{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ETEOFKW2KPQWMIGEHOP3FMWGB7","short_pith_number":"pith:ETEOFKW2","schema_version":"1.0","canonical_sha256":"24c8e2aada53e16620c43b9fb2b2c60fc75ad30fa312ff9b08546b8aa0622018","source":{"kind":"arxiv","id":"2410.04332","version":2},"attestation_state":"computed","paper":{"title":"Gradient Routing: Masking Gradients to Localize Computation in Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexander Matt Turner, Alex Cloud, Ev\\v{z}en Wybitul, Jacob Goldman-Wetzler, Joseph Miller","submitted_at":"2024-10-06T02:43:49Z","abstract_excerpt":"Neural networks are trained primarily based on their inputs and outputs, without regard for their internal mechanisms. These neglected mechanisms determine properties that are critical for safety, like (i) transparency; (ii) the absence of sensitive information or harmful capabilities; and (iii) reliable generalization of goals beyond the training distribution. To address this shortcoming, we introduce gradient routing, a training method that isolates capabilities to specific subregions of a neural network. Gradient routing applies data-dependent, weighted masks to gradients during backpropaga"},"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":"2410.04332","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-06T02:43:49Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"36c67b9a4df4e82af3c3fb06c8e4ae5b88d38bb3b9267bfd499f4878d1b3e76f","abstract_canon_sha256":"52c6f49cfaf636ae875628e3bbbbc064927c6b35a356123d3d040238e6c7dc57"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:42:15.156733Z","signature_b64":"3C1SDHhlpBQRdWRi2mzIzhT9tRQ50p8WJceJVaFcOd4PTjKsg/dHdhJ+z4ZxLDbtloJaesv+/gpSAx9eheIXCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24c8e2aada53e16620c43b9fb2b2c60fc75ad30fa312ff9b08546b8aa0622018","last_reissued_at":"2026-07-05T09:42:15.156261Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:42:15.156261Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Routing: Masking Gradients to Localize Computation in Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexander Matt Turner, Alex Cloud, Ev\\v{z}en Wybitul, Jacob Goldman-Wetzler, Joseph Miller","submitted_at":"2024-10-06T02:43:49Z","abstract_excerpt":"Neural networks are trained primarily based on their inputs and outputs, without regard for their internal mechanisms. These neglected mechanisms determine properties that are critical for safety, like (i) transparency; (ii) the absence of sensitive information or harmful capabilities; and (iii) reliable generalization of goals beyond the training distribution. To address this shortcoming, we introduce gradient routing, a training method that isolates capabilities to specific subregions of a neural network. Gradient routing applies data-dependent, weighted masks to gradients during backpropaga"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.04332","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/2410.04332/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":"2410.04332","created_at":"2026-07-05T09:42:15.156327+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.04332v2","created_at":"2026-07-05T09:42:15.156327+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.04332","created_at":"2026-07-05T09:42:15.156327+00:00"},{"alias_kind":"pith_short_12","alias_value":"ETEOFKW2KPQW","created_at":"2026-07-05T09:42:15.156327+00:00"},{"alias_kind":"pith_short_16","alias_value":"ETEOFKW2KPQWMIGE","created_at":"2026-07-05T09:42:15.156327+00:00"},{"alias_kind":"pith_short_8","alias_value":"ETEOFKW2","created_at":"2026-07-05T09:42:15.156327+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08077","citing_title":"Modular Pretraining Enables Access Control","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2607.02513","citing_title":"LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12360","citing_title":"Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00510","citing_title":"Prototype Language Models","ref_index":136,"is_internal_anchor":false},{"citing_arxiv_id":"2602.16197","citing_title":"ModalImmune: Immunity Driven Unlearning via Self Destructive Training","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10403","citing_title":"Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7","json":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7.json","graph_json":"https://pith.science/api/pith-number/ETEOFKW2KPQWMIGEHOP3FMWGB7/graph.json","events_json":"https://pith.science/api/pith-number/ETEOFKW2KPQWMIGEHOP3FMWGB7/events.json","paper":"https://pith.science/paper/ETEOFKW2"},"agent_actions":{"view_html":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7","download_json":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7.json","view_paper":"https://pith.science/paper/ETEOFKW2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.04332&json=true","fetch_graph":"https://pith.science/api/pith-number/ETEOFKW2KPQWMIGEHOP3FMWGB7/graph.json","fetch_events":"https://pith.science/api/pith-number/ETEOFKW2KPQWMIGEHOP3FMWGB7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7/action/storage_attestation","attest_author":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7/action/author_attestation","sign_citation":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7/action/citation_signature","submit_replication":"https://pith.science/pith/ETEOFKW2KPQWMIGEHOP3FMWGB7/action/replication_record"}},"created_at":"2026-07-05T09:42:15.156327+00:00","updated_at":"2026-07-05T09:42:15.156327+00:00"}