{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:6IDSR27M2LNDIX6PSOOGDBAJM4","short_pith_number":"pith:6IDSR27M","schema_version":"1.0","canonical_sha256":"f20728ebecd2da345fcf939c6184096707041293818040fd25c00a06908c8da6","source":{"kind":"arxiv","id":"2110.06500","version":2},"attestation_state":"computed","paper":{"title":"Differentially Private Fine-tuning of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andre Manoel, Arturs Backurs, Da Yu, Gautam Kamath, Huishuai Zhang, Huseyin A. Inan, Janardhan Kulkarni, Lukas Wutschitz, Saurabh Naik, Sergey Yekhanin, Sivakanth Gopi, Yin Tat Lee","submitted_at":"2021-10-13T05:15:00Z","abstract_excerpt":"We give simpler, sparser, and faster algorithms for differentially private fine-tuning of large-scale pre-trained language models, which achieve the state-of-the-art privacy versus utility tradeoffs on many standard NLP tasks. We propose a meta-framework for this problem, inspired by the recent success of highly parameter-efficient methods for fine-tuning. Our experiments show that differentially private adaptations of these approaches outperform previous private algorithms in three important dimensions: utility, privacy, and the computational and memory cost of private training. On many commo"},"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":"2110.06500","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-10-13T05:15:00Z","cross_cats_sorted":["cs.CL","cs.CR","stat.ML"],"title_canon_sha256":"32b031c12323bc88d8fa8b7e39c580e40deb78bb92c84b15fac207c7b68bffc4","abstract_canon_sha256":"a338fc6052a62c406de4668aae52184f5f836db925fd8388f142c6c6ec84ea48"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:40:24.064767Z","signature_b64":"8GpoeJlR1K5u/E6xbbEYfMeOrfA9GCBJb936rpqilyJAyIXLdvJj3zJx5ccI2MzDGvKnGRNz0qCc9xT9BPzFAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f20728ebecd2da345fcf939c6184096707041293818040fd25c00a06908c8da6","last_reissued_at":"2026-07-05T04:40:24.064291Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:40:24.064291Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Differentially Private Fine-tuning of Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andre Manoel, Arturs Backurs, Da Yu, Gautam Kamath, Huishuai Zhang, Huseyin A. Inan, Janardhan Kulkarni, Lukas Wutschitz, Saurabh Naik, Sergey Yekhanin, Sivakanth Gopi, Yin Tat Lee","submitted_at":"2021-10-13T05:15:00Z","abstract_excerpt":"We give simpler, sparser, and faster algorithms for differentially private fine-tuning of large-scale pre-trained language models, which achieve the state-of-the-art privacy versus utility tradeoffs on many standard NLP tasks. We propose a meta-framework for this problem, inspired by the recent success of highly parameter-efficient methods for fine-tuning. Our experiments show that differentially private adaptations of these approaches outperform previous private algorithms in three important dimensions: utility, privacy, and the computational and memory cost of private training. On many commo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2110.06500","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/2110.06500/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":"2110.06500","created_at":"2026-07-05T04:40:24.064361+00:00"},{"alias_kind":"arxiv_version","alias_value":"2110.06500v2","created_at":"2026-07-05T04:40:24.064361+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2110.06500","created_at":"2026-07-05T04:40:24.064361+00:00"},{"alias_kind":"pith_short_12","alias_value":"6IDSR27M2LND","created_at":"2026-07-05T04:40:24.064361+00:00"},{"alias_kind":"pith_short_16","alias_value":"6IDSR27M2LNDIX6P","created_at":"2026-07-05T04:40:24.064361+00:00"},{"alias_kind":"pith_short_8","alias_value":"6IDSR27M","created_at":"2026-07-05T04:40:24.064361+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22153","citing_title":"$\\pi$-RAG: Oblivious Retrieval via Semantic Quantization and Transcendental Addressing for Large Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20897","citing_title":"PeerCheck: Enhancing LLM-Generated Academic Reviews Towards Human-Level Quality","ref_index":86,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09125","citing_title":"Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05004","citing_title":"SharedRequest: Privacy-Preserving Model-Agnostic Inference for Large Language Models","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01908","citing_title":"Private and Stable Test-Time Adaptation with Differential Privacy","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2509.12958","citing_title":"Forget What's Sensitive, Remember What Matters: Token-Level Differential Privacy in Memory Sculpting for Continual Learning","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2401.00870","citing_title":"ConfusionPrompt: Practical Private Inference for Online Large Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2402.16391","citing_title":"Industry Practitioners Perspectives on AI Model Quality: Perceptions, Challenges, and Solutions","ref_index":139,"is_internal_anchor":false},{"citing_arxiv_id":"2501.02407","citing_title":"Towards the Anonymization of the Language Modeling","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2506.05640","citing_title":"FedShield-LLM: A Secure and Scalable Federated Fine-Tuned Large Language Model","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2506.15588","citing_title":"Memory-Efficient Differentially Private Training with Gradient Random Projection","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17432","citing_title":"DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15248","citing_title":"Probing Privacy Leaks in LLM-based Code Generation via Test Generation","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2305.15717","citing_title":"The False Promise of Imitating Proprietary LLMs","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2603.01059","citing_title":"GroupGPT: A Token-efficient and Privacy-preserving Agentic Framework for Multi-User Chat Assistant","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12512","citing_title":"Beyond Individual Mimicry: Constructing Human-Like Social network with Graph-Augmented LLM Agents","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21905","citing_title":"Low-Rank Adaptation Redux for Large Models","ref_index":226,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09952","citing_title":"SLM Finetuning for Natural Language to Domain Specific Code Generation in Production","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4","json":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4.json","graph_json":"https://pith.science/api/pith-number/6IDSR27M2LNDIX6PSOOGDBAJM4/graph.json","events_json":"https://pith.science/api/pith-number/6IDSR27M2LNDIX6PSOOGDBAJM4/events.json","paper":"https://pith.science/paper/6IDSR27M"},"agent_actions":{"view_html":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4","download_json":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4.json","view_paper":"https://pith.science/paper/6IDSR27M","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2110.06500&json=true","fetch_graph":"https://pith.science/api/pith-number/6IDSR27M2LNDIX6PSOOGDBAJM4/graph.json","fetch_events":"https://pith.science/api/pith-number/6IDSR27M2LNDIX6PSOOGDBAJM4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4/action/storage_attestation","attest_author":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4/action/author_attestation","sign_citation":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4/action/citation_signature","submit_replication":"https://pith.science/pith/6IDSR27M2LNDIX6PSOOGDBAJM4/action/replication_record"}},"created_at":"2026-07-05T04:40:24.064361+00:00","updated_at":"2026-07-05T04:40:24.064361+00:00"}