{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:SRHBKOBX5KJJXTVIALVE3D6MIA","short_pith_number":"pith:SRHBKOBX","schema_version":"1.0","canonical_sha256":"944e153837ea929bcea802ea4d8fcc402f34168af0c506660c20ea5ea6890307","source":{"kind":"arxiv","id":"2312.05405","version":1},"attestation_state":"computed","paper":{"title":"Guaranteed Trust Region Optimization via Two-Phase KL Penalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Gaurav S. Sukhatme, K.R. Zentner, Ujjwal Puri, Zhehui Huang","submitted_at":"2023-12-08T23:29:57Z","abstract_excerpt":"On-policy reinforcement learning (RL) has become a popular framework for solving sequential decision problems due to its computational efficiency and theoretical simplicity. Some on-policy methods guarantee every policy update is constrained to a trust region relative to the prior policy to ensure training stability. These methods often require computationally intensive non-linear optimization or require a particular form of action distribution. In this work, we show that applying KL penalization alone is nearly sufficient to enforce such trust regions. Then, we show that introducing a \"fixup\""},"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":"2312.05405","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-12-08T23:29:57Z","cross_cats_sorted":[],"title_canon_sha256":"dfeff39f9b23a060712cf3330fda8ba481d99a81af6c5d38db40a9f6c17279b8","abstract_canon_sha256":"2a8c7613c7d01d8297b16a0e594028b52eaf2f8f53e73b6ce66181d3f879a9b8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:22:03.944761Z","signature_b64":"LtXt02oxMY1UkhebUvy3cIe0wkVY/Og7mvhay01OwC0k4y7mVRCT3/y6nYDh2JXvNE1+twxeIWjAkYscEUF1CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"944e153837ea929bcea802ea4d8fcc402f34168af0c506660c20ea5ea6890307","last_reissued_at":"2026-07-05T07:22:03.944301Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:22:03.944301Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Guaranteed Trust Region Optimization via Two-Phase KL Penalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Gaurav S. Sukhatme, K.R. Zentner, Ujjwal Puri, Zhehui Huang","submitted_at":"2023-12-08T23:29:57Z","abstract_excerpt":"On-policy reinforcement learning (RL) has become a popular framework for solving sequential decision problems due to its computational efficiency and theoretical simplicity. Some on-policy methods guarantee every policy update is constrained to a trust region relative to the prior policy to ensure training stability. These methods often require computationally intensive non-linear optimization or require a particular form of action distribution. In this work, we show that applying KL penalization alone is nearly sufficient to enforce such trust regions. Then, we show that introducing a \"fixup\""},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.05405","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/2312.05405/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":"2312.05405","created_at":"2026-07-05T07:22:03.944369+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.05405v1","created_at":"2026-07-05T07:22:03.944369+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.05405","created_at":"2026-07-05T07:22:03.944369+00:00"},{"alias_kind":"pith_short_12","alias_value":"SRHBKOBX5KJJ","created_at":"2026-07-05T07:22:03.944369+00:00"},{"alias_kind":"pith_short_16","alias_value":"SRHBKOBX5KJJXTVI","created_at":"2026-07-05T07:22:03.944369+00:00"},{"alias_kind":"pith_short_8","alias_value":"SRHBKOBX","created_at":"2026-07-05T07:22:03.944369+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.26784","citing_title":"Ratio-Variance Regularized Policy Optimization","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA","json":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA.json","graph_json":"https://pith.science/api/pith-number/SRHBKOBX5KJJXTVIALVE3D6MIA/graph.json","events_json":"https://pith.science/api/pith-number/SRHBKOBX5KJJXTVIALVE3D6MIA/events.json","paper":"https://pith.science/paper/SRHBKOBX"},"agent_actions":{"view_html":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA","download_json":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA.json","view_paper":"https://pith.science/paper/SRHBKOBX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.05405&json=true","fetch_graph":"https://pith.science/api/pith-number/SRHBKOBX5KJJXTVIALVE3D6MIA/graph.json","fetch_events":"https://pith.science/api/pith-number/SRHBKOBX5KJJXTVIALVE3D6MIA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA/action/storage_attestation","attest_author":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA/action/author_attestation","sign_citation":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA/action/citation_signature","submit_replication":"https://pith.science/pith/SRHBKOBX5KJJXTVIALVE3D6MIA/action/replication_record"}},"created_at":"2026-07-05T07:22:03.944369+00:00","updated_at":"2026-07-05T07:22:03.944369+00:00"}