{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LGCXGK3HSPCZSHPFQUFB4LW62A","short_pith_number":"pith:LGCXGK3H","schema_version":"1.0","canonical_sha256":"5985732b6793c5991de5850a1e2eded039f7051156a827a3bb3a48d32b533845","source":{"kind":"arxiv","id":"2504.12532","version":1},"attestation_state":"computed","paper":{"title":"Generalization through variance: how noise shapes inductive biases in diffusion models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.AI"],"primary_cat":"cs.LG","authors_text":"John J. Vastola","submitted_at":"2025-04-16T23:41:10Z","abstract_excerpt":"How diffusion models generalize beyond their training set is not known, and is somewhat mysterious given two facts: the optimum of the denoising score matching (DSM) objective usually used to train diffusion models is the score function of the training distribution; and the networks usually used to learn the score function are expressive enough to learn this score to high accuracy. We claim that a certain feature of the DSM objective -- the fact that its target is not the training distribution's score, but a noisy quantity only equal to it in expectation -- strongly impacts whether and to what"},"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":"2504.12532","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-16T23:41:10Z","cross_cats_sorted":["cond-mat.dis-nn","cs.AI"],"title_canon_sha256":"a2449f94b5444f73db39eca7ddeebd3087dea8aa6caa374d4f19b82ea11fe238","abstract_canon_sha256":"ecab4e63223cfdb05d5b1f2e419a25ff3dbc5162c6e78da3e7ae965ec1fc3245"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:50:17.902602Z","signature_b64":"cTew6TEc6MWUtLzn4pANL8u1k/4qDjFmy1A+EeSVzkxCI0qQUCbmZm/2Z0jIK/wY4E7uAkqk+RSRBwrnCebVCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5985732b6793c5991de5850a1e2eded039f7051156a827a3bb3a48d32b533845","last_reissued_at":"2026-07-05T10:50:17.902063Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:50:17.902063Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Generalization through variance: how noise shapes inductive biases in diffusion models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.AI"],"primary_cat":"cs.LG","authors_text":"John J. Vastola","submitted_at":"2025-04-16T23:41:10Z","abstract_excerpt":"How diffusion models generalize beyond their training set is not known, and is somewhat mysterious given two facts: the optimum of the denoising score matching (DSM) objective usually used to train diffusion models is the score function of the training distribution; and the networks usually used to learn the score function are expressive enough to learn this score to high accuracy. We claim that a certain feature of the DSM objective -- the fact that its target is not the training distribution's score, but a noisy quantity only equal to it in expectation -- strongly impacts whether and to what"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.12532","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/2504.12532/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":"2504.12532","created_at":"2026-07-05T10:50:17.902128+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.12532v1","created_at":"2026-07-05T10:50:17.902128+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.12532","created_at":"2026-07-05T10:50:17.902128+00:00"},{"alias_kind":"pith_short_12","alias_value":"LGCXGK3HSPCZ","created_at":"2026-07-05T10:50:17.902128+00:00"},{"alias_kind":"pith_short_16","alias_value":"LGCXGK3HSPCZSHPF","created_at":"2026-07-05T10:50:17.902128+00:00"},{"alias_kind":"pith_short_8","alias_value":"LGCXGK3H","created_at":"2026-07-05T10:50:17.902128+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05196","citing_title":"Unified Audio Intelligence Without Regressing on Text Intelligence","ref_index":169,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10019","citing_title":"The two clocks and the innovation window: When and how generative models learn rules","ref_index":118,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A","json":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A.json","graph_json":"https://pith.science/api/pith-number/LGCXGK3HSPCZSHPFQUFB4LW62A/graph.json","events_json":"https://pith.science/api/pith-number/LGCXGK3HSPCZSHPFQUFB4LW62A/events.json","paper":"https://pith.science/paper/LGCXGK3H"},"agent_actions":{"view_html":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A","download_json":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A.json","view_paper":"https://pith.science/paper/LGCXGK3H","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.12532&json=true","fetch_graph":"https://pith.science/api/pith-number/LGCXGK3HSPCZSHPFQUFB4LW62A/graph.json","fetch_events":"https://pith.science/api/pith-number/LGCXGK3HSPCZSHPFQUFB4LW62A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A/action/storage_attestation","attest_author":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A/action/author_attestation","sign_citation":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A/action/citation_signature","submit_replication":"https://pith.science/pith/LGCXGK3HSPCZSHPFQUFB4LW62A/action/replication_record"}},"created_at":"2026-07-05T10:50:17.902128+00:00","updated_at":"2026-07-05T10:50:17.902128+00:00"}