{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VPU6VNK2JRMMP4VEDR4IKWHRUJ","short_pith_number":"pith:VPU6VNK2","schema_version":"1.0","canonical_sha256":"abe9eab55a4c58c7f2a41c788558f1a24ae977932e506ef1525235a92e44bb6e","source":{"kind":"arxiv","id":"2503.08942","version":3},"attestation_state":"computed","paper":{"title":"Extragradient Preference Optimization (EGPO): Beyond Last-Iterate Convergence for Nash Learning from Human Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Maryam Fazel, Runlong Zhou, Simon S. Du","submitted_at":"2025-03-11T22:44:54Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has become essential for improving language model capabilities, but traditional approaches rely on the assumption that human preferences follow a transitive Bradley-Terry model. This assumption fails to capture the non-transitive nature of populational human preferences. Nash learning from human feedback (NLHF), targeting non-transitive preferences, is a problem of computing the Nash equilibrium (NE) of the two-player constant-sum game defined by the human preference. We introduce Extragradient preference optimization (EGPO), a novel algorithm "},"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":"2503.08942","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-03-11T22:44:54Z","cross_cats_sorted":[],"title_canon_sha256":"c12b9008e8b86bdcc927c6b588b0dbe473d73c1e846e37936fcf6b4477b2a8b3","abstract_canon_sha256":"a33b7983543c55db71089152283878e961906c069dd354a3e96eadd81575a64c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:34:05.674265Z","signature_b64":"UCIfWrBSs621CYlhArworOW4SHs0zlOTQtJWAAva5IU/lqtGjVMGI3TcnnmAfJl8HetUUszAyzWjCNgaWfK3CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abe9eab55a4c58c7f2a41c788558f1a24ae977932e506ef1525235a92e44bb6e","last_reissued_at":"2026-07-05T11:34:05.673785Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:34:05.673785Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Extragradient Preference Optimization (EGPO): Beyond Last-Iterate Convergence for Nash Learning from Human Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Maryam Fazel, Runlong Zhou, Simon S. Du","submitted_at":"2025-03-11T22:44:54Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has become essential for improving language model capabilities, but traditional approaches rely on the assumption that human preferences follow a transitive Bradley-Terry model. This assumption fails to capture the non-transitive nature of populational human preferences. Nash learning from human feedback (NLHF), targeting non-transitive preferences, is a problem of computing the Nash equilibrium (NE) of the two-player constant-sum game defined by the human preference. We introduce Extragradient preference optimization (EGPO), a novel algorithm "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.08942","kind":"arxiv","version":3},"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/2503.08942/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":"2503.08942","created_at":"2026-07-05T11:34:05.673843+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.08942v3","created_at":"2026-07-05T11:34:05.673843+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.08942","created_at":"2026-07-05T11:34:05.673843+00:00"},{"alias_kind":"pith_short_12","alias_value":"VPU6VNK2JRMM","created_at":"2026-07-05T11:34:05.673843+00:00"},{"alias_kind":"pith_short_16","alias_value":"VPU6VNK2JRMMP4VE","created_at":"2026-07-05T11:34:05.673843+00:00"},{"alias_kind":"pith_short_8","alias_value":"VPU6VNK2","created_at":"2026-07-05T11:34:05.673843+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17342","citing_title":"Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment","ref_index":108,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23102","citing_title":"Multiplayer Nash Preference Optimization","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04494","citing_title":"Towards General Preference Alignment: Diffusion Models at Nash Equilibrium","ref_index":55,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ","json":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ.json","graph_json":"https://pith.science/api/pith-number/VPU6VNK2JRMMP4VEDR4IKWHRUJ/graph.json","events_json":"https://pith.science/api/pith-number/VPU6VNK2JRMMP4VEDR4IKWHRUJ/events.json","paper":"https://pith.science/paper/VPU6VNK2"},"agent_actions":{"view_html":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ","download_json":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ.json","view_paper":"https://pith.science/paper/VPU6VNK2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.08942&json=true","fetch_graph":"https://pith.science/api/pith-number/VPU6VNK2JRMMP4VEDR4IKWHRUJ/graph.json","fetch_events":"https://pith.science/api/pith-number/VPU6VNK2JRMMP4VEDR4IKWHRUJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ/action/storage_attestation","attest_author":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ/action/author_attestation","sign_citation":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ/action/citation_signature","submit_replication":"https://pith.science/pith/VPU6VNK2JRMMP4VEDR4IKWHRUJ/action/replication_record"}},"created_at":"2026-07-05T11:34:05.673843+00:00","updated_at":"2026-07-05T11:34:05.673843+00:00"}