{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YKLK3NXOM57UFDDH7FNG4UYFE3","short_pith_number":"pith:YKLK3NXO","schema_version":"1.0","canonical_sha256":"c296adb6ee677f428c67f95a6e530526f9fa256d2b08e35260b1a451189fb0fd","source":{"kind":"arxiv","id":"2602.20062","version":2},"attestation_state":"computed","paper":{"title":"A Theory of How Pretraining Shapes Inductive Bias in Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew M. Saxe, Clementine Domine, Flavia Mancini, Francesco Locatello, Marco Mondelli, Nicolas Anguita, Samuel Lippl","submitted_at":"2026-02-23T17:19:33Z","abstract_excerpt":"Pretraining and fine-tuning are central stages in modern machine learning systems. In practice, feature learning plays an important role across both stages: deep neural networks learn a broad range of useful features during pretraining and further refine those features during fine-tuning. However, an end-to-end theoretical understanding of how choices of initialization impact the ability to reuse and refine features during fine-tuning has remained elusive. Here we develop an analytical theory of the pretraining fine-tuning pipeline in diagonal linear networks, deriving exact expressions for th"},"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":"2602.20062","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-23T17:19:33Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"1f183143c92b7f530713610e546df9513040ff8e23ae1a872f0da0511e2cb5c1","abstract_canon_sha256":"a9571f26c14af7946598f656bc436ec318e174528c3180951b0f646aa6687fe4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-01T01:17:47.519440Z","signature_b64":"3B8uMpS5JqD5l66CvlcDSxmJBqwnsFkCTKx+mm38eARuMlixAjG429XTVl2EW43arhVj7mTE4PYSCx/fZ+AgDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c296adb6ee677f428c67f95a6e530526f9fa256d2b08e35260b1a451189fb0fd","last_reissued_at":"2026-07-01T01:17:47.518945Z","signature_status":"signed_v1","first_computed_at":"2026-07-01T01:17:47.518945Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Theory of How Pretraining Shapes Inductive Bias in Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew M. Saxe, Clementine Domine, Flavia Mancini, Francesco Locatello, Marco Mondelli, Nicolas Anguita, Samuel Lippl","submitted_at":"2026-02-23T17:19:33Z","abstract_excerpt":"Pretraining and fine-tuning are central stages in modern machine learning systems. In practice, feature learning plays an important role across both stages: deep neural networks learn a broad range of useful features during pretraining and further refine those features during fine-tuning. However, an end-to-end theoretical understanding of how choices of initialization impact the ability to reuse and refine features during fine-tuning has remained elusive. Here we develop an analytical theory of the pretraining fine-tuning pipeline in diagonal linear networks, deriving exact expressions for th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.20062","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/2602.20062/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":"2602.20062","created_at":"2026-07-01T01:17:47.519010+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.20062v2","created_at":"2026-07-01T01:17:47.519010+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.20062","created_at":"2026-07-01T01:17:47.519010+00:00"},{"alias_kind":"pith_short_12","alias_value":"YKLK3NXOM57U","created_at":"2026-07-01T01:17:47.519010+00:00"},{"alias_kind":"pith_short_16","alias_value":"YKLK3NXOM57UFDDH","created_at":"2026-07-01T01:17:47.519010+00:00"},{"alias_kind":"pith_short_8","alias_value":"YKLK3NXO","created_at":"2026-07-01T01:17:47.519010+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.22972","citing_title":"A mathematical theory of balancing relational generalization and memorization","ref_index":93,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20105","citing_title":"Optimal Representation Size: High-Dimensional Analysis of Pretraining and Linear Probing","ref_index":63,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3","json":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3.json","graph_json":"https://pith.science/api/pith-number/YKLK3NXOM57UFDDH7FNG4UYFE3/graph.json","events_json":"https://pith.science/api/pith-number/YKLK3NXOM57UFDDH7FNG4UYFE3/events.json","paper":"https://pith.science/paper/YKLK3NXO"},"agent_actions":{"view_html":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3","download_json":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3.json","view_paper":"https://pith.science/paper/YKLK3NXO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.20062&json=true","fetch_graph":"https://pith.science/api/pith-number/YKLK3NXOM57UFDDH7FNG4UYFE3/graph.json","fetch_events":"https://pith.science/api/pith-number/YKLK3NXOM57UFDDH7FNG4UYFE3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3/action/storage_attestation","attest_author":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3/action/author_attestation","sign_citation":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3/action/citation_signature","submit_replication":"https://pith.science/pith/YKLK3NXOM57UFDDH7FNG4UYFE3/action/replication_record"}},"created_at":"2026-07-01T01:17:47.519010+00:00","updated_at":"2026-07-01T01:17:47.519010+00:00"}