{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6KAY4W3HOISRER77RS6GJE7OUU","short_pith_number":"pith:6KAY4W3H","schema_version":"1.0","canonical_sha256":"f2818e5b6772251247ff8cbc6493eea506ef7ed507afe553d84c316a1c5ad3d3","source":{"kind":"arxiv","id":"2606.05434","version":1},"attestation_state":"computed","paper":{"title":"Selective-Advantage Entropy-Adaptive Horizon GRPO: Asymmetric Token-Level Discounting for Efficient Reinforcement Learning of Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chirag Chawla, Madhav S. Baidya, Rohan Charudatt Salvi","submitted_at":"2026-06-03T20:57:57Z","abstract_excerpt":"Group Relative Policy Optimisation (GRPO) has emerged as an effective reinforcement-learning algorithm for aligning language models on reasoning tasks, but it treats every token position and every sampled rollout symmetrically. We introduce two complementary extensions: (i) Adaptive-Horizon GRPO (AH-GRPO), which weights each token's policy gradient using a cumulative entropy-based discount that reduces the effective horizon when the model is uncertain, and (ii) Selective-Advantage AH-GRPO (SA-AH-GRPO), which applies this discounting only to negative-advantage rollouts, leaving positive-advanta"},"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":"2606.05434","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-03T20:57:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6a680eaa346257d11b2a6706e7d81efdad50682bdb862bbcd818e490000e4c01","abstract_canon_sha256":"afa466e83cf3cac1ee7a8636d154963c1fd22511518eff960fbf48953152f33f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T00:13:57.645992Z","signature_b64":"6MqdghlG4PMXNV2vC9WV8lzrnP8PcLH8srHOk0bz/G1aGwrwu6gVIGMLKddgCvXrbdUwBa3M/25rYUZhAtDBCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f2818e5b6772251247ff8cbc6493eea506ef7ed507afe553d84c316a1c5ad3d3","last_reissued_at":"2026-06-05T00:13:57.645337Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T00:13:57.645337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Selective-Advantage Entropy-Adaptive Horizon GRPO: Asymmetric Token-Level Discounting for Efficient Reinforcement Learning of Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chirag Chawla, Madhav S. Baidya, Rohan Charudatt Salvi","submitted_at":"2026-06-03T20:57:57Z","abstract_excerpt":"Group Relative Policy Optimisation (GRPO) has emerged as an effective reinforcement-learning algorithm for aligning language models on reasoning tasks, but it treats every token position and every sampled rollout symmetrically. We introduce two complementary extensions: (i) Adaptive-Horizon GRPO (AH-GRPO), which weights each token's policy gradient using a cumulative entropy-based discount that reduces the effective horizon when the model is uncertain, and (ii) Selective-Advantage AH-GRPO (SA-AH-GRPO), which applies this discounting only to negative-advantage rollouts, leaving positive-advanta"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.05434","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/2606.05434/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":"2606.05434","created_at":"2026-06-05T00:13:57.645436+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.05434v1","created_at":"2026-06-05T00:13:57.645436+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.05434","created_at":"2026-06-05T00:13:57.645436+00:00"},{"alias_kind":"pith_short_12","alias_value":"6KAY4W3HOISR","created_at":"2026-06-05T00:13:57.645436+00:00"},{"alias_kind":"pith_short_16","alias_value":"6KAY4W3HOISRER77","created_at":"2026-06-05T00:13:57.645436+00:00"},{"alias_kind":"pith_short_8","alias_value":"6KAY4W3H","created_at":"2026-06-05T00:13:57.645436+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/6KAY4W3HOISRER77RS6GJE7OUU","json":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU.json","graph_json":"https://pith.science/api/pith-number/6KAY4W3HOISRER77RS6GJE7OUU/graph.json","events_json":"https://pith.science/api/pith-number/6KAY4W3HOISRER77RS6GJE7OUU/events.json","paper":"https://pith.science/paper/6KAY4W3H"},"agent_actions":{"view_html":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU","download_json":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU.json","view_paper":"https://pith.science/paper/6KAY4W3H","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.05434&json=true","fetch_graph":"https://pith.science/api/pith-number/6KAY4W3HOISRER77RS6GJE7OUU/graph.json","fetch_events":"https://pith.science/api/pith-number/6KAY4W3HOISRER77RS6GJE7OUU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU/action/storage_attestation","attest_author":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU/action/author_attestation","sign_citation":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU/action/citation_signature","submit_replication":"https://pith.science/pith/6KAY4W3HOISRER77RS6GJE7OUU/action/replication_record"}},"created_at":"2026-06-05T00:13:57.645436+00:00","updated_at":"2026-06-05T00:13:57.645436+00:00"}