{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7MHJFZN2H6RU3DFL563HQ53NOI","short_pith_number":"pith:7MHJFZN2","schema_version":"1.0","canonical_sha256":"fb0e92e5ba3fa34d8cabefb678776d720f97be1a689bf44a2dd822cb2a22f33a","source":{"kind":"arxiv","id":"2505.22841","version":1},"attestation_state":"computed","paper":{"title":"Kernel-Smoothed Scores for Denoising Diffusion: A Bias-Variance Study","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["math.PR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Emmanuel Schertzer, Fran\\c{c}ois Ged, Franck Gabriel, Maria Han Veiga","submitted_at":"2025-05-28T20:22:18Z","abstract_excerpt":"Diffusion models now set the benchmark in high-fidelity generative sampling, yet they can, in principle, be prone to memorization. In this case, their learned score overfits the finite dataset so that the reverse-time SDE samples are mostly training points. In this paper, we interpret the empirical score as a noisy version of the true score and show that its covariance matrix is asymptotically a re-weighted data PCA. In large dimension, the small time limit makes the noise variance blow up while simultaneously reducing spatial correlation. To reduce this variance, we introduce a kernel-smoothe"},"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":"2505.22841","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-28T20:22:18Z","cross_cats_sorted":["math.PR","stat.ML"],"title_canon_sha256":"495c22c2bce28a6fcecb1cbe7333259364b27d367879597c91b503eb1d5cb661","abstract_canon_sha256":"f88f5e294f5bf9f051f3550b60aa2ba649f0fffb480f1beafacd6aa928145bfb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:50.732697Z","signature_b64":"eyKTO88Z+WoU2RBei0w28smoCqUtCnbQu7lgRuR4PeV2xwoNmqptVwm/Nalte7iuxJPjGAUik52ALfHXds+qBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fb0e92e5ba3fa34d8cabefb678776d720f97be1a689bf44a2dd822cb2a22f33a","last_reissued_at":"2026-07-05T11:11:50.732152Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:50.732152Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Kernel-Smoothed Scores for Denoising Diffusion: A Bias-Variance Study","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["math.PR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Emmanuel Schertzer, Fran\\c{c}ois Ged, Franck Gabriel, Maria Han Veiga","submitted_at":"2025-05-28T20:22:18Z","abstract_excerpt":"Diffusion models now set the benchmark in high-fidelity generative sampling, yet they can, in principle, be prone to memorization. In this case, their learned score overfits the finite dataset so that the reverse-time SDE samples are mostly training points. In this paper, we interpret the empirical score as a noisy version of the true score and show that its covariance matrix is asymptotically a re-weighted data PCA. In large dimension, the small time limit makes the noise variance blow up while simultaneously reducing spatial correlation. To reduce this variance, we introduce a kernel-smoothe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.22841","kind":"arxiv","version":1},"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/2505.22841/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":"2505.22841","created_at":"2026-07-05T11:11:50.732216+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22841v1","created_at":"2026-07-05T11:11:50.732216+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22841","created_at":"2026-07-05T11:11:50.732216+00:00"},{"alias_kind":"pith_short_12","alias_value":"7MHJFZN2H6RU","created_at":"2026-07-05T11:11:50.732216+00:00"},{"alias_kind":"pith_short_16","alias_value":"7MHJFZN2H6RU3DFL","created_at":"2026-07-05T11:11:50.732216+00:00"},{"alias_kind":"pith_short_8","alias_value":"7MHJFZN2","created_at":"2026-07-05T11:11:50.732216+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13433","citing_title":"Smoothed-KL Reweighting: A Principled Account and Matching Rule for SNR-Based Diffusion Training","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14276","citing_title":"Training-Free Generative Sampling via Moment-Matched Score Smoothing","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI","json":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI.json","graph_json":"https://pith.science/api/pith-number/7MHJFZN2H6RU3DFL563HQ53NOI/graph.json","events_json":"https://pith.science/api/pith-number/7MHJFZN2H6RU3DFL563HQ53NOI/events.json","paper":"https://pith.science/paper/7MHJFZN2"},"agent_actions":{"view_html":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI","download_json":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI.json","view_paper":"https://pith.science/paper/7MHJFZN2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22841&json=true","fetch_graph":"https://pith.science/api/pith-number/7MHJFZN2H6RU3DFL563HQ53NOI/graph.json","fetch_events":"https://pith.science/api/pith-number/7MHJFZN2H6RU3DFL563HQ53NOI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI/action/storage_attestation","attest_author":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI/action/author_attestation","sign_citation":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI/action/citation_signature","submit_replication":"https://pith.science/pith/7MHJFZN2H6RU3DFL563HQ53NOI/action/replication_record"}},"created_at":"2026-07-05T11:11:50.732216+00:00","updated_at":"2026-07-05T11:11:50.732216+00:00"}