{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:J4NPN3DHLZYYW236C26QTJWF2D","short_pith_number":"pith:J4NPN3DH","schema_version":"1.0","canonical_sha256":"4f1af6ec675e718b6b7e16bd09a6c5d0f06a6316fbd813172880ca507a17623c","source":{"kind":"arxiv","id":"2403.02051","version":2},"attestation_state":"computed","paper":{"title":"Privacy of SGD under Gaussian or Heavy-Tailed Noise: Guarantees without Gradient Clipping","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG","math.ST","stat.TH"],"primary_cat":"stat.ML","authors_text":"Lingjiong Zhu, Mert G\\\"urb\\\"uzbalaban, Sinan Y{\\i}ld{\\i}r{\\i}m, Umut \\c{S}im\\c{s}ekli","submitted_at":"2024-03-04T13:53:41Z","abstract_excerpt":"The injection of heavy-tailed noise into the iterates of stochastic gradient descent (SGD) has garnered growing interest in recent years due to its theoretical and empirical benefits for optimization and generalization. However, its implications for privacy preservation remain largely unexplored. Aiming to bridge this gap, we provide differential privacy (DP) guarantees for noisy SGD, when the injected noise follows an $\\alpha$-stable distribution, which includes a spectrum of heavy-tailed distributions (with infinite variance) as well as the light-tailed Gaussian distribution. Considering the"},"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":"2403.02051","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2024-03-04T13:53:41Z","cross_cats_sorted":["cs.CR","cs.LG","math.ST","stat.TH"],"title_canon_sha256":"5a2651f1297db43253cb47431fdb710299a0e2fc4d695751786fdee2d10f0dda","abstract_canon_sha256":"91f4e69ca1d6d0abf01b08aaca1bbbb4f776b1ff6151eb06f6faa2320480edf1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:01:23.418679Z","signature_b64":"+Pni1TpDMyTx91CZe4Ocwg5rmpsWSy+HAswgxCwKjSaF2RrJgguAY+wdk7UXihSRVGGHABtDdO49ZR2cf1DGBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4f1af6ec675e718b6b7e16bd09a6c5d0f06a6316fbd813172880ca507a17623c","last_reissued_at":"2026-07-05T11:01:23.418201Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:01:23.418201Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Privacy of SGD under Gaussian or Heavy-Tailed Noise: Guarantees without Gradient Clipping","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG","math.ST","stat.TH"],"primary_cat":"stat.ML","authors_text":"Lingjiong Zhu, Mert G\\\"urb\\\"uzbalaban, Sinan Y{\\i}ld{\\i}r{\\i}m, Umut \\c{S}im\\c{s}ekli","submitted_at":"2024-03-04T13:53:41Z","abstract_excerpt":"The injection of heavy-tailed noise into the iterates of stochastic gradient descent (SGD) has garnered growing interest in recent years due to its theoretical and empirical benefits for optimization and generalization. However, its implications for privacy preservation remain largely unexplored. Aiming to bridge this gap, we provide differential privacy (DP) guarantees for noisy SGD, when the injected noise follows an $\\alpha$-stable distribution, which includes a spectrum of heavy-tailed distributions (with infinite variance) as well as the light-tailed Gaussian distribution. Considering the"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.02051","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/2403.02051/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":"2403.02051","created_at":"2026-07-05T11:01:23.418255+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.02051v2","created_at":"2026-07-05T11:01:23.418255+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.02051","created_at":"2026-07-05T11:01:23.418255+00:00"},{"alias_kind":"pith_short_12","alias_value":"J4NPN3DHLZYY","created_at":"2026-07-05T11:01:23.418255+00:00"},{"alias_kind":"pith_short_16","alias_value":"J4NPN3DHLZYYW236","created_at":"2026-07-05T11:01:23.418255+00:00"},{"alias_kind":"pith_short_8","alias_value":"J4NPN3DH","created_at":"2026-07-05T11:01:23.418255+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.00885","citing_title":"Algorithmic Stability of Stochastic Gradient Descent with Momentum under Heavy-Tailed Noise","ref_index":50,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D","json":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D.json","graph_json":"https://pith.science/api/pith-number/J4NPN3DHLZYYW236C26QTJWF2D/graph.json","events_json":"https://pith.science/api/pith-number/J4NPN3DHLZYYW236C26QTJWF2D/events.json","paper":"https://pith.science/paper/J4NPN3DH"},"agent_actions":{"view_html":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D","download_json":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D.json","view_paper":"https://pith.science/paper/J4NPN3DH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.02051&json=true","fetch_graph":"https://pith.science/api/pith-number/J4NPN3DHLZYYW236C26QTJWF2D/graph.json","fetch_events":"https://pith.science/api/pith-number/J4NPN3DHLZYYW236C26QTJWF2D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D/action/storage_attestation","attest_author":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D/action/author_attestation","sign_citation":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D/action/citation_signature","submit_replication":"https://pith.science/pith/J4NPN3DHLZYYW236C26QTJWF2D/action/replication_record"}},"created_at":"2026-07-05T11:01:23.418255+00:00","updated_at":"2026-07-05T11:01:23.418255+00:00"}