{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PLHS2GO23H3ZXEQFZVDAEWYOGD","short_pith_number":"pith:PLHS2GO2","schema_version":"1.0","canonical_sha256":"7acf2d19dad9f79b9205cd46025b0e30cfade55a417301f524d1bc76ea4a6ce1","source":{"kind":"arxiv","id":"2412.20292","version":2},"attestation_state":"computed","paper":{"title":"An analytic theory of creativity in convolutional diffusion models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.AI","q-bio.NC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Mason Kamb, Surya Ganguli","submitted_at":"2024-12-28T22:33:29Z","abstract_excerpt":"We obtain an analytic, interpretable and predictive theory of creativity in convolutional diffusion models. Indeed, score-matching diffusion models can generate highly original images that lie far from their training data. However, optimal score-matching theory suggests that these models should only be able to produce memorized training examples. To reconcile this theory-experiment gap, we identify two simple inductive biases, locality and equivariance, that: (1) induce a form of combinatorial creativity by preventing optimal score-matching; (2) result in fully analytic, completely mechanistic"},"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":"2412.20292","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-12-28T22:33:29Z","cross_cats_sorted":["cond-mat.dis-nn","cs.AI","q-bio.NC","stat.ML"],"title_canon_sha256":"a5e0910d50154fb2668aa608966ec12a66c1bd162cb14ec7287b2876d99eb769","abstract_canon_sha256":"cb00d0842d4be2ebf32b43a14c5c903dd1673efd233a4de4b893bd843f550bc1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:05.095778Z","signature_b64":"LeyqFtPrRX3x3NqDcmV+GCbg8YPKN9ONn0qR3vTojNsuM/yrATTuzvDM8daLuwi5RwuGk3HEo7+Re5tQewpEBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7acf2d19dad9f79b9205cd46025b0e30cfade55a417301f524d1bc76ea4a6ce1","last_reissued_at":"2026-07-05T11:16:05.094816Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:05.094816Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"An analytic theory of creativity in convolutional diffusion models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.AI","q-bio.NC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Mason Kamb, Surya Ganguli","submitted_at":"2024-12-28T22:33:29Z","abstract_excerpt":"We obtain an analytic, interpretable and predictive theory of creativity in convolutional diffusion models. Indeed, score-matching diffusion models can generate highly original images that lie far from their training data. However, optimal score-matching theory suggests that these models should only be able to produce memorized training examples. To reconcile this theory-experiment gap, we identify two simple inductive biases, locality and equivariance, that: (1) induce a form of combinatorial creativity by preventing optimal score-matching; (2) result in fully analytic, completely mechanistic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.20292","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/2412.20292/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":"2412.20292","created_at":"2026-07-05T11:16:05.094925+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.20292v2","created_at":"2026-07-05T11:16:05.094925+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.20292","created_at":"2026-07-05T11:16:05.094925+00:00"},{"alias_kind":"pith_short_12","alias_value":"PLHS2GO23H3Z","created_at":"2026-07-05T11:16:05.094925+00:00"},{"alias_kind":"pith_short_16","alias_value":"PLHS2GO23H3ZXEQF","created_at":"2026-07-05T11:16:05.094925+00:00"},{"alias_kind":"pith_short_8","alias_value":"PLHS2GO2","created_at":"2026-07-05T11:16:05.094925+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06624","citing_title":"Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00273","citing_title":"When Do Diffusion Models learn to Generate Multiple Objects?","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01238","citing_title":"Training-Free Imitation Learning with Closed-Form Diffusion Policies","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19594","citing_title":"Unsupervised Causal Abstractions Discovery","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2503.03206","citing_title":"An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2603.13419","citing_title":"Diffusion Models Memorize in Training -- and Generalize in Inference","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20299","citing_title":"Mechanisms of Misgeneralization in Physical Sequence Modeling","ref_index":106,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16348","citing_title":"Flow-Direct: Feedback-Efficient and Reusable Guidance for Flow Models via Non-Parametric Guidance Field","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2508.06614","citing_title":"Local Diffusion Models and Phases of Data Distributions","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2509.25003","citing_title":"Score-based Membership Inference on Diffusion Models","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2512.11415","citing_title":"Emergence of Nonequilibrium Latent Cycles in Unsupervised Generative Modeling","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26841","citing_title":"Language Diffusion Models are Associative Memories Capable of Retrieving Unseen Data","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04830","citing_title":"Concurrence of Symmetry Breaking and Nonlocality Phase Transitions in Diffusion Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09979","citing_title":"A Minimal Model of Representation Collapse: Frustration, Stop-Gradient, and Dynamics","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15306","citing_title":"Generalization in LLM Problem Solving: The Case of the Shortest Path","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD","json":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD.json","graph_json":"https://pith.science/api/pith-number/PLHS2GO23H3ZXEQFZVDAEWYOGD/graph.json","events_json":"https://pith.science/api/pith-number/PLHS2GO23H3ZXEQFZVDAEWYOGD/events.json","paper":"https://pith.science/paper/PLHS2GO2"},"agent_actions":{"view_html":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD","download_json":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD.json","view_paper":"https://pith.science/paper/PLHS2GO2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.20292&json=true","fetch_graph":"https://pith.science/api/pith-number/PLHS2GO23H3ZXEQFZVDAEWYOGD/graph.json","fetch_events":"https://pith.science/api/pith-number/PLHS2GO23H3ZXEQFZVDAEWYOGD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD/action/storage_attestation","attest_author":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD/action/author_attestation","sign_citation":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD/action/citation_signature","submit_replication":"https://pith.science/pith/PLHS2GO23H3ZXEQFZVDAEWYOGD/action/replication_record"}},"created_at":"2026-07-05T11:16:05.094925+00:00","updated_at":"2026-07-05T11:16:05.094925+00:00"}