{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6LNEJMZSNSV7AKXHK4FHOSVFNJ","short_pith_number":"pith:6LNEJMZS","schema_version":"1.0","canonical_sha256":"f2da44b3326cabf02ae7570a774aa56a69893084297d80a6b5a238102b8806bf","source":{"kind":"arxiv","id":"2501.18914","version":1},"attestation_state":"computed","paper":{"title":"Scaling Laws for Differentially Private Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Amer Sinha, Badih Ghazi, Borja Balle, Chiyuan Zhang, Christopher A. Choquette-Choo, Da Yu, George Kaissis, Ravi Kumar, Ruibo Liu, Ryan McKenna, Yangsibo Huang, Zachary Charles","submitted_at":"2025-01-31T06:32:46Z","abstract_excerpt":"Scaling laws have emerged as important components of large language model (LLM) training as they can predict performance gains through scale, and provide guidance on important hyper-parameter choices that would otherwise be expensive. LLMs also rely on large, high-quality training datasets, like those sourced from (sometimes sensitive) user data. Training models on this sensitive user data requires careful privacy protections like differential privacy (DP). However, the dynamics of DP training are significantly different, and consequently their scaling laws are not yet fully understood. In thi"},"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":"2501.18914","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-01-31T06:32:46Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"90084de20f8b6246e8d2c512875a8fe767775833ab5cab1d195be8f68e7d4538","abstract_canon_sha256":"0c82875eca3d1fd04a9c99b910f848afedfaec73e9e8e2bf19d90c4f6562485e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:07:53.576969Z","signature_b64":"8yptBRw7kXu89D0t7LnnQrMG7QChk7I1DVx4YUfQFueuAxX307tftjtqumH6OzXAqowl2GrPyEiUg1rurnrMDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f2da44b3326cabf02ae7570a774aa56a69893084297d80a6b5a238102b8806bf","last_reissued_at":"2026-07-05T10:07:53.576435Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:07:53.576435Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaling Laws for Differentially Private Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Amer Sinha, Badih Ghazi, Borja Balle, Chiyuan Zhang, Christopher A. Choquette-Choo, Da Yu, George Kaissis, Ravi Kumar, Ruibo Liu, Ryan McKenna, Yangsibo Huang, Zachary Charles","submitted_at":"2025-01-31T06:32:46Z","abstract_excerpt":"Scaling laws have emerged as important components of large language model (LLM) training as they can predict performance gains through scale, and provide guidance on important hyper-parameter choices that would otherwise be expensive. LLMs also rely on large, high-quality training datasets, like those sourced from (sometimes sensitive) user data. Training models on this sensitive user data requires careful privacy protections like differential privacy (DP). However, the dynamics of DP training are significantly different, and consequently their scaling laws are not yet fully understood. In thi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.18914","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/2501.18914/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":"2501.18914","created_at":"2026-07-05T10:07:53.576504+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.18914v1","created_at":"2026-07-05T10:07:53.576504+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.18914","created_at":"2026-07-05T10:07:53.576504+00:00"},{"alias_kind":"pith_short_12","alias_value":"6LNEJMZSNSV7","created_at":"2026-07-05T10:07:53.576504+00:00"},{"alias_kind":"pith_short_16","alias_value":"6LNEJMZSNSV7AKXH","created_at":"2026-07-05T10:07:53.576504+00:00"},{"alias_kind":"pith_short_8","alias_value":"6LNEJMZS","created_at":"2026-07-05T10:07:53.576504+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2505.16329","citing_title":"High-Dimensional Private Linear Regression with Optimal Rates","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2601.10237","citing_title":"Fundamental Limitations of Favorable Privacy-Utility Guarantees for DP-SGD","ref_index":47,"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":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ","json":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ.json","graph_json":"https://pith.science/api/pith-number/6LNEJMZSNSV7AKXHK4FHOSVFNJ/graph.json","events_json":"https://pith.science/api/pith-number/6LNEJMZSNSV7AKXHK4FHOSVFNJ/events.json","paper":"https://pith.science/paper/6LNEJMZS"},"agent_actions":{"view_html":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ","download_json":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ.json","view_paper":"https://pith.science/paper/6LNEJMZS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.18914&json=true","fetch_graph":"https://pith.science/api/pith-number/6LNEJMZSNSV7AKXHK4FHOSVFNJ/graph.json","fetch_events":"https://pith.science/api/pith-number/6LNEJMZSNSV7AKXHK4FHOSVFNJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ/action/storage_attestation","attest_author":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ/action/author_attestation","sign_citation":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ/action/citation_signature","submit_replication":"https://pith.science/pith/6LNEJMZSNSV7AKXHK4FHOSVFNJ/action/replication_record"}},"created_at":"2026-07-05T10:07:53.576504+00:00","updated_at":"2026-07-05T10:07:53.576504+00:00"}