{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:BX2MMW7HNHXNSAURKSJ2EOAW2B","short_pith_number":"pith:BX2MMW7H","schema_version":"1.0","canonical_sha256":"0df4c65be769eed902915493a23816d0785afe950ddebd3336b2780dc9843a70","source":{"kind":"arxiv","id":"2003.07337","version":1},"attestation_state":"computed","paper":{"title":"Is Temporal Difference Learning Optimal? An Instance-Dependent Analysis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","math.OC"],"primary_cat":"stat.ML","authors_text":"Ashwin Pananjady, Feng Ruan, Koulik Khamaru, Martin J. Wainwright, Michael I. Jordan","submitted_at":"2020-03-16T17:15:28Z","abstract_excerpt":"We address the problem of policy evaluation in discounted Markov decision processes, and provide instance-dependent guarantees on the $\\ell_\\infty$-error under a generative model. We establish both asymptotic and non-asymptotic versions of local minimax lower bounds for policy evaluation, thereby providing an instance-dependent baseline by which to compare algorithms. Theory-inspired simulations show that the widely-used temporal difference (TD) algorithm is strictly suboptimal when evaluated in a non-asymptotic setting, even when combined with Polyak-Ruppert iterate averaging. We remedy this "},"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":"2003.07337","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2020-03-16T17:15:28Z","cross_cats_sorted":["cs.LG","math.OC"],"title_canon_sha256":"d0c5834d0e7924f612f72757d087e354767cc2c4c505b315ab984870d82ca83a","abstract_canon_sha256":"6a4947a8bbdf84d981b7682d02c58785a14f4a45d6a61ae9ef6c12081a99a38c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:48:11.649256Z","signature_b64":"oj8z/7M7e0k6DxHcKw0YHsIUY+HULXLu8SvlMY/jdl+BsTszdK3eUOjN9hEsAxuhWMJ6opWBlK1koH28Zh17Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0df4c65be769eed902915493a23816d0785afe950ddebd3336b2780dc9843a70","last_reissued_at":"2026-07-05T00:48:11.648699Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:48:11.648699Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Temporal Difference Learning Optimal? An Instance-Dependent Analysis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","math.OC"],"primary_cat":"stat.ML","authors_text":"Ashwin Pananjady, Feng Ruan, Koulik Khamaru, Martin J. Wainwright, Michael I. Jordan","submitted_at":"2020-03-16T17:15:28Z","abstract_excerpt":"We address the problem of policy evaluation in discounted Markov decision processes, and provide instance-dependent guarantees on the $\\ell_\\infty$-error under a generative model. We establish both asymptotic and non-asymptotic versions of local minimax lower bounds for policy evaluation, thereby providing an instance-dependent baseline by which to compare algorithms. Theory-inspired simulations show that the widely-used temporal difference (TD) algorithm is strictly suboptimal when evaluated in a non-asymptotic setting, even when combined with Polyak-Ruppert iterate averaging. We remedy this "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2003.07337","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/2003.07337/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":"2003.07337","created_at":"2026-07-05T00:48:11.648762+00:00"},{"alias_kind":"arxiv_version","alias_value":"2003.07337v1","created_at":"2026-07-05T00:48:11.648762+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2003.07337","created_at":"2026-07-05T00:48:11.648762+00:00"},{"alias_kind":"pith_short_12","alias_value":"BX2MMW7HNHXN","created_at":"2026-07-05T00:48:11.648762+00:00"},{"alias_kind":"pith_short_16","alias_value":"BX2MMW7HNHXNSAUR","created_at":"2026-07-05T00:48:11.648762+00:00"},{"alias_kind":"pith_short_8","alias_value":"BX2MMW7H","created_at":"2026-07-05T00:48:11.648762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.08933","citing_title":"Corruption-Tolerant Asynchronous Q-Learning with Near-Optimal Rates","ref_index":82,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B","json":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B.json","graph_json":"https://pith.science/api/pith-number/BX2MMW7HNHXNSAURKSJ2EOAW2B/graph.json","events_json":"https://pith.science/api/pith-number/BX2MMW7HNHXNSAURKSJ2EOAW2B/events.json","paper":"https://pith.science/paper/BX2MMW7H"},"agent_actions":{"view_html":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B","download_json":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B.json","view_paper":"https://pith.science/paper/BX2MMW7H","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2003.07337&json=true","fetch_graph":"https://pith.science/api/pith-number/BX2MMW7HNHXNSAURKSJ2EOAW2B/graph.json","fetch_events":"https://pith.science/api/pith-number/BX2MMW7HNHXNSAURKSJ2EOAW2B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B/action/storage_attestation","attest_author":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B/action/author_attestation","sign_citation":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B/action/citation_signature","submit_replication":"https://pith.science/pith/BX2MMW7HNHXNSAURKSJ2EOAW2B/action/replication_record"}},"created_at":"2026-07-05T00:48:11.648762+00:00","updated_at":"2026-07-05T00:48:11.648762+00:00"}