{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XTYU6EUZWRI6EBU2PYG3UUWNUO","short_pith_number":"pith:XTYU6EUZ","schema_version":"1.0","canonical_sha256":"bcf14f1299b451e2069a7e0dba52cda386c3aa8cd71a9373336daaff223821a7","source":{"kind":"arxiv","id":"2603.13419","version":2},"attestation_state":"computed","paper":{"title":"Diffusion Models Memorize in Training -- and Generalize in Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Markus Kollmann, Tim Kaiser","submitted_at":"2026-03-12T21:02:17Z","abstract_excerpt":"Diffusion models generalize well in practice. However, an optimal diffusion model fully memorizes the training data and therefore fails to generalize, raising the question of what induces generalization in a real diffusion model. We show that, despite generalizing at the sample level, diffusion models progressively overfit the denoising training objective and thereby create a generalization gap between the performance on validation and training samples. This gap is most pronounced at intermediate noise levels. Using a fully analytic error-prone toy model, we trace the factors affecting the gen"},"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":"2603.13419","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-12T21:02:17Z","cross_cats_sorted":[],"title_canon_sha256":"906e7888a176fa0ff84047551e496efad52e274328aab0c8186e993a9dcf080f","abstract_canon_sha256":"09a1eaeaf825a7dc2c0155581590a41c7e1d73a0713b6dc2abfaee4c61d6955d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:05:17.690768Z","signature_b64":"Sk2zBuDTP3kXucgMTzHXD66IQ9C97FJ2GtYP230k82OTJdsOvSX7q5NyRscHAi391MbgphPVsc0pWgGjCIsyAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bcf14f1299b451e2069a7e0dba52cda386c3aa8cd71a9373336daaff223821a7","last_reissued_at":"2026-05-21T01:05:17.690176Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:05:17.690176Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Diffusion Models Memorize in Training -- and Generalize in Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Markus Kollmann, Tim Kaiser","submitted_at":"2026-03-12T21:02:17Z","abstract_excerpt":"Diffusion models generalize well in practice. However, an optimal diffusion model fully memorizes the training data and therefore fails to generalize, raising the question of what induces generalization in a real diffusion model. We show that, despite generalizing at the sample level, diffusion models progressively overfit the denoising training objective and thereby create a generalization gap between the performance on validation and training samples. This gap is most pronounced at intermediate noise levels. Using a fully analytic error-prone toy model, we trace the factors affecting the gen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.13419","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/2603.13419/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":"2603.13419","created_at":"2026-05-21T01:05:17.690271+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.13419v2","created_at":"2026-05-21T01:05:17.690271+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.13419","created_at":"2026-05-21T01:05:17.690271+00:00"},{"alias_kind":"pith_short_12","alias_value":"XTYU6EUZWRI6","created_at":"2026-05-21T01:05:17.690271+00:00"},{"alias_kind":"pith_short_16","alias_value":"XTYU6EUZWRI6EBU2","created_at":"2026-05-21T01:05:17.690271+00:00"},{"alias_kind":"pith_short_8","alias_value":"XTYU6EUZ","created_at":"2026-05-21T01:05:17.690271+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/XTYU6EUZWRI6EBU2PYG3UUWNUO","json":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO.json","graph_json":"https://pith.science/api/pith-number/XTYU6EUZWRI6EBU2PYG3UUWNUO/graph.json","events_json":"https://pith.science/api/pith-number/XTYU6EUZWRI6EBU2PYG3UUWNUO/events.json","paper":"https://pith.science/paper/XTYU6EUZ"},"agent_actions":{"view_html":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO","download_json":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO.json","view_paper":"https://pith.science/paper/XTYU6EUZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.13419&json=true","fetch_graph":"https://pith.science/api/pith-number/XTYU6EUZWRI6EBU2PYG3UUWNUO/graph.json","fetch_events":"https://pith.science/api/pith-number/XTYU6EUZWRI6EBU2PYG3UUWNUO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO/action/storage_attestation","attest_author":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO/action/author_attestation","sign_citation":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO/action/citation_signature","submit_replication":"https://pith.science/pith/XTYU6EUZWRI6EBU2PYG3UUWNUO/action/replication_record"}},"created_at":"2026-05-21T01:05:17.690271+00:00","updated_at":"2026-05-21T01:05:17.690271+00:00"}