{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:RFRXM2NFXZTJDX7YZ3QMVZQNFV","short_pith_number":"pith:RFRXM2NF","schema_version":"1.0","canonical_sha256":"89637669a5be6691dff8cee0cae60d2d7a0ef572a8f154ad56bfaa3aab91c52e","source":{"kind":"arxiv","id":"2302.13861","version":1},"attestation_state":"computed","paper":{"title":"Differentially Private Diffusion Models Generate Useful Synthetic Images","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Borja Balle, Ira Ktena, Jamie Hayes, Leonard Berrada, Olivia Wiles, Robert Stanforth, Sahra Ghalebikesabi, Samuel L. Smith, Soham De, Sven Gowal","submitted_at":"2023-02-27T15:02:04Z","abstract_excerpt":"The ability to generate privacy-preserving synthetic versions of sensitive image datasets could unlock numerous ML applications currently constrained by data availability. Due to their astonishing image generation quality, diffusion models are a prime candidate for generating high-quality synthetic data. However, recent studies have found that, by default, the outputs of some diffusion models do not preserve training data privacy. By privately fine-tuning ImageNet pre-trained diffusion models with more than 80M parameters, we obtain SOTA results on CIFAR-10 and Camelyon17 in terms of both FID "},"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":"2302.13861","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-02-27T15:02:04Z","cross_cats_sorted":["cs.CR","cs.CV","stat.ML"],"title_canon_sha256":"3e063ac0997146eb137f8097a25ce322f1e781bd80dab2d9bafac1e479999851","abstract_canon_sha256":"178cf5704199ee7cc37b086c74fd894f7c3660887adff4b6eb8211436c5eb595"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:45:51.775879Z","signature_b64":"mpnoevtTve57+oLgfwxCmaFne8TRnbRm+XpM3X+4l6bq4quxrO79Alc3kjt75vNkxiAuwAxxooCbzO834Xl1BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"89637669a5be6691dff8cee0cae60d2d7a0ef572a8f154ad56bfaa3aab91c52e","last_reissued_at":"2026-07-05T05:45:51.775432Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:45:51.775432Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Differentially Private Diffusion Models Generate Useful Synthetic Images","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Borja Balle, Ira Ktena, Jamie Hayes, Leonard Berrada, Olivia Wiles, Robert Stanforth, Sahra Ghalebikesabi, Samuel L. Smith, Soham De, Sven Gowal","submitted_at":"2023-02-27T15:02:04Z","abstract_excerpt":"The ability to generate privacy-preserving synthetic versions of sensitive image datasets could unlock numerous ML applications currently constrained by data availability. Due to their astonishing image generation quality, diffusion models are a prime candidate for generating high-quality synthetic data. However, recent studies have found that, by default, the outputs of some diffusion models do not preserve training data privacy. By privately fine-tuning ImageNet pre-trained diffusion models with more than 80M parameters, we obtain SOTA results on CIFAR-10 and Camelyon17 in terms of both FID "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2302.13861","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/2302.13861/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":"2302.13861","created_at":"2026-07-05T05:45:51.775489+00:00"},{"alias_kind":"arxiv_version","alias_value":"2302.13861v1","created_at":"2026-07-05T05:45:51.775489+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2302.13861","created_at":"2026-07-05T05:45:51.775489+00:00"},{"alias_kind":"pith_short_12","alias_value":"RFRXM2NFXZTJ","created_at":"2026-07-05T05:45:51.775489+00:00"},{"alias_kind":"pith_short_16","alias_value":"RFRXM2NFXZTJDX7Y","created_at":"2026-07-05T05:45:51.775489+00:00"},{"alias_kind":"pith_short_8","alias_value":"RFRXM2NF","created_at":"2026-07-05T05:45:51.775489+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06616","citing_title":"WHERE to Generate Matters: Budget-Aware Synthetic Augmentation for Label Skewed Federated Learning","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20897","citing_title":"PeerCheck: Enhancing LLM-Generated Academic Reviews Towards Human-Level Quality","ref_index":87,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30312","citing_title":"DP-SAPF: Saliency-Aware Parameter Fine-tuning of Public Models for Differentially Private Image Synthesis","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2406.14429","citing_title":"CollaFuse: Collaborative Diffusion Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2411.04680","citing_title":"Privacy Leakage via Output Label Space and Differentially Private Continual Learning","ref_index":47,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV","json":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV.json","graph_json":"https://pith.science/api/pith-number/RFRXM2NFXZTJDX7YZ3QMVZQNFV/graph.json","events_json":"https://pith.science/api/pith-number/RFRXM2NFXZTJDX7YZ3QMVZQNFV/events.json","paper":"https://pith.science/paper/RFRXM2NF"},"agent_actions":{"view_html":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV","download_json":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV.json","view_paper":"https://pith.science/paper/RFRXM2NF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2302.13861&json=true","fetch_graph":"https://pith.science/api/pith-number/RFRXM2NFXZTJDX7YZ3QMVZQNFV/graph.json","fetch_events":"https://pith.science/api/pith-number/RFRXM2NFXZTJDX7YZ3QMVZQNFV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV/action/storage_attestation","attest_author":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV/action/author_attestation","sign_citation":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV/action/citation_signature","submit_replication":"https://pith.science/pith/RFRXM2NFXZTJDX7YZ3QMVZQNFV/action/replication_record"}},"created_at":"2026-07-05T05:45:51.775489+00:00","updated_at":"2026-07-05T05:45:51.775489+00:00"}