{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZG3PB2WWP3WQCSD57TMYPLXB27","short_pith_number":"pith:ZG3PB2WW","schema_version":"1.0","canonical_sha256":"c9b6f0ead67eed01487dfcd987aee1d7fff42b4373ce1ddc5d4808604977ff3b","source":{"kind":"arxiv","id":"2402.10342","version":2},"attestation_state":"computed","paper":{"title":"Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Anna Winnicki, Gal Dalal, R. Srikant, Shie Mannor, Yihan Du","submitted_at":"2024-02-15T22:11:18Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has achieved impressive empirical successes while relying on a small amount of human feedback. However, there is limited theoretical justification for this phenomenon. Additionally, most recent studies focus on value-based algorithms despite the recent empirical successes of policy-based algorithms. In this work, we consider an RLHF algorithm based on policy optimization (PO-RLHF). The algorithm is based on the popular Policy Cover-Policy Gradient (PC-PG) algorithm, which assumes knowledge of the reward function. In PO-RLHF, knowledge of the re"},"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":"2402.10342","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-15T22:11:18Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"2d4818253cdc937f683b3b3d7fd5b1a654b6d79bf601a9761f6c20f826ef2ddc","abstract_canon_sha256":"d3c50aed247563754a0d03e4b3e7804edab177502ce3f82a4f11b5ecfe548785"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:43:53.234148Z","signature_b64":"VHvC/nyTQtA3vKo+qTIKkwc3ADCAkdqiBsylHc3Q/8fVNvwu80y2srKNjUw3O3TCPefttc0Cthzf59zCbE1WBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c9b6f0ead67eed01487dfcd987aee1d7fff42b4373ce1ddc5d4808604977ff3b","last_reissued_at":"2026-07-05T08:43:53.233685Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:43:53.233685Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Anna Winnicki, Gal Dalal, R. Srikant, Shie Mannor, Yihan Du","submitted_at":"2024-02-15T22:11:18Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has achieved impressive empirical successes while relying on a small amount of human feedback. However, there is limited theoretical justification for this phenomenon. Additionally, most recent studies focus on value-based algorithms despite the recent empirical successes of policy-based algorithms. In this work, we consider an RLHF algorithm based on policy optimization (PO-RLHF). The algorithm is based on the popular Policy Cover-Policy Gradient (PC-PG) algorithm, which assumes knowledge of the reward function. In PO-RLHF, knowledge of the re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.10342","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/2402.10342/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":"2402.10342","created_at":"2026-07-05T08:43:53.233750+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.10342v2","created_at":"2026-07-05T08:43:53.233750+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.10342","created_at":"2026-07-05T08:43:53.233750+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZG3PB2WWP3WQ","created_at":"2026-07-05T08:43:53.233750+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZG3PB2WWP3WQCSD5","created_at":"2026-07-05T08:43:53.233750+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZG3PB2WW","created_at":"2026-07-05T08:43:53.233750+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18963","citing_title":"Online Reward-Punishment Learning from Fixed-Channel Perceptual Event Streams without Environment Rewards","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08933","citing_title":"Corruption-Tolerant Asynchronous Q-Learning with Near-Optimal Rates","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15697","citing_title":"Distributed Zeroth-Order Policy Gradient for Networked Multi-agent Reinforcement Learning from Human Feedback","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19024","citing_title":"Policy Gradient Primal-Dual Method for Safe Reinforcement Learning from Human Feedback","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27","json":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27.json","graph_json":"https://pith.science/api/pith-number/ZG3PB2WWP3WQCSD57TMYPLXB27/graph.json","events_json":"https://pith.science/api/pith-number/ZG3PB2WWP3WQCSD57TMYPLXB27/events.json","paper":"https://pith.science/paper/ZG3PB2WW"},"agent_actions":{"view_html":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27","download_json":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27.json","view_paper":"https://pith.science/paper/ZG3PB2WW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.10342&json=true","fetch_graph":"https://pith.science/api/pith-number/ZG3PB2WWP3WQCSD57TMYPLXB27/graph.json","fetch_events":"https://pith.science/api/pith-number/ZG3PB2WWP3WQCSD57TMYPLXB27/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27/action/storage_attestation","attest_author":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27/action/author_attestation","sign_citation":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27/action/citation_signature","submit_replication":"https://pith.science/pith/ZG3PB2WWP3WQCSD57TMYPLXB27/action/replication_record"}},"created_at":"2026-07-05T08:43:53.233750+00:00","updated_at":"2026-07-05T08:43:53.233750+00:00"}