{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VZBY4UEXHEHMKXN2VMKFQI2GW2","short_pith_number":"pith:VZBY4UEX","schema_version":"1.0","canonical_sha256":"ae438e5097390ec55dbaab14582346b697197ea8ba89612a038e0d556aa03553","source":{"kind":"arxiv","id":"2507.10536","version":1},"attestation_state":"computed","paper":{"title":"On the Performance of Differentially Private Optimization with Heavy-Tail Class Imbalance","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alain Zhiyanov, Mathias L\\'ecuyer, Qiaoyue Tang","submitted_at":"2025-07-14T17:57:08Z","abstract_excerpt":"In this work, we analyze the optimization behaviour of common private learning optimization algorithms under heavy-tail class imbalanced distribution. We show that, in a stylized model, optimizing with Gradient Descent with differential privacy (DP-GD) suffers when learning low-frequency classes, whereas optimization algorithms that estimate second-order information do not. In particular, DP-AdamBC that removes the DP bias from estimating loss curvature is a crucial component to avoid the ill-condition caused by heavy-tail class imbalance, and empirically fits the data better with $\\approx8\\%$"},"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":"2507.10536","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-07-14T17:57:08Z","cross_cats_sorted":[],"title_canon_sha256":"3834122fa9f8ce04544088ba47bd4eb824e2d5dac1fbd2bf5ddf974e5eca24f2","abstract_canon_sha256":"8e19d6979fd15223daaa3c86c34092bd51ee5514ae0a841ae824657c1443f68b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:36:53.242268Z","signature_b64":"pmq3/QhwzIzcmxGsSSTpaY1TzSgqEYDQ1PAIhRYOpUQ5ysmNPtPbUgYPmXNp2G3AFTLTQ2JAa5D6EeYl7G1mBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ae438e5097390ec55dbaab14582346b697197ea8ba89612a038e0d556aa03553","last_reissued_at":"2026-07-05T11:36:53.241647Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:36:53.241647Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Performance of Differentially Private Optimization with Heavy-Tail Class Imbalance","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alain Zhiyanov, Mathias L\\'ecuyer, Qiaoyue Tang","submitted_at":"2025-07-14T17:57:08Z","abstract_excerpt":"In this work, we analyze the optimization behaviour of common private learning optimization algorithms under heavy-tail class imbalanced distribution. We show that, in a stylized model, optimizing with Gradient Descent with differential privacy (DP-GD) suffers when learning low-frequency classes, whereas optimization algorithms that estimate second-order information do not. In particular, DP-AdamBC that removes the DP bias from estimating loss curvature is a crucial component to avoid the ill-condition caused by heavy-tail class imbalance, and empirically fits the data better with $\\approx8\\%$"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.10536","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/2507.10536/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":"2507.10536","created_at":"2026-07-05T11:36:53.241707+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.10536v1","created_at":"2026-07-05T11:36:53.241707+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.10536","created_at":"2026-07-05T11:36:53.241707+00:00"},{"alias_kind":"pith_short_12","alias_value":"VZBY4UEXHEHM","created_at":"2026-07-05T11:36:53.241707+00:00"},{"alias_kind":"pith_short_16","alias_value":"VZBY4UEXHEHMKXN2","created_at":"2026-07-05T11:36:53.241707+00:00"},{"alias_kind":"pith_short_8","alias_value":"VZBY4UEX","created_at":"2026-07-05T11:36:53.241707+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2","json":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2.json","graph_json":"https://pith.science/api/pith-number/VZBY4UEXHEHMKXN2VMKFQI2GW2/graph.json","events_json":"https://pith.science/api/pith-number/VZBY4UEXHEHMKXN2VMKFQI2GW2/events.json","paper":"https://pith.science/paper/VZBY4UEX"},"agent_actions":{"view_html":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2","download_json":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2.json","view_paper":"https://pith.science/paper/VZBY4UEX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.10536&json=true","fetch_graph":"https://pith.science/api/pith-number/VZBY4UEXHEHMKXN2VMKFQI2GW2/graph.json","fetch_events":"https://pith.science/api/pith-number/VZBY4UEXHEHMKXN2VMKFQI2GW2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2/action/storage_attestation","attest_author":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2/action/author_attestation","sign_citation":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2/action/citation_signature","submit_replication":"https://pith.science/pith/VZBY4UEXHEHMKXN2VMKFQI2GW2/action/replication_record"}},"created_at":"2026-07-05T11:36:53.241707+00:00","updated_at":"2026-07-05T11:36:53.241707+00:00"}