{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7ZCBIX56TYBCPKLWPE32E4Z2RM","short_pith_number":"pith:7ZCBIX56","schema_version":"1.0","canonical_sha256":"fe44145fbe9e0227a9767937a2733a8b0a96a87b2148fab9d72dd472cb289071","source":{"kind":"arxiv","id":"2506.02864","version":1},"attestation_state":"computed","paper":{"title":"BNPO: Beta Normalization Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Changyi Xiao, Mengdi Zhang, Yixin Cao","submitted_at":"2025-06-03T13:28:57Z","abstract_excerpt":"Recent studies, including DeepSeek-R1 and Kimi-k1.5, have demonstrated that reinforcement learning with rule-based, binary-valued reward functions can significantly enhance the reasoning capabilities of large language models. These models primarily utilize REINFORCE-based policy optimization techniques, such as REINFORCE with baseline and group relative policy optimization (GRPO). However, a key limitation remains: current policy optimization methods either neglect reward normalization or employ static normalization strategies, which fail to adapt to the dynamic nature of policy updates during"},"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":"2506.02864","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-03T13:28:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d72a6408d5b8109e434ea2d89dd72c54fd683b258c4cc7f659150970e767cb5f","abstract_canon_sha256":"3e2b7af543dafe8e013407c79ffdd2db05837cf53875a799896b08b8f492375f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:15:10.643813Z","signature_b64":"4UT1c1mCpQ6+js76lNyVyuiNTQI+AsTJkP6Lqm+4LgwGRkKGW2RsT4wCOLnAKiPC3NAxPYMOhmj4Zgg0dM73Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fe44145fbe9e0227a9767937a2733a8b0a96a87b2148fab9d72dd472cb289071","last_reissued_at":"2026-07-05T11:15:10.643278Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:15:10.643278Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BNPO: Beta Normalization Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Changyi Xiao, Mengdi Zhang, Yixin Cao","submitted_at":"2025-06-03T13:28:57Z","abstract_excerpt":"Recent studies, including DeepSeek-R1 and Kimi-k1.5, have demonstrated that reinforcement learning with rule-based, binary-valued reward functions can significantly enhance the reasoning capabilities of large language models. These models primarily utilize REINFORCE-based policy optimization techniques, such as REINFORCE with baseline and group relative policy optimization (GRPO). However, a key limitation remains: current policy optimization methods either neglect reward normalization or employ static normalization strategies, which fail to adapt to the dynamic nature of policy updates during"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02864","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/2506.02864/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":"2506.02864","created_at":"2026-07-05T11:15:10.643367+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.02864v1","created_at":"2026-07-05T11:15:10.643367+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02864","created_at":"2026-07-05T11:15:10.643367+00:00"},{"alias_kind":"pith_short_12","alias_value":"7ZCBIX56TYBC","created_at":"2026-07-05T11:15:10.643367+00:00"},{"alias_kind":"pith_short_16","alias_value":"7ZCBIX56TYBCPKLW","created_at":"2026-07-05T11:15:10.643367+00:00"},{"alias_kind":"pith_short_8","alias_value":"7ZCBIX56","created_at":"2026-07-05T11:15:10.643367+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":232,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04889","citing_title":"GRAIL: Gradient-Reweighted Advantages for Reinforcement Learning with Verifiable Rewards","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29758","citing_title":"PS-PPO: Prefix-Sampling PPO for Critic-Free RLHF","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28421","citing_title":"DenoiseRL: Bootstrapping Reasoning Models to Recover from Noisy Prefixes","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12058","citing_title":"Holder Policy Optimisation","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12058","citing_title":"Holder Policy Optimisation","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08873","citing_title":"CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy Optimization","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM","json":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM.json","graph_json":"https://pith.science/api/pith-number/7ZCBIX56TYBCPKLWPE32E4Z2RM/graph.json","events_json":"https://pith.science/api/pith-number/7ZCBIX56TYBCPKLWPE32E4Z2RM/events.json","paper":"https://pith.science/paper/7ZCBIX56"},"agent_actions":{"view_html":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM","download_json":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM.json","view_paper":"https://pith.science/paper/7ZCBIX56","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.02864&json=true","fetch_graph":"https://pith.science/api/pith-number/7ZCBIX56TYBCPKLWPE32E4Z2RM/graph.json","fetch_events":"https://pith.science/api/pith-number/7ZCBIX56TYBCPKLWPE32E4Z2RM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM/action/storage_attestation","attest_author":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM/action/author_attestation","sign_citation":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM/action/citation_signature","submit_replication":"https://pith.science/pith/7ZCBIX56TYBCPKLWPE32E4Z2RM/action/replication_record"}},"created_at":"2026-07-05T11:15:10.643367+00:00","updated_at":"2026-07-05T11:15:10.643367+00:00"}