{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:72AYSJU3L7MEIB7TMN6YEXMXFH","short_pith_number":"pith:72AYSJU3","schema_version":"1.0","canonical_sha256":"fe8189269b5fd84407f3637d825d9729ca7482b77dd7e31a5e710f488beb239d","source":{"kind":"arxiv","id":"2110.05679","version":6},"attestation_state":"computed","paper":{"title":"Large Language Models Can Be Strong Differentially Private Learners","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Florian Tram\\`er, Percy Liang, Tatsunori Hashimoto, Xuechen Li","submitted_at":"2021-10-12T01:45:27Z","abstract_excerpt":"Differentially Private (DP) learning has seen limited success for building large deep learning models of text, and straightforward attempts at applying Differentially Private Stochastic Gradient Descent (DP-SGD) to NLP tasks have resulted in large performance drops and high computational overhead. We show that this performance drop can be mitigated with (1) the use of large pretrained language models; (2) non-standard hyperparameters that suit DP optimization; and (3) fine-tuning objectives which are aligned with the pretraining procedure. With the above, we obtain NLP models that outperform s"},"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":"2110.05679","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-10-12T01:45:27Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e76a7d45705645cb0476dd761a0a3548b266165ee772c17b9258e230643f973a","abstract_canon_sha256":"52d6156d90c73c0838d0a286bc24be5d2a4e2205bb08c5c63ba8760346dc2dae"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:14:55.702426Z","signature_b64":"lnW5FpsrqqRq5ceiw5mzTvsgymqUDgOBs7TNEudRSkt1SvZoSR+3MvBdGtkFux64kiawevzepjiFsqjgaXSNAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fe8189269b5fd84407f3637d825d9729ca7482b77dd7e31a5e710f488beb239d","last_reissued_at":"2026-07-05T05:14:55.701991Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:14:55.701991Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Large Language Models Can Be Strong Differentially Private Learners","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Florian Tram\\`er, Percy Liang, Tatsunori Hashimoto, Xuechen Li","submitted_at":"2021-10-12T01:45:27Z","abstract_excerpt":"Differentially Private (DP) learning has seen limited success for building large deep learning models of text, and straightforward attempts at applying Differentially Private Stochastic Gradient Descent (DP-SGD) to NLP tasks have resulted in large performance drops and high computational overhead. We show that this performance drop can be mitigated with (1) the use of large pretrained language models; (2) non-standard hyperparameters that suit DP optimization; and (3) fine-tuning objectives which are aligned with the pretraining procedure. With the above, we obtain NLP models that outperform s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2110.05679","kind":"arxiv","version":6},"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/2110.05679/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":"2110.05679","created_at":"2026-07-05T05:14:55.702047+00:00"},{"alias_kind":"arxiv_version","alias_value":"2110.05679v6","created_at":"2026-07-05T05:14:55.702047+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2110.05679","created_at":"2026-07-05T05:14:55.702047+00:00"},{"alias_kind":"pith_short_12","alias_value":"72AYSJU3L7ME","created_at":"2026-07-05T05:14:55.702047+00:00"},{"alias_kind":"pith_short_16","alias_value":"72AYSJU3L7MEIB7T","created_at":"2026-07-05T05:14:55.702047+00:00"},{"alias_kind":"pith_short_8","alias_value":"72AYSJU3","created_at":"2026-07-05T05:14:55.702047+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26627","citing_title":"Agents That Know Too Much: A Data-Centric Survey of Privacy in LLM Agents","ref_index":66,"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":39,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03399","citing_title":"Selective Token-Level Cryptographic Redaction for Privacy-Preserving Clinical Deployment of Large Language Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00426","citing_title":"Canonicalized Stable-List Replay for Private Federated Continual Learning over Language-Model Embeddings","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31208","citing_title":"Probing Memorization of Tabular In-Context Learning","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24879","citing_title":"Efficient DP-SGD for LLMs with Randomized Clipping","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2506.15588","citing_title":"Memory-Efficient Differentially Private Training with Gradient Random Projection","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15851","citing_title":"DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17432","citing_title":"DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Models","ref_index":18,"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":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14514","citing_title":"Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07072","citing_title":"Less Random, More Private: What is the Optimal Subsampling Scheme for DP-SGD?","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04461","citing_title":"DP-OPD: Differentially Private On-Policy Distillation for Language Models","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15851","citing_title":"DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH","json":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH.json","graph_json":"https://pith.science/api/pith-number/72AYSJU3L7MEIB7TMN6YEXMXFH/graph.json","events_json":"https://pith.science/api/pith-number/72AYSJU3L7MEIB7TMN6YEXMXFH/events.json","paper":"https://pith.science/paper/72AYSJU3"},"agent_actions":{"view_html":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH","download_json":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH.json","view_paper":"https://pith.science/paper/72AYSJU3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2110.05679&json=true","fetch_graph":"https://pith.science/api/pith-number/72AYSJU3L7MEIB7TMN6YEXMXFH/graph.json","fetch_events":"https://pith.science/api/pith-number/72AYSJU3L7MEIB7TMN6YEXMXFH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH/action/storage_attestation","attest_author":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH/action/author_attestation","sign_citation":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH/action/citation_signature","submit_replication":"https://pith.science/pith/72AYSJU3L7MEIB7TMN6YEXMXFH/action/replication_record"}},"created_at":"2026-07-05T05:14:55.702047+00:00","updated_at":"2026-07-05T05:14:55.702047+00:00"}