{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "choi-et-al-selfimproving-robust-preference-2024", "title": "Choi et al.: self-improving robust preference optimization", "summary": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision.", "boundary": "Robustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.", "area": "Language-model preference optimization", "topic": "Artificial intelligence", "year": 2024, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "choi-et-al-selfimproving-robust-preference-2024", "target": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "humans": [ "Eugene Choi", "Arash Ahmadian", "Matthieu Geist", "Oilvier Pietquin", "Mohammad Gheshlaghi Azar" ], "authors_complete": true, "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261015-arxiv-2406-01660v1", "title": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } ], "points": "65000", "published_grade": { "id": "choi-et-al-selfimproving-robust-preference-2024", "entry_id": "choi-et-al-selfimproving-robust-preference-2024", "band": "B", "modifier": "+", "target": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "what_changed": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision.", "scope": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision. Robustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-16T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Eugene Choi", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Arash Ahmadian", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Matthieu Geist", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Oilvier Pietquin", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Mohammad Gheshlaghi Azar", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Robustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.", "domain": "Language-model preference optimization" }, "sources": [ { "citation": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1" } ], "statement": "# Choi et al.: self-improving robust preference optimization\n\nJointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision.\n\n## Scientific scope\n\nRobustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nEugene Choi; Arash Ahmadian; Matthieu Geist; Oilvier Pietquin; Mohammad Gheshlaghi Azar\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [Self-Improving Robust Preference Optimization](https://arxiv.org/abs/2406.01660v1)\n", "summary": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision.", "title": "Choi et al.: self-improving robust preference optimization" }, "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "source_ref": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:b8c4aef19b4f8ef5c41e67fcbeac245c8beca370f71a84b7f51b477ae35c85e0" }, "target": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "choi-et-al-selfimproving-robust-preference-2024", "entry_id": "choi-et-al-selfimproving-robust-preference-2024", "revision": 1, "title": "Choi et al.: self-improving robust preference optimization", "context": "Language-model preference optimization", "track": "Human-designed research method or resource", "band": "B", "modifier": "+", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision.", "scope": "Jointly optimized a generation policy and a policy that improves its outputs through a regularized adversarial preference objective, deriving an offline supervised training formulation and evaluating iterative self-revision. Robustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.", "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Robustness refers to behavior-policy dependence in the stated fixed-preference formulation, not arbitrary prompt-distribution, preference, support or model-class shifts. Theoretical identities are author-reported and not independently audited here; summarization results do not certify general factual correctness.", "references": [ { "title": "Self-Improving Robust Preference Optimization", "url": "https://arxiv.org/abs/2406.01660v1", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." } ], "roles": [ { "who": "Source-listed authors and technical contributors", "credit": "SCIENCE", "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities." } ], "target": { "record_id": "urn:uuid:d28d77db-9e80-4255-a39b-3877492bf664", "version": 1, "content_hash": "sha256:9212d2676a1e0a5607dc9d19b0838444187f41285e29b0c8b91606c3ee723fc3" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "65000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }