{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OXKMP2V2KWHZFRKR65KMBSYM7S","short_pith_number":"pith:OXKMP2V2","schema_version":"1.0","canonical_sha256":"75d4c7eaba558f92c551f754c0cb0cfcad71955f96828c5e5672271c9716bb9f","source":{"kind":"arxiv","id":"2502.03435","version":2},"attestation_state":"computed","paper":{"title":"Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Claire Boyer, G\\'erard Biau, Pierre Marion, Yu-Han Wu","submitted_at":"2025-02-05T18:29:35Z","abstract_excerpt":"Denoising score matching plays a pivotal role in the performance of diffusion-based generative models. However, the empirical optimal score--the exact solution to the denoising score matching--leads to memorization, where generated samples replicate the training data. Yet, in practice, only a moderate degree of memorization is observed, even without explicit regularization. In this paper, we investigate this phenomenon by uncovering an implicit regularization mechanism driven by large learning rates. Specifically, we show that in the small-noise regime, the empirical optimal score exhibits hig"},"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":"2502.03435","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2025-02-05T18:29:35Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"a41f4bce40d89d3fd14496c9c2410e080e64206823169d362f2e13b685de42f8","abstract_canon_sha256":"8ba829a21d2a491ffe1bd14daec5086f0f7afa43925ffed6c27f6db84b23dd78"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:59:05.436148Z","signature_b64":"H560EWxlLmW6E3ksFXiU7sEfx4CT0MyhU10TRt0YWrnwHzFqbhvZYu3vd6HgdybT8kz6ZPkjZoo4JlrtwvVPBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"75d4c7eaba558f92c551f754c0cb0cfcad71955f96828c5e5672271c9716bb9f","last_reissued_at":"2026-07-05T10:59:05.435647Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:59:05.435647Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Claire Boyer, G\\'erard Biau, Pierre Marion, Yu-Han Wu","submitted_at":"2025-02-05T18:29:35Z","abstract_excerpt":"Denoising score matching plays a pivotal role in the performance of diffusion-based generative models. However, the empirical optimal score--the exact solution to the denoising score matching--leads to memorization, where generated samples replicate the training data. Yet, in practice, only a moderate degree of memorization is observed, even without explicit regularization. In this paper, we investigate this phenomenon by uncovering an implicit regularization mechanism driven by large learning rates. Specifically, we show that in the small-noise regime, the empirical optimal score exhibits hig"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.03435","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/2502.03435/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":"2502.03435","created_at":"2026-07-05T10:59:05.435703+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.03435v2","created_at":"2026-07-05T10:59:05.435703+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.03435","created_at":"2026-07-05T10:59:05.435703+00:00"},{"alias_kind":"pith_short_12","alias_value":"OXKMP2V2KWHZ","created_at":"2026-07-05T10:59:05.435703+00:00"},{"alias_kind":"pith_short_16","alias_value":"OXKMP2V2KWHZFRKR","created_at":"2026-07-05T10:59:05.435703+00:00"},{"alias_kind":"pith_short_8","alias_value":"OXKMP2V2","created_at":"2026-07-05T10:59:05.435703+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2503.11615","citing_title":"From Score Matching to Diffusion: A Fine-Grained Error Analysis in the Gaussian Setting","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S","json":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S.json","graph_json":"https://pith.science/api/pith-number/OXKMP2V2KWHZFRKR65KMBSYM7S/graph.json","events_json":"https://pith.science/api/pith-number/OXKMP2V2KWHZFRKR65KMBSYM7S/events.json","paper":"https://pith.science/paper/OXKMP2V2"},"agent_actions":{"view_html":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S","download_json":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S.json","view_paper":"https://pith.science/paper/OXKMP2V2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.03435&json=true","fetch_graph":"https://pith.science/api/pith-number/OXKMP2V2KWHZFRKR65KMBSYM7S/graph.json","fetch_events":"https://pith.science/api/pith-number/OXKMP2V2KWHZFRKR65KMBSYM7S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S/action/storage_attestation","attest_author":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S/action/author_attestation","sign_citation":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S/action/citation_signature","submit_replication":"https://pith.science/pith/OXKMP2V2KWHZFRKR65KMBSYM7S/action/replication_record"}},"created_at":"2026-07-05T10:59:05.435703+00:00","updated_at":"2026-07-05T10:59:05.435703+00:00"}