{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:3YRDTMD44J2FPCWYPMEW5IPT7D","short_pith_number":"pith:3YRDTMD4","schema_version":"1.0","canonical_sha256":"de2239b07ce274578ad87b096ea1f3f8c4311cfffd3e1fc7431e6b9ff7a6e8e1","source":{"kind":"arxiv","id":"2102.00479","version":2},"attestation_state":"computed","paper":{"title":"Fast Rates for the Regret of Offline Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Masatoshi Uehara, Nathan Kallus, Yichun Hu","submitted_at":"2021-01-31T16:17:56Z","abstract_excerpt":"We study the regret of reinforcement learning from offline data generated by a fixed behavior policy in an infinite-horizon discounted Markov decision process (MDP). While existing analyses of common approaches, such as fitted $Q$-iteration (FQI), suggest a $O(1/\\sqrt{n})$ convergence for regret, empirical behavior exhibits \\emph{much} faster convergence. In this paper, we present a finer regret analysis that exactly characterizes this phenomenon by providing fast rates for the regret convergence. First, we show that given any estimate for the optimal quality function $Q^*$, the regret of the "},"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":"2102.00479","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-01-31T16:17:56Z","cross_cats_sorted":["cs.AI","math.OC","stat.ML"],"title_canon_sha256":"0cfe26f18752674f2444fdd1735b9db74e6ae6e574f9cbb5e213b2bfe841e5a3","abstract_canon_sha256":"ea5d6221122bdb9393fa688ccbe3f9e081eb5d6ef950d4d41868e4c90047590b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:29:59.609015Z","signature_b64":"Q5wp2PDs3in/mGnixoztPzmxQoM6VmHlBiI01noYiX/fs984TXvK9HYX+wgk83Sc8oNWLIm0OW83wOnuH022BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"de2239b07ce274578ad87b096ea1f3f8c4311cfffd3e1fc7431e6b9ff7a6e8e1","last_reissued_at":"2026-07-05T06:29:59.608548Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:29:59.608548Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fast Rates for the Regret of Offline Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Masatoshi Uehara, Nathan Kallus, Yichun Hu","submitted_at":"2021-01-31T16:17:56Z","abstract_excerpt":"We study the regret of reinforcement learning from offline data generated by a fixed behavior policy in an infinite-horizon discounted Markov decision process (MDP). While existing analyses of common approaches, such as fitted $Q$-iteration (FQI), suggest a $O(1/\\sqrt{n})$ convergence for regret, empirical behavior exhibits \\emph{much} faster convergence. In this paper, we present a finer regret analysis that exactly characterizes this phenomenon by providing fast rates for the regret convergence. First, we show that given any estimate for the optimal quality function $Q^*$, the regret of the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2102.00479","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/2102.00479/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":"2102.00479","created_at":"2026-07-05T06:29:59.608612+00:00"},{"alias_kind":"arxiv_version","alias_value":"2102.00479v2","created_at":"2026-07-05T06:29:59.608612+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2102.00479","created_at":"2026-07-05T06:29:59.608612+00:00"},{"alias_kind":"pith_short_12","alias_value":"3YRDTMD44J2F","created_at":"2026-07-05T06:29:59.608612+00:00"},{"alias_kind":"pith_short_16","alias_value":"3YRDTMD44J2FPCWY","created_at":"2026-07-05T06:29:59.608612+00:00"},{"alias_kind":"pith_short_8","alias_value":"3YRDTMD4","created_at":"2026-07-05T06:29:59.608612+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22579","citing_title":"Stationary Robust Mean-Field Games under Model Mismatches","ref_index":233,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00268","citing_title":"Fast Rates in $\\alpha$-Potential Games via Regularized Mirror Descent","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00264","citing_title":"Pessimism-Free Offline Learning in General-Sum Games via KL Regularization","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13025","citing_title":"Offline Two-Player Zero-Sum Markov Games with KL Regularization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05573","citing_title":"AstroAlertBench: Evaluating the Accuracy, Reasoning, and Honesty of Multimodal LLMs in Astronomical Classification","ref_index":86,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00268","citing_title":"Fast Rates in $\\alpha$-Potential Games via Regularized Mirror Descent","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00264","citing_title":"Pessimism-Free Offline Learning in General-Sum Games via KL Regularization","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D","json":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D.json","graph_json":"https://pith.science/api/pith-number/3YRDTMD44J2FPCWYPMEW5IPT7D/graph.json","events_json":"https://pith.science/api/pith-number/3YRDTMD44J2FPCWYPMEW5IPT7D/events.json","paper":"https://pith.science/paper/3YRDTMD4"},"agent_actions":{"view_html":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D","download_json":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D.json","view_paper":"https://pith.science/paper/3YRDTMD4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2102.00479&json=true","fetch_graph":"https://pith.science/api/pith-number/3YRDTMD44J2FPCWYPMEW5IPT7D/graph.json","fetch_events":"https://pith.science/api/pith-number/3YRDTMD44J2FPCWYPMEW5IPT7D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D/action/storage_attestation","attest_author":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D/action/author_attestation","sign_citation":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D/action/citation_signature","submit_replication":"https://pith.science/pith/3YRDTMD44J2FPCWYPMEW5IPT7D/action/replication_record"}},"created_at":"2026-07-05T06:29:59.608612+00:00","updated_at":"2026-07-05T06:29:59.608612+00:00"}