{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GFVGZIKGHJY3VPEYWMHI3QPUBC","short_pith_number":"pith:GFVGZIKG","schema_version":"1.0","canonical_sha256":"316a6ca1463a71babc98b30e8dc1f408b3950059ee9620ff09b3e6b558e2cec4","source":{"kind":"arxiv","id":"2505.15694","version":1},"attestation_state":"computed","paper":{"title":"A Unified Theoretical Analysis of Private and Robust Offline Alignment: from RLHF to DPO","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Francesco Orabona, Xingyu Zhou, Yulian Wu","submitted_at":"2025-05-21T16:07:47Z","abstract_excerpt":"In this paper, we theoretically investigate the effects of noisy labels in offline alignment, with a focus on the interplay between privacy and robustness against adversarial corruption. Specifically, under linear modeling assumptions, we present a unified analysis covering both reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) under different privacy-corruption scenarios, such as Local differential privacy-then-Corruption (LTC), where human preference labels are privatized before being corrupted by an adversary, and Corruption-then-Local differential p"},"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":"2505.15694","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-21T16:07:47Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"133161794aaa6b1439f62c1d5600fd5452cd253fc373613a6d37581e7244902f","abstract_canon_sha256":"204952a969e2aec0e81a0fcf658829cdad9747f905f49c62d47ed348c94a3ae1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:49.689891Z","signature_b64":"6CWc9DaroRNca1ls/ToYQGvxXKaCy3s4ZHASwND1wPdbP/ts5n35a7E1ssC+buZj9oyA+6tdWw+hmI4DqQpvCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"316a6ca1463a71babc98b30e8dc1f408b3950059ee9620ff09b3e6b558e2cec4","last_reissued_at":"2026-07-05T11:06:49.689400Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:49.689400Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Unified Theoretical Analysis of Private and Robust Offline Alignment: from RLHF to DPO","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Francesco Orabona, Xingyu Zhou, Yulian Wu","submitted_at":"2025-05-21T16:07:47Z","abstract_excerpt":"In this paper, we theoretically investigate the effects of noisy labels in offline alignment, with a focus on the interplay between privacy and robustness against adversarial corruption. Specifically, under linear modeling assumptions, we present a unified analysis covering both reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) under different privacy-corruption scenarios, such as Local differential privacy-then-Corruption (LTC), where human preference labels are privatized before being corrupted by an adversary, and Corruption-then-Local differential p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15694","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/2505.15694/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":"2505.15694","created_at":"2026-07-05T11:06:49.689477+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15694v1","created_at":"2026-07-05T11:06:49.689477+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15694","created_at":"2026-07-05T11:06:49.689477+00:00"},{"alias_kind":"pith_short_12","alias_value":"GFVGZIKGHJY3","created_at":"2026-07-05T11:06:49.689477+00:00"},{"alias_kind":"pith_short_16","alias_value":"GFVGZIKGHJY3VPEY","created_at":"2026-07-05T11:06:49.689477+00:00"},{"alias_kind":"pith_short_8","alias_value":"GFVGZIKG","created_at":"2026-07-05T11:06:49.689477+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.02507","citing_title":"Reinforcement Learning from Human Feedback: A Statistical Perspective","ref_index":95,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11134","citing_title":"Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC","json":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC.json","graph_json":"https://pith.science/api/pith-number/GFVGZIKGHJY3VPEYWMHI3QPUBC/graph.json","events_json":"https://pith.science/api/pith-number/GFVGZIKGHJY3VPEYWMHI3QPUBC/events.json","paper":"https://pith.science/paper/GFVGZIKG"},"agent_actions":{"view_html":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC","download_json":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC.json","view_paper":"https://pith.science/paper/GFVGZIKG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15694&json=true","fetch_graph":"https://pith.science/api/pith-number/GFVGZIKGHJY3VPEYWMHI3QPUBC/graph.json","fetch_events":"https://pith.science/api/pith-number/GFVGZIKGHJY3VPEYWMHI3QPUBC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC/action/storage_attestation","attest_author":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC/action/author_attestation","sign_citation":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC/action/citation_signature","submit_replication":"https://pith.science/pith/GFVGZIKGHJY3VPEYWMHI3QPUBC/action/replication_record"}},"created_at":"2026-07-05T11:06:49.689477+00:00","updated_at":"2026-07-05T11:06:49.689477+00:00"}