{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:R7XGCYSGGNMZUQHPP7DXFXTVR4","short_pith_number":"pith:R7XGCYSG","schema_version":"1.0","canonical_sha256":"8fee61624633599a40ef7fc772de758f034bc8aec35c16d0d4af7cc8eb633b86","source":{"kind":"arxiv","id":"2403.17673","version":2},"attestation_state":"computed","paper":{"title":"How Private are DP-SGD Implementations?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.DS"],"primary_cat":"cs.LG","authors_text":"Amer Sinha, Badih Ghazi, Chiyuan Zhang, Lynn Chua, Pasin Manurangsi, Pritish Kamath, Ravi Kumar","submitted_at":"2024-03-26T13:02:43Z","abstract_excerpt":"We demonstrate a substantial gap between the privacy guarantees of the Adaptive Batch Linear Queries (ABLQ) mechanism under different types of batch sampling: (i) Shuffling, and (ii) Poisson subsampling; the typical analysis of Differentially Private Stochastic Gradient Descent (DP-SGD) follows by interpreting it as a post-processing of ABLQ. While shuffling-based DP-SGD is more commonly used in practical implementations, it has not been amenable to easy privacy analysis, either analytically or even numerically. On the other hand, Poisson subsampling-based DP-SGD is challenging to scalably imp"},"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":"2403.17673","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-26T13:02:43Z","cross_cats_sorted":["cs.CR","cs.DS"],"title_canon_sha256":"ad68a9b78e747b9452d08e147ab76f91828dfbddcff0d8284affe89dc930df66","abstract_canon_sha256":"cb3daece8867a3d0804b4ba1c5de5a7cf0f9d5b9bf1d600c0df38a4aca9bf7b3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:27.348392Z","signature_b64":"4hKGzIAFwNi0U4aT8Giw3xjft0uPXi7RpXo4CjShGq3XMhYCf7t9j0XQXihK9jJedUPHZ4th0mqMNTuSmwTxCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8fee61624633599a40ef7fc772de758f034bc8aec35c16d0d4af7cc8eb633b86","last_reissued_at":"2026-07-05T08:28:27.347682Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:27.347682Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Private are DP-SGD Implementations?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.DS"],"primary_cat":"cs.LG","authors_text":"Amer Sinha, Badih Ghazi, Chiyuan Zhang, Lynn Chua, Pasin Manurangsi, Pritish Kamath, Ravi Kumar","submitted_at":"2024-03-26T13:02:43Z","abstract_excerpt":"We demonstrate a substantial gap between the privacy guarantees of the Adaptive Batch Linear Queries (ABLQ) mechanism under different types of batch sampling: (i) Shuffling, and (ii) Poisson subsampling; the typical analysis of Differentially Private Stochastic Gradient Descent (DP-SGD) follows by interpreting it as a post-processing of ABLQ. While shuffling-based DP-SGD is more commonly used in practical implementations, it has not been amenable to easy privacy analysis, either analytically or even numerically. On the other hand, Poisson subsampling-based DP-SGD is challenging to scalably imp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.17673","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/2403.17673/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":"2403.17673","created_at":"2026-07-05T08:28:27.347771+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.17673v2","created_at":"2026-07-05T08:28:27.347771+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.17673","created_at":"2026-07-05T08:28:27.347771+00:00"},{"alias_kind":"pith_short_12","alias_value":"R7XGCYSGGNMZ","created_at":"2026-07-05T08:28:27.347771+00:00"},{"alias_kind":"pith_short_16","alias_value":"R7XGCYSGGNMZUQHP","created_at":"2026-07-05T08:28:27.347771+00:00"},{"alias_kind":"pith_short_8","alias_value":"R7XGCYSG","created_at":"2026-07-05T08:28:27.347771+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.08201","citing_title":"Correlated Noise Mechanisms for Differentially Private Learning","ref_index":6,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4","json":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4.json","graph_json":"https://pith.science/api/pith-number/R7XGCYSGGNMZUQHPP7DXFXTVR4/graph.json","events_json":"https://pith.science/api/pith-number/R7XGCYSGGNMZUQHPP7DXFXTVR4/events.json","paper":"https://pith.science/paper/R7XGCYSG"},"agent_actions":{"view_html":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4","download_json":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4.json","view_paper":"https://pith.science/paper/R7XGCYSG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.17673&json=true","fetch_graph":"https://pith.science/api/pith-number/R7XGCYSGGNMZUQHPP7DXFXTVR4/graph.json","fetch_events":"https://pith.science/api/pith-number/R7XGCYSGGNMZUQHPP7DXFXTVR4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4/action/storage_attestation","attest_author":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4/action/author_attestation","sign_citation":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4/action/citation_signature","submit_replication":"https://pith.science/pith/R7XGCYSGGNMZUQHPP7DXFXTVR4/action/replication_record"}},"created_at":"2026-07-05T08:28:27.347771+00:00","updated_at":"2026-07-05T08:28:27.347771+00:00"}