{"paper":{"title":"Toy Combinatorial Interpretability Models Reveal Lottery Tickets in Early Feature Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Winning tickets correspond to precursor locations in feature space already near the final codes at initialization.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alon Bebchuk, Nir Shavit","submitted_at":"2026-05-18T00:00:53Z","abstract_excerpt":"The lottery ticket hypothesis posits that dense networks contain sparse subnetworks, ``winning tickets,'' that, when rewound to their initial weights and retrained in isolation, match the performance of the full model. We ask a more mechanistic question: what internal object does a winning ticket preserve? We work in a combinatorial, clause-structured toy setting that admits an interpretable feature-space representation with well-defined combinatorial distances between features. We show that winning tickets in weight space correspond to precursor locations in feature space that are already nea"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"winning tickets in weight space correspond to precursor locations in feature space that are already near, at initialization, to the final feature-channel codes. A winning ticket is thus a family of compatible code locations that jointly balance proximity to final codes with low inter-feature interference.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The combinatorial, clause-structured toy setting admits an interpretable feature-space representation with well-defined combinatorial distances between features that capture the relevant dynamics of real neural networks under superposition.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"In a combinatorial toy setting, winning lottery tickets preserve families of compatible feature locations in early feature space that balance proximity to final codes with low interference, rather than specific weight subnetworks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Winning tickets correspond to precursor locations in feature space already near the final codes at initialization.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8368c865ca9c63ef62ac228ffba84ec337f60e9ccfe53ba188a3996aaaff97a0"},"source":{"id":"2605.17704","kind":"arxiv","version":1},"verdict":{"id":"ca06049a-8b98-4cf9-91f2-eabf34870def","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T22:06:17.836785Z","strongest_claim":"winning tickets in weight space correspond to precursor locations in feature space that are already near, at initialization, to the final feature-channel codes. A winning ticket is thus a family of compatible code locations that jointly balance proximity to final codes with low inter-feature interference.","one_line_summary":"In a combinatorial toy setting, winning lottery tickets preserve families of compatible feature locations in early feature space that balance proximity to final codes with low interference, rather than specific weight subnetworks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The combinatorial, clause-structured toy setting admits an interpretable feature-space representation with well-defined combinatorial distances between features that capture the relevant dynamics of real neural networks under superposition.","pith_extraction_headline":"Winning tickets correspond to precursor locations in feature space already near the final codes at initialization."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.17704/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:19.405696Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T22:10:57.458735Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"shingle_duplication","ran_at":"2026-05-19T21:49:43.479897Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T21:49:43.309341Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.512988Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T21:21:59.173374Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T21:21:57.420154Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"d12a80bad7596fa66e800253bec3b70f84da1d583d68da6f48c4873a1a360fd1"},"references":{"count":91,"sample":[{"doi":"","year":null,"title":"International Conference on Learning Representations , year=","work_id":"61b9ca7f-3fca-4d6c-8df5-e91576c98a1b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"International Conference on Learning Representations , year=","work_id":"9145ab82-d376-4eef-ab5b-332886510d7b","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"International Conference on Machine Learning , year=","work_id":"4b9f0050-4324-4b4d-95a0-77ec2030a415","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"International Conference on Learning Representations , year=","work_id":"a8140a03-8d67-4bf8-93d3-896dbb7b2dbe","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems , year=","work_id":"9ac2d762-b569-4d13-9f39-17080231cf34","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":91,"snapshot_sha256":"049f8f95a42883fe34c333d33c3083246569f043978576b244084ad54264fa8f","internal_anchors":6},"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"}