{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7UY7XQMD3L5L7QXJLUOCXRL6SF","short_pith_number":"pith:7UY7XQMD","schema_version":"1.0","canonical_sha256":"fd31fbc183dafabfc2e95d1c2bc57e915ed0240e8644c2713c9a759c4fd6acf6","source":{"kind":"arxiv","id":"2503.00030","version":2},"attestation_state":"computed","paper":{"title":"RSPO: Regularized Self-Play Alignment of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Huizhuo Yuan, Ilija Bogunovic, Quanquan Gu, Sangwoong Yoon, Seongho Son, Xiaohang Tang","submitted_at":"2025-02-24T22:43:21Z","abstract_excerpt":"Self-play alignment has emerged as an effective approach for fine-tuning large language models (LLMs), formulating preference optimization as a two-player game. However, the regularization with respect to the reference policy, which is crucial for mitigating over-optimization, has been insufficiently investigated in self-play alignment. To study the impact of different regularization strategies, we propose \\textbf{Regularized Self-Play Policy Optimization (RSPO)}, a general and modular framework that unifies prior methods and enables simple plug-and-play integration of various regularizers, me"},"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":"2503.00030","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-24T22:43:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"24580a7b237c5529cde6d2c0a2e0487d3e54d23d91d1f31dd071dbbd6c88aaa9","abstract_canon_sha256":"4e5ad56209f570e5087434d0228ae6882d6e1d1d3f60c3c8dae19eda5bef2aa5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:33:20.878632Z","signature_b64":"UIhlkhyWxIHb8rkRNbKtyz/EcBRE4Pa8co4qMcjbRzlvwXTdMx5YS/FS+VTBKWkteLSXR3nQoR6kPFMJymz7BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd31fbc183dafabfc2e95d1c2bc57e915ed0240e8644c2713c9a759c4fd6acf6","last_reissued_at":"2026-07-05T11:33:20.878167Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:33:20.878167Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RSPO: Regularized Self-Play Alignment of Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Huizhuo Yuan, Ilija Bogunovic, Quanquan Gu, Sangwoong Yoon, Seongho Son, Xiaohang Tang","submitted_at":"2025-02-24T22:43:21Z","abstract_excerpt":"Self-play alignment has emerged as an effective approach for fine-tuning large language models (LLMs), formulating preference optimization as a two-player game. However, the regularization with respect to the reference policy, which is crucial for mitigating over-optimization, has been insufficiently investigated in self-play alignment. To study the impact of different regularization strategies, we propose \\textbf{Regularized Self-Play Policy Optimization (RSPO)}, a general and modular framework that unifies prior methods and enables simple plug-and-play integration of various regularizers, me"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.00030","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/2503.00030/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":"2503.00030","created_at":"2026-07-05T11:33:20.878231+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.00030v2","created_at":"2026-07-05T11:33:20.878231+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.00030","created_at":"2026-07-05T11:33:20.878231+00:00"},{"alias_kind":"pith_short_12","alias_value":"7UY7XQMD3L5L","created_at":"2026-07-05T11:33:20.878231+00:00"},{"alias_kind":"pith_short_16","alias_value":"7UY7XQMD3L5L7QXJ","created_at":"2026-07-05T11:33:20.878231+00:00"},{"alias_kind":"pith_short_8","alias_value":"7UY7XQMD","created_at":"2026-07-05T11:33:20.878231+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01561","citing_title":"S-SPPO: Semantic-Calibrated Self-Play Preference Optimization","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29398","citing_title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20933","citing_title":"IRIS: Interpolative R\\'enyi Iterative Self-play for Large Language Model Fine-Tuning","ref_index":68,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF","json":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF.json","graph_json":"https://pith.science/api/pith-number/7UY7XQMD3L5L7QXJLUOCXRL6SF/graph.json","events_json":"https://pith.science/api/pith-number/7UY7XQMD3L5L7QXJLUOCXRL6SF/events.json","paper":"https://pith.science/paper/7UY7XQMD"},"agent_actions":{"view_html":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF","download_json":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF.json","view_paper":"https://pith.science/paper/7UY7XQMD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.00030&json=true","fetch_graph":"https://pith.science/api/pith-number/7UY7XQMD3L5L7QXJLUOCXRL6SF/graph.json","fetch_events":"https://pith.science/api/pith-number/7UY7XQMD3L5L7QXJLUOCXRL6SF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF/action/storage_attestation","attest_author":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF/action/author_attestation","sign_citation":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF/action/citation_signature","submit_replication":"https://pith.science/pith/7UY7XQMD3L5L7QXJLUOCXRL6SF/action/replication_record"}},"created_at":"2026-07-05T11:33:20.878231+00:00","updated_at":"2026-07-05T11:33:20.878231+00:00"}