{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IXEWMFX46CDPXJ45RT446IK4N2","short_pith_number":"pith:IXEWMFX4","schema_version":"1.0","canonical_sha256":"45c96616fcf086fba79d8cf9cf215c6e8a306ef48ab895bfb0a129e7d0dd8602","source":{"kind":"arxiv","id":"2512.23816","version":2},"attestation_state":"computed","paper":{"title":"Improved Bounds for Private and Robust Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Wenqian Weng, Xingyu Zhou, Yi He","submitted_at":"2025-12-29T19:20:35Z","abstract_excerpt":"In this paper, we study the private and robust alignment of language models from a theoretical perspective by establishing upper bounds on the suboptimality gap in both offline and online settings. We consider preference labels subject to privacy constraints and/or adversarial corruption, and analyze two distinct interplays between them: privacy-first and corruption-first. For the privacy-only setting, we show that log loss with an MLE-style algorithm achieves near-optimal rates, in contrast to conventional wisdom. For the joint privacy-and-corruption setting, we first demonstrate that existin"},"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":"2512.23816","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-12-29T19:20:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"30185f3c716f46015b8d7234b067283debb027f735c55e3c639ffc93e58dcf1f","abstract_canon_sha256":"babbb32e62b416a4dd8d625914c139f79d552e62e1cd0947e9016cbd1347686e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:15:48.179699Z","signature_b64":"RPqAAOrgfGz9fvbIP1V7EV/jNrs20zCpNF6uMc3qMkUIGb4u0X1sw+MOTRJEdnh75+p1V379brcLEbHK09xRCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"45c96616fcf086fba79d8cf9cf215c6e8a306ef48ab895bfb0a129e7d0dd8602","last_reissued_at":"2026-06-26T01:15:48.179219Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:15:48.179219Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improved Bounds for Private and Robust Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Wenqian Weng, Xingyu Zhou, Yi He","submitted_at":"2025-12-29T19:20:35Z","abstract_excerpt":"In this paper, we study the private and robust alignment of language models from a theoretical perspective by establishing upper bounds on the suboptimality gap in both offline and online settings. We consider preference labels subject to privacy constraints and/or adversarial corruption, and analyze two distinct interplays between them: privacy-first and corruption-first. For the privacy-only setting, we show that log loss with an MLE-style algorithm achieves near-optimal rates, in contrast to conventional wisdom. For the joint privacy-and-corruption setting, we first demonstrate that existin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.23816","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/2512.23816/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":"2512.23816","created_at":"2026-06-26T01:15:48.179276+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.23816v2","created_at":"2026-06-26T01:15:48.179276+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.23816","created_at":"2026-06-26T01:15:48.179276+00:00"},{"alias_kind":"pith_short_12","alias_value":"IXEWMFX46CDP","created_at":"2026-06-26T01:15:48.179276+00:00"},{"alias_kind":"pith_short_16","alias_value":"IXEWMFX46CDPXJ45","created_at":"2026-06-26T01:15:48.179276+00:00"},{"alias_kind":"pith_short_8","alias_value":"IXEWMFX4","created_at":"2026-06-26T01:15:48.179276+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.02141","citing_title":"On the Optimal Sample Complexity of Offline Multi-Armed Bandits with KL Regularization","ref_index":54,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2","json":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2.json","graph_json":"https://pith.science/api/pith-number/IXEWMFX46CDPXJ45RT446IK4N2/graph.json","events_json":"https://pith.science/api/pith-number/IXEWMFX46CDPXJ45RT446IK4N2/events.json","paper":"https://pith.science/paper/IXEWMFX4"},"agent_actions":{"view_html":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2","download_json":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2.json","view_paper":"https://pith.science/paper/IXEWMFX4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.23816&json=true","fetch_graph":"https://pith.science/api/pith-number/IXEWMFX46CDPXJ45RT446IK4N2/graph.json","fetch_events":"https://pith.science/api/pith-number/IXEWMFX46CDPXJ45RT446IK4N2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2/action/storage_attestation","attest_author":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2/action/author_attestation","sign_citation":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2/action/citation_signature","submit_replication":"https://pith.science/pith/IXEWMFX46CDPXJ45RT446IK4N2/action/replication_record"}},"created_at":"2026-06-26T01:15:48.179276+00:00","updated_at":"2026-06-26T01:15:48.179276+00:00"}