{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7RS54ATOSFMSLTYA6U5667MWIS","short_pith_number":"pith:7RS54ATO","schema_version":"1.0","canonical_sha256":"fc65de026e915925cf00f53bef7d964495e1dd1f251426476586a0c39844c2a1","source":{"kind":"arxiv","id":"2411.19339","version":3},"attestation_state":"computed","paper":{"title":"Towards a Mechanistic Explanation of Diffusion Model Generalization","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Berend Zwartsenberg, Frank Wood, Kevin Murphy, Matthew Niedoba","submitted_at":"2024-11-28T19:22:17Z","abstract_excerpt":"We propose a simple, training-free mechanism which explains the generalization behaviour of diffusion models. By comparing pre-trained diffusion models to their theoretically optimal empirical counterparts, we identify a shared local inductive bias across a variety of network architectures. From this observation, we hypothesize that network denoisers generalize through localized denoising operations, as these operations approximate the training objective well over much of the training distribution. To validate our hypothesis, we introduce novel denoising algorithms which aggregate local empiri"},"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":"2411.19339","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-28T19:22:17Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"eadf792641edbbadc664bffd2f793c3aa336e2ae5f2b3da7c6a4ededca076866","abstract_canon_sha256":"99bd503e0e9f7b58f8fa556a464229e007240d0967dc9dd9cec48837f0e14e7a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:18:40.648580Z","signature_b64":"aEeIT+kO0E5ZqQEsj6RMjK8Tkhf3y4uPXYofxKE4NsERqDBe9FpXErcKiftqiBGC2NUQMU/gzBuw0NrFO6sJCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc65de026e915925cf00f53bef7d964495e1dd1f251426476586a0c39844c2a1","last_reissued_at":"2026-07-05T11:18:40.648086Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:18:40.648086Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards a Mechanistic Explanation of Diffusion Model Generalization","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Berend Zwartsenberg, Frank Wood, Kevin Murphy, Matthew Niedoba","submitted_at":"2024-11-28T19:22:17Z","abstract_excerpt":"We propose a simple, training-free mechanism which explains the generalization behaviour of diffusion models. By comparing pre-trained diffusion models to their theoretically optimal empirical counterparts, we identify a shared local inductive bias across a variety of network architectures. From this observation, we hypothesize that network denoisers generalize through localized denoising operations, as these operations approximate the training objective well over much of the training distribution. To validate our hypothesis, we introduce novel denoising algorithms which aggregate local empiri"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.19339","kind":"arxiv","version":3},"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/2411.19339/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":"2411.19339","created_at":"2026-07-05T11:18:40.648139+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.19339v3","created_at":"2026-07-05T11:18:40.648139+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.19339","created_at":"2026-07-05T11:18:40.648139+00:00"},{"alias_kind":"pith_short_12","alias_value":"7RS54ATOSFMS","created_at":"2026-07-05T11:18:40.648139+00:00"},{"alias_kind":"pith_short_16","alias_value":"7RS54ATOSFMSLTYA","created_at":"2026-07-05T11:18:40.648139+00:00"},{"alias_kind":"pith_short_8","alias_value":"7RS54ATO","created_at":"2026-07-05T11:18:40.648139+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08041","citing_title":"An exact information theory of generalization phase transitions in Bayesian diffusion models","ref_index":48,"is_internal_anchor":true},{"citing_arxiv_id":"2606.06624","citing_title":"Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2508.06614","citing_title":"Local Diffusion Models and Phases of Data Distributions","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10019","citing_title":"The two clocks and the innovation window: When and how generative models learn rules","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS","json":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS.json","graph_json":"https://pith.science/api/pith-number/7RS54ATOSFMSLTYA6U5667MWIS/graph.json","events_json":"https://pith.science/api/pith-number/7RS54ATOSFMSLTYA6U5667MWIS/events.json","paper":"https://pith.science/paper/7RS54ATO"},"agent_actions":{"view_html":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS","download_json":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS.json","view_paper":"https://pith.science/paper/7RS54ATO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.19339&json=true","fetch_graph":"https://pith.science/api/pith-number/7RS54ATOSFMSLTYA6U5667MWIS/graph.json","fetch_events":"https://pith.science/api/pith-number/7RS54ATOSFMSLTYA6U5667MWIS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS/action/storage_attestation","attest_author":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS/action/author_attestation","sign_citation":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS/action/citation_signature","submit_replication":"https://pith.science/pith/7RS54ATOSFMSLTYA6U5667MWIS/action/replication_record"}},"created_at":"2026-07-05T11:18:40.648139+00:00","updated_at":"2026-07-05T11:18:40.648139+00:00"}