{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:O7Y2R2CY7XPIJQ46CKNZDOOLNP","short_pith_number":"pith:O7Y2R2CY","schema_version":"1.0","canonical_sha256":"77f1a8e858fdde84c39e129b91b9cb6be7ebca71bff729d3565c3766bc95dbd5","source":{"kind":"arxiv","id":"2308.12018","version":1},"attestation_state":"computed","paper":{"title":"Bias-Aware Minimisation: Understanding and Mitigating Estimator Bias in Private SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Alexander Ziller, Daniel Rueckert, Georgios Kaissis, Moritz Knolle, Robert Dorfman","submitted_at":"2023-08-23T09:20:41Z","abstract_excerpt":"Differentially private SGD (DP-SGD) holds the promise of enabling the safe and responsible application of machine learning to sensitive datasets. However, DP-SGD only provides a biased, noisy estimate of a mini-batch gradient. This renders optimisation steps less effective and limits model utility as a result. With this work, we show a connection between per-sample gradient norms and the estimation bias of the private gradient oracle used in DP-SGD. Here, we propose Bias-Aware Minimisation (BAM) that allows for the provable reduction of private gradient estimator bias. We show how to efficient"},"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":"2308.12018","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-08-23T09:20:41Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"d4a607ec55c9d8605a9803146a53d1de50055ee62e1a60a4180a80472279e7b7","abstract_canon_sha256":"0146a6bfd87b04e6d3200c14dafb920389b7aafb968ff0076421e2cb7eeb74fa"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:43:58.243365Z","signature_b64":"5BDqoBrdUmBUYgH6f3QFwDlQnXtnUcGkMkKc39wEK3K+0pcFq3k6tcQnMDZHbgUUA6kjEZVaNtB0AcmGSGPGDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"77f1a8e858fdde84c39e129b91b9cb6be7ebca71bff729d3565c3766bc95dbd5","last_reissued_at":"2026-07-05T06:43:58.242914Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:43:58.242914Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Bias-Aware Minimisation: Understanding and Mitigating Estimator Bias in Private SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Alexander Ziller, Daniel Rueckert, Georgios Kaissis, Moritz Knolle, Robert Dorfman","submitted_at":"2023-08-23T09:20:41Z","abstract_excerpt":"Differentially private SGD (DP-SGD) holds the promise of enabling the safe and responsible application of machine learning to sensitive datasets. However, DP-SGD only provides a biased, noisy estimate of a mini-batch gradient. This renders optimisation steps less effective and limits model utility as a result. With this work, we show a connection between per-sample gradient norms and the estimation bias of the private gradient oracle used in DP-SGD. Here, we propose Bias-Aware Minimisation (BAM) that allows for the provable reduction of private gradient estimator bias. We show how to efficient"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2308.12018","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/2308.12018/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":"2308.12018","created_at":"2026-07-05T06:43:58.242969+00:00"},{"alias_kind":"arxiv_version","alias_value":"2308.12018v1","created_at":"2026-07-05T06:43:58.242969+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2308.12018","created_at":"2026-07-05T06:43:58.242969+00:00"},{"alias_kind":"pith_short_12","alias_value":"O7Y2R2CY7XPI","created_at":"2026-07-05T06:43:58.242969+00:00"},{"alias_kind":"pith_short_16","alias_value":"O7Y2R2CY7XPIJQ46","created_at":"2026-07-05T06:43:58.242969+00:00"},{"alias_kind":"pith_short_8","alias_value":"O7Y2R2CY","created_at":"2026-07-05T06:43:58.242969+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07930","citing_title":"INO-SGD: Addressing Utility Imbalance under Individualized Differential Privacy","ref_index":102,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP","json":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP.json","graph_json":"https://pith.science/api/pith-number/O7Y2R2CY7XPIJQ46CKNZDOOLNP/graph.json","events_json":"https://pith.science/api/pith-number/O7Y2R2CY7XPIJQ46CKNZDOOLNP/events.json","paper":"https://pith.science/paper/O7Y2R2CY"},"agent_actions":{"view_html":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP","download_json":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP.json","view_paper":"https://pith.science/paper/O7Y2R2CY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2308.12018&json=true","fetch_graph":"https://pith.science/api/pith-number/O7Y2R2CY7XPIJQ46CKNZDOOLNP/graph.json","fetch_events":"https://pith.science/api/pith-number/O7Y2R2CY7XPIJQ46CKNZDOOLNP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP/action/storage_attestation","attest_author":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP/action/author_attestation","sign_citation":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP/action/citation_signature","submit_replication":"https://pith.science/pith/O7Y2R2CY7XPIJQ46CKNZDOOLNP/action/replication_record"}},"created_at":"2026-07-05T06:43:58.242969+00:00","updated_at":"2026-07-05T06:43:58.242969+00:00"}