{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:UORISE2RFJ2FXVMNSI77L5WAOA","short_pith_number":"pith:UORISE2R","schema_version":"1.0","canonical_sha256":"a3a28913512a745bd58d923ff5f6c0702cf67fa3883cae3281e39aa1365cb19f","source":{"kind":"arxiv","id":"2605.20722","version":1},"attestation_state":"computed","paper":{"title":"AGPO: Adaptive Group Policy Optimization with Dual Statistical Feedback","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bokun Wang, Daren Zha, Jun Xiao, Miaobo Hu, Ruohan Wang, Shuhao Hu, Xiaobo Guo, Xin Wang","submitted_at":"2026-05-20T05:20:46Z","abstract_excerpt":"Reinforcement learning improves LLM reasoning, but PPO/GRPO typically use fixed clipping and decoding temperature, which makes training brittle and tuning-heavy. We propose Adaptive Group Policy Optimization (AGPO), a critic-free refinement of GRPO that uses group-level statistics to control both update magnitude and exploration. AGPO uses a shared probe-derived statistical state to drive two controllers: (i) adaptive clipping, which sets the trust-region size from reward dispersion and skewness, probe vote entropy, policy entropy, and step-wise KL drift; and (ii) bidirectional adaptive temper"},"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":"2605.20722","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T05:20:46Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"dc23ef803de7f1fdb6ac59cdcf6b00040324e336854bfabe0ba12d4029858435","abstract_canon_sha256":"3cf09e6653bfe1d689f6c2837d677563b8e7ee1926abcd14d2c3cc15d4f123d4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:04:50.797142Z","signature_b64":"L2SObAkbXzv2/BA6Ww93E6fkLSoE79Frc8yGOwYrc4rQR2ma8RC/EoDHIgh7M0Tw3g8Y3g+lKFz+y4bJ2MKQBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3a28913512a745bd58d923ff5f6c0702cf67fa3883cae3281e39aa1365cb19f","last_reissued_at":"2026-05-21T01:04:50.796406Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:04:50.796406Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AGPO: Adaptive Group Policy Optimization with Dual Statistical Feedback","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bokun Wang, Daren Zha, Jun Xiao, Miaobo Hu, Ruohan Wang, Shuhao Hu, Xiaobo Guo, Xin Wang","submitted_at":"2026-05-20T05:20:46Z","abstract_excerpt":"Reinforcement learning improves LLM reasoning, but PPO/GRPO typically use fixed clipping and decoding temperature, which makes training brittle and tuning-heavy. We propose Adaptive Group Policy Optimization (AGPO), a critic-free refinement of GRPO that uses group-level statistics to control both update magnitude and exploration. AGPO uses a shared probe-derived statistical state to drive two controllers: (i) adaptive clipping, which sets the trust-region size from reward dispersion and skewness, probe vote entropy, policy entropy, and step-wise KL drift; and (ii) bidirectional adaptive temper"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20722","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/2605.20722/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":"2605.20722","created_at":"2026-05-21T01:04:50.796539+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20722v1","created_at":"2026-05-21T01:04:50.796539+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20722","created_at":"2026-05-21T01:04:50.796539+00:00"},{"alias_kind":"pith_short_12","alias_value":"UORISE2RFJ2F","created_at":"2026-05-21T01:04:50.796539+00:00"},{"alias_kind":"pith_short_16","alias_value":"UORISE2RFJ2FXVMN","created_at":"2026-05-21T01:04:50.796539+00:00"},{"alias_kind":"pith_short_8","alias_value":"UORISE2R","created_at":"2026-05-21T01:04:50.796539+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA","json":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA.json","graph_json":"https://pith.science/api/pith-number/UORISE2RFJ2FXVMNSI77L5WAOA/graph.json","events_json":"https://pith.science/api/pith-number/UORISE2RFJ2FXVMNSI77L5WAOA/events.json","paper":"https://pith.science/paper/UORISE2R"},"agent_actions":{"view_html":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA","download_json":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA.json","view_paper":"https://pith.science/paper/UORISE2R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20722&json=true","fetch_graph":"https://pith.science/api/pith-number/UORISE2RFJ2FXVMNSI77L5WAOA/graph.json","fetch_events":"https://pith.science/api/pith-number/UORISE2RFJ2FXVMNSI77L5WAOA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA/action/storage_attestation","attest_author":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA/action/author_attestation","sign_citation":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA/action/citation_signature","submit_replication":"https://pith.science/pith/UORISE2RFJ2FXVMNSI77L5WAOA/action/replication_record"}},"created_at":"2026-05-21T01:04:50.796539+00:00","updated_at":"2026-05-21T01:04:50.796539+00:00"}