7a9258bc2c3b46be545967b9cc857b5587ef635d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:2SOGRFIV2DF6XNLRFT22CR5G7E","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"LLM Unlearning using Gradient Ratio-Based Influence Estimation and Noise Injection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ameya Anjarlekar, Sandeep Pombra","submitted_at":"2025-08-08T17:15:32Z","abstract_excerpt":"The growing legal and ethical scrutiny of large language models (LLMs) necessitates effective machine unlearning, particularly for sensitive or unauthorized data. Existing empirical methods often yield incomplete forgetting or unintended degradation of unrelated knowledge due to poor localization. In this work, we propose GRIN: a modular and targeted framework for LLM unlearning. GRIN introduces a novel gradient-ratio-based metric to identify parameters most responsible for memorizing forget data. We then perform selective noise injection into these parameters prior to fine-tuning, which impro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.06467","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/2508.06467/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:50:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"hmdY2DBkx2B6BFA0zynKaJ2nYkmKws197Q48cox9wyNFZltZSUAKhLQqiwOeSRBEEubRZxM9/Fdj4QDljjQVDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-12T08:39:20.122766Z"},"content_sha256":"1db057246fb2d93c2ecb4563e3f1ef16425fe6e157bcef3811bf8d14e024648b","schema_version":"1.0","event_id":"sha256:1db057246fb2d93c2ecb4563e3f1ef16425fe6e157bcef3811bf8d14e024648b"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/bundle.json","state_url":"https://pith.science/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-12T08:39:20Z","links":{"resolver":"https://pith.science/pith/2SOGRFIV2DF6XNLRFT22CR5G7E","bundle":"https://pith.science/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/bundle.json","state":"https://pith.science/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/state.json","well_known_bundle":"https://pith.science/.well-known/pith/2SOGRFIV2DF6XNLRFT22CR5G7E/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:2SOGRFIV2DF6XNLRFT22CR5G7E","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a4afd24b36c8bffdaa96d51c04eec0e5eb4120366de1b507fb99ebbfa86cefd9","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T17:15:32Z","title_canon_sha256":"f610b6e9720dacd766a9b2ec2b154f841ec2986d80b5911cb49cd8392a31063d"},"schema_version":"1.0","source":{"id":"2508.06467","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2508.06467","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"arxiv_version","alias_value":"2508.06467v1","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.06467","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_12","alias_value":"2SOGRFIV2DF6","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_16","alias_value":"2SOGRFIV2DF6XNLR","created_at":"2026-07-05T11:50:56Z"},{"alias_kind":"pith_short_8","alias_value":"2SOGRFIV","created_at":"2026-07-05T11:50:56Z"}],"graph_snapshots":[{"event_id":"sha256:1db057246fb2d93c2ecb4563e3f1ef16425fe6e157bcef3811bf8d14e024648b","target":"graph","created_at":"2026-07-05T11:50:56Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2508.06467/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"The growing legal and ethical scrutiny of large language models (LLMs) necessitates effective machine unlearning, particularly for sensitive or unauthorized data. Existing empirical methods often yield incomplete forgetting or unintended degradation of unrelated knowledge due to poor localization. In this work, we propose GRIN: a modular and targeted framework for LLM unlearning. GRIN introduces a novel gradient-ratio-based metric to identify parameters most responsible for memorizing forget data. We then perform selective noise injection into these parameters prior to fine-tuning, which impro","authors_text":"Ameya Anjarlekar, Sandeep Pombra","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T17:15:32Z","title":"LLM Unlearning using Gradient Ratio-Based Influence Estimation and Noise Injection"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.06467","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:8abfa62dda9e0641c2b1c4cc7a9258bc2c3b46be545967b9cc857b5587ef635d","target":"record","created_at":"2026-07-05T11:50:56Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"a4afd24b36c8bffdaa96d51c04eec0e5eb4120366de1b507fb99ebbfa86cefd9","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-08T17:15:32Z","title_canon_sha256":"f610b6e9720dacd766a9b2ec2b154f841ec2986d80b5911cb49cd8392a31063d"},"schema_version":"1.0","source":{"id":"2508.06467","kind":"arxiv","version":1}},"canonical_sha256":"d49c689515d0cbebb5712cf5a147a6f93dfca853c78a0d004da556ef237561ad","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d49c689515d0cbebb5712cf5a147a6f93dfca853c78a0d004da556ef237561ad","first_computed_at":"2026-07-05T11:50:56.832562Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:50:56.832562Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"4JKEddHMC0TNEUiPHcPBItkrL8/5VmINSrgHrtAXMU+kDa+HNVtly0LmkphwLbJ23imApBNJQo6pjLuyuzJkBQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:50:56.833040Z","signed_message":"canonical_sha256_bytes"},"source_id":"2508.06467","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8abfa62dda9e0641c2b1c4cc7a9258bc2c3b46be545967b9cc857b5587ef635d","sha256:1db057246fb2d93c2ecb4563e3f1ef16425fe6e157bcef3811bf8d14e024648b"],"state_sha256":"110026011ab14452aa3b73e3906bc3aeeb5516fb09f7cbd4553b92c9d4743d37"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"YElR6cmcM5+pj2doG2izPmdoLGuhw3d1Id9ffU38imCjq8PY8F+CvfDC8y+KynXYL0rWDHAJr4+lSE8p2QBrDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-12T08:39:20.124841Z","bundle_sha256":"62bcaeefec012ae89938829d033822ee5a60f1f321f2e6acd6e6224b54b6590b"}}