{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:OSRK4UFQ4JSNBHSIJL3UXZL4RW","short_pith_number":"pith:OSRK4UFQ","schema_version":"1.0","canonical_sha256":"74a2ae50b0e264d09e484af74be57c8db4d98e653926a0ba233dee72d8bc8648","source":{"kind":"arxiv","id":"2310.02557","version":3},"attestation_state":"computed","paper":{"title":"Generalization in diffusion models arises from geometry-adaptive harmonic representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Eero P. Simoncelli, Florentin Guth, St\\'ephane Mallat, Zahra Kadkhodaie","submitted_at":"2023-10-04T03:30:32Z","abstract_excerpt":"Deep neural networks (DNNs) trained for image denoising are able to generate high-quality samples with score-based reverse diffusion algorithms. These impressive capabilities seem to imply an escape from the curse of dimensionality, but recent reports of memorization of the training set raise the question of whether these networks are learning the \"true\" continuous density of the data. Here, we show that two DNNs trained on non-overlapping subsets of a dataset learn nearly the same score function, and thus the same density, when the number of training images is large enough. In this regime of "},"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":"2310.02557","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2023-10-04T03:30:32Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"81d7597b1c768baf5a941c6cafac032e725a6a21b16485d52304b7030c59129d","abstract_canon_sha256":"5f0eceee7bc407b395dfaa1dc4311d53f3209ad62fb2aab5cdce76656e9744f5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:21:51.914654Z","signature_b64":"zaUNOdHluf/qGBQ+Kkh/DwA6OXRZtFnY6lwyKd4ts7/tps0Vvwv3WmihlaMiXCY68wYlhPwkRtqfO+/LHA2KDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"74a2ae50b0e264d09e484af74be57c8db4d98e653926a0ba233dee72d8bc8648","last_reissued_at":"2026-07-05T10:21:51.914157Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:21:51.914157Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Generalization in diffusion models arises from geometry-adaptive harmonic representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Eero P. Simoncelli, Florentin Guth, St\\'ephane Mallat, Zahra Kadkhodaie","submitted_at":"2023-10-04T03:30:32Z","abstract_excerpt":"Deep neural networks (DNNs) trained for image denoising are able to generate high-quality samples with score-based reverse diffusion algorithms. These impressive capabilities seem to imply an escape from the curse of dimensionality, but recent reports of memorization of the training set raise the question of whether these networks are learning the \"true\" continuous density of the data. Here, we show that two DNNs trained on non-overlapping subsets of a dataset learn nearly the same score function, and thus the same density, when the number of training images is large enough. In this regime of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.02557","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/2310.02557/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":"2310.02557","created_at":"2026-07-05T10:21:51.914216+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.02557v3","created_at":"2026-07-05T10:21:51.914216+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.02557","created_at":"2026-07-05T10:21:51.914216+00:00"},{"alias_kind":"pith_short_12","alias_value":"OSRK4UFQ4JSN","created_at":"2026-07-05T10:21:51.914216+00:00"},{"alias_kind":"pith_short_16","alias_value":"OSRK4UFQ4JSNBHSI","created_at":"2026-07-05T10:21:51.914216+00:00"},{"alias_kind":"pith_short_8","alias_value":"OSRK4UFQ","created_at":"2026-07-05T10:21:51.914216+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"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":4,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08554","citing_title":"A Theoretical Analysis of Memory and Overfitting Phenomena in Stochastic Interpolation Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03848","citing_title":"Generating quantum ensembles via reverse-time quantum diffusions","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28092","citing_title":"Diffusion Model Attribution via Spectral Coupling of Denoiser Responses","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2503.11615","citing_title":"From Score Matching to Diffusion: A Fine-Grained Error Analysis in the Gaussian Setting","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2603.13419","citing_title":"Diffusion Models Memorize in Training -- and Generalize in Inference","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20299","citing_title":"Mechanisms of Misgeneralization in Physical Sequence Modeling","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2510.21366","citing_title":"BADiff: Bandwidth Adaptive Diffusion Model","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26841","citing_title":"Language Diffusion Models are Associative Memories Capable of Retrieving Unseen Data","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17673","citing_title":"Grokking of Diffusion Models: Case Study on Modular Addition","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW","json":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW.json","graph_json":"https://pith.science/api/pith-number/OSRK4UFQ4JSNBHSIJL3UXZL4RW/graph.json","events_json":"https://pith.science/api/pith-number/OSRK4UFQ4JSNBHSIJL3UXZL4RW/events.json","paper":"https://pith.science/paper/OSRK4UFQ"},"agent_actions":{"view_html":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW","download_json":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW.json","view_paper":"https://pith.science/paper/OSRK4UFQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.02557&json=true","fetch_graph":"https://pith.science/api/pith-number/OSRK4UFQ4JSNBHSIJL3UXZL4RW/graph.json","fetch_events":"https://pith.science/api/pith-number/OSRK4UFQ4JSNBHSIJL3UXZL4RW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW/action/storage_attestation","attest_author":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW/action/author_attestation","sign_citation":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW/action/citation_signature","submit_replication":"https://pith.science/pith/OSRK4UFQ4JSNBHSIJL3UXZL4RW/action/replication_record"}},"created_at":"2026-07-05T10:21:51.914216+00:00","updated_at":"2026-07-05T10:21:51.914216+00:00"}