{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:FKGZ2LNXE7OP7R4KKEJPTUKT5F","short_pith_number":"pith:FKGZ2LNX","schema_version":"1.0","canonical_sha256":"2a8d9d2db727dcffc78a5112f9d153e961178a5ad04b18bcdc23b26fc1e121bf","source":{"kind":"arxiv","id":"2111.00173","version":3},"attestation_state":"computed","paper":{"title":"Dynamic Differential-Privacy Preserving SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Jian Du, Mingyi Hong, Siheng Chen, Song Li, Xiangyi Chen","submitted_at":"2021-10-30T04:45:11Z","abstract_excerpt":"The vanilla Differentially-Private Stochastic Gradient Descent (DP-SGD), including DP-Adam and other variants, ensures the privacy of training data by uniformly distributing privacy costs across training steps. The equivalent privacy costs controlled by maintaining the same gradient clipping thresholds and noise powers in each step result in unstable updates and a lower model accuracy when compared to the non-DP counterpart. In this paper, we propose the dynamic DP-SGD (along with dynamic DP-Adam, and others) to reduce the performance loss gap while maintaining privacy by dynamically adjusting"},"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":"2111.00173","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-10-30T04:45:11Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"250b41b0ba116426ac6f66dfe41be087652e820e69adacd9538f8c07a830d63c","abstract_canon_sha256":"9bcae157215a8ce2f33add81e2147156ead59d8aa3b726cff0f8f0422d9bb202"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:48:46.694325Z","signature_b64":"aPcTZA0D7JPJKM5oNILG+/xANHPE8K+HhThomzmBm9Gf/RQEMYh1R5roNIuGve3QnibF4n7E75hZ1MpD9dqYAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2a8d9d2db727dcffc78a5112f9d153e961178a5ad04b18bcdc23b26fc1e121bf","last_reissued_at":"2026-07-05T03:48:46.693884Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:48:46.693884Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dynamic Differential-Privacy Preserving SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Jian Du, Mingyi Hong, Siheng Chen, Song Li, Xiangyi Chen","submitted_at":"2021-10-30T04:45:11Z","abstract_excerpt":"The vanilla Differentially-Private Stochastic Gradient Descent (DP-SGD), including DP-Adam and other variants, ensures the privacy of training data by uniformly distributing privacy costs across training steps. The equivalent privacy costs controlled by maintaining the same gradient clipping thresholds and noise powers in each step result in unstable updates and a lower model accuracy when compared to the non-DP counterpart. In this paper, we propose the dynamic DP-SGD (along with dynamic DP-Adam, and others) to reduce the performance loss gap while maintaining privacy by dynamically adjusting"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2111.00173","kind":"arxiv","version":3},"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/2111.00173/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":"2111.00173","created_at":"2026-07-05T03:48:46.693940+00:00"},{"alias_kind":"arxiv_version","alias_value":"2111.00173v3","created_at":"2026-07-05T03:48:46.693940+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2111.00173","created_at":"2026-07-05T03:48:46.693940+00:00"},{"alias_kind":"pith_short_12","alias_value":"FKGZ2LNXE7OP","created_at":"2026-07-05T03:48:46.693940+00:00"},{"alias_kind":"pith_short_16","alias_value":"FKGZ2LNXE7OP7R4K","created_at":"2026-07-05T03:48:46.693940+00:00"},{"alias_kind":"pith_short_8","alias_value":"FKGZ2LNX","created_at":"2026-07-05T03:48:46.693940+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.03472","citing_title":"DPQuant: Efficient and Differentially-Private Model Training via Dynamic Quantization Scheduling","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10272","citing_title":"DP-LAC: Lightweight Adaptive Clipping for Differentially Private Federated Fine-tuning of Language Models","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F","json":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F.json","graph_json":"https://pith.science/api/pith-number/FKGZ2LNXE7OP7R4KKEJPTUKT5F/graph.json","events_json":"https://pith.science/api/pith-number/FKGZ2LNXE7OP7R4KKEJPTUKT5F/events.json","paper":"https://pith.science/paper/FKGZ2LNX"},"agent_actions":{"view_html":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F","download_json":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F.json","view_paper":"https://pith.science/paper/FKGZ2LNX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2111.00173&json=true","fetch_graph":"https://pith.science/api/pith-number/FKGZ2LNXE7OP7R4KKEJPTUKT5F/graph.json","fetch_events":"https://pith.science/api/pith-number/FKGZ2LNXE7OP7R4KKEJPTUKT5F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F/action/storage_attestation","attest_author":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F/action/author_attestation","sign_citation":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F/action/citation_signature","submit_replication":"https://pith.science/pith/FKGZ2LNXE7OP7R4KKEJPTUKT5F/action/replication_record"}},"created_at":"2026-07-05T03:48:46.693940+00:00","updated_at":"2026-07-05T03:48:46.693940+00:00"}