{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6M5OGOZON2AAYNUMW7T4I2K55Z","short_pith_number":"pith:6M5OGOZO","schema_version":"1.0","canonical_sha256":"f33ae33b2e6e800c368cb7e7c4695dee47a79048f67f62a673fecab0ab6939a4","source":{"kind":"arxiv","id":"2505.00509","version":1},"attestation_state":"computed","paper":{"title":"Self-Ablating Transformers: More Interpretability, Less Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jeremias Ferrao, Keenan Pepper, Luhan Mikaelson, Natalia Perez-Campanero Antolin","submitted_at":"2025-05-01T13:25:37Z","abstract_excerpt":"A growing intuition in machine learning suggests a link between sparsity and interpretability. We introduce a novel self-ablation mechanism to investigate this connection ante-hoc in the context of language transformers. Our approach dynamically enforces a k-winner-takes-all constraint, forcing the model to demonstrate selective activation across neuron and attention units. Unlike post-hoc methods that analyze already-trained models, our approach integrates interpretability directly into model training, promoting feature localization from inception. Training small models on the TinyStories dat"},"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":"2505.00509","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-01T13:25:37Z","cross_cats_sorted":[],"title_canon_sha256":"e3c8249273ec102e0a4844325580e3d73027bb3cc3757a976986813a6117a6a9","abstract_canon_sha256":"d248fe8b09ff22a2d099db730aee6f8ddd20eb5261b17daeaa3f88a196b4bb7c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:57:14.716114Z","signature_b64":"MsrwqiHCQzziCJdl8mVJQLpRhWtnbMd9lKvIV15TXix6h2YZVRcjNwDZLFBW0d20eOWFPGdBMJDUIrtUq8E6AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f33ae33b2e6e800c368cb7e7c4695dee47a79048f67f62a673fecab0ab6939a4","last_reissued_at":"2026-07-05T10:57:14.715526Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:57:14.715526Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Ablating Transformers: More Interpretability, Less Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jeremias Ferrao, Keenan Pepper, Luhan Mikaelson, Natalia Perez-Campanero Antolin","submitted_at":"2025-05-01T13:25:37Z","abstract_excerpt":"A growing intuition in machine learning suggests a link between sparsity and interpretability. We introduce a novel self-ablation mechanism to investigate this connection ante-hoc in the context of language transformers. Our approach dynamically enforces a k-winner-takes-all constraint, forcing the model to demonstrate selective activation across neuron and attention units. Unlike post-hoc methods that analyze already-trained models, our approach integrates interpretability directly into model training, promoting feature localization from inception. Training small models on the TinyStories dat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.00509","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/2505.00509/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":"2505.00509","created_at":"2026-07-05T10:57:14.715592+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.00509v1","created_at":"2026-07-05T10:57:14.715592+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.00509","created_at":"2026-07-05T10:57:14.715592+00:00"},{"alias_kind":"pith_short_12","alias_value":"6M5OGOZON2AA","created_at":"2026-07-05T10:57:14.715592+00:00"},{"alias_kind":"pith_short_16","alias_value":"6M5OGOZON2AAYNUM","created_at":"2026-07-05T10:57:14.715592+00:00"},{"alias_kind":"pith_short_8","alias_value":"6M5OGOZO","created_at":"2026-07-05T10:57:14.715592+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/6M5OGOZON2AAYNUMW7T4I2K55Z","json":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z.json","graph_json":"https://pith.science/api/pith-number/6M5OGOZON2AAYNUMW7T4I2K55Z/graph.json","events_json":"https://pith.science/api/pith-number/6M5OGOZON2AAYNUMW7T4I2K55Z/events.json","paper":"https://pith.science/paper/6M5OGOZO"},"agent_actions":{"view_html":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z","download_json":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z.json","view_paper":"https://pith.science/paper/6M5OGOZO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.00509&json=true","fetch_graph":"https://pith.science/api/pith-number/6M5OGOZON2AAYNUMW7T4I2K55Z/graph.json","fetch_events":"https://pith.science/api/pith-number/6M5OGOZON2AAYNUMW7T4I2K55Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z/action/storage_attestation","attest_author":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z/action/author_attestation","sign_citation":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z/action/citation_signature","submit_replication":"https://pith.science/pith/6M5OGOZON2AAYNUMW7T4I2K55Z/action/replication_record"}},"created_at":"2026-07-05T10:57:14.715592+00:00","updated_at":"2026-07-05T10:57:14.715592+00:00"}