{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VGTYJQXRS376HPESLWRUON2DTF","short_pith_number":"pith:VGTYJQXR","schema_version":"1.0","canonical_sha256":"a9a784c2f196ffe3bc925da34737439967d32d38f1f7c5bf1acc1390c862195d","source":{"kind":"arxiv","id":"2409.09951","version":1},"attestation_state":"computed","paper":{"title":"Optimal ablation for interpretability","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Lucas Janson, Maximilian Li","submitted_at":"2024-09-16T02:45:54Z","abstract_excerpt":"Interpretability studies often involve tracing the flow of information through machine learning models to identify specific model components that perform relevant computations for tasks of interest. Prior work quantifies the importance of a model component on a particular task by measuring the impact of performing ablation on that component, or simulating model inference with the component disabled. We propose a new method, optimal ablation (OA), and show that OA-based component importance has theoretical and empirical advantages over measuring importance via other ablation methods. We also sh"},"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":"2409.09951","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-09-16T02:45:54Z","cross_cats_sorted":[],"title_canon_sha256":"91894ebf57a0c3d7bf6d2a9d2361c75b2c1a6af01355a1fe1fd5422f74491a8e","abstract_canon_sha256":"9210bcbe09536425467987d0d4f7f33ddc6993fb2914e7c93e87b85caf806e22"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:07:16.768994Z","signature_b64":"xqdtQQSJM/SE5FKuomRwzDt+TaF520U97LxQQ/nCGTWcNyx56pDVI7olQY3PVxA26mAtnDsoHk1bjI4iHfPvCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a9a784c2f196ffe3bc925da34737439967d32d38f1f7c5bf1acc1390c862195d","last_reissued_at":"2026-07-05T09:07:16.768581Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:07:16.768581Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimal ablation for interpretability","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Lucas Janson, Maximilian Li","submitted_at":"2024-09-16T02:45:54Z","abstract_excerpt":"Interpretability studies often involve tracing the flow of information through machine learning models to identify specific model components that perform relevant computations for tasks of interest. Prior work quantifies the importance of a model component on a particular task by measuring the impact of performing ablation on that component, or simulating model inference with the component disabled. We propose a new method, optimal ablation (OA), and show that OA-based component importance has theoretical and empirical advantages over measuring importance via other ablation methods. We also sh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.09951","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/2409.09951/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":"2409.09951","created_at":"2026-07-05T09:07:16.768638+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.09951v1","created_at":"2026-07-05T09:07:16.768638+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.09951","created_at":"2026-07-05T09:07:16.768638+00:00"},{"alias_kind":"pith_short_12","alias_value":"VGTYJQXRS376","created_at":"2026-07-05T09:07:16.768638+00:00"},{"alias_kind":"pith_short_16","alias_value":"VGTYJQXRS376HPES","created_at":"2026-07-05T09:07:16.768638+00:00"},{"alias_kind":"pith_short_8","alias_value":"VGTYJQXR","created_at":"2026-07-05T09:07:16.768638+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.12469","citing_title":"Sparse Concept Anchoring for Interpretable and Controllable Neural Representations","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05341","citing_title":"Feature Starvation as Geometric Instability in Sparse Autoencoders","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF","json":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF.json","graph_json":"https://pith.science/api/pith-number/VGTYJQXRS376HPESLWRUON2DTF/graph.json","events_json":"https://pith.science/api/pith-number/VGTYJQXRS376HPESLWRUON2DTF/events.json","paper":"https://pith.science/paper/VGTYJQXR"},"agent_actions":{"view_html":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF","download_json":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF.json","view_paper":"https://pith.science/paper/VGTYJQXR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.09951&json=true","fetch_graph":"https://pith.science/api/pith-number/VGTYJQXRS376HPESLWRUON2DTF/graph.json","fetch_events":"https://pith.science/api/pith-number/VGTYJQXRS376HPESLWRUON2DTF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF/action/storage_attestation","attest_author":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF/action/author_attestation","sign_citation":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF/action/citation_signature","submit_replication":"https://pith.science/pith/VGTYJQXRS376HPESLWRUON2DTF/action/replication_record"}},"created_at":"2026-07-05T09:07:16.768638+00:00","updated_at":"2026-07-05T09:07:16.768638+00:00"}