{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:AIGVCECJDA3SHJGRMXCCYAXP6R","short_pith_number":"pith:AIGVCECJ","schema_version":"1.0","canonical_sha256":"020d511049183723a4d165c42c02eff47a48959f930b8bfa97fba6d7ddeff2ee","source":{"kind":"arxiv","id":"2105.10148","version":2},"attestation_state":"computed","paper":{"title":"On Instrumental Variable Regression for Deep Offline Policy Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Arnaud Doucet, Arthur Gretton, Caglar Gulcehre, Liyuan Xu, Nando de Freitas, Tom Le Paine, Yutian Chen","submitted_at":"2021-05-21T06:22:34Z","abstract_excerpt":"We show that the popular reinforcement learning (RL) strategy of estimating the state-action value (Q-function) by minimizing the mean squared Bellman error leads to a regression problem with confounding, the inputs and output noise being correlated. Hence, direct minimization of the Bellman error can result in significantly biased Q-function estimates. We explain why fixing the target Q-network in Deep Q-Networks and Fitted Q Evaluation provides a way of overcoming this confounding, thus shedding new light on this popular but not well understood trick in the deep RL literature. An alternative"},"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":"2105.10148","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-05-21T06:22:34Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"5a6e30ac6ca390ba2b6c5bb1c21fec3ce545148082a55aa40418f8d8c83d6b74","abstract_canon_sha256":"bf28fecda263fab97a72713a3827c2fa3c4f7a9ee41b11add606a3c1a174c248"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:20:58.271234Z","signature_b64":"HVgrV41Bt8id0Jfb2KNKxHlFvVmi2CK0bZ081ec8FUIwjJe2fJyByTPv7vR1DS3z99aaZ9cqY0UyMo4a8SyQAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"020d511049183723a4d165c42c02eff47a48959f930b8bfa97fba6d7ddeff2ee","last_reissued_at":"2026-07-05T05:20:58.270065Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:20:58.270065Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Instrumental Variable Regression for Deep Offline Policy Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Arnaud Doucet, Arthur Gretton, Caglar Gulcehre, Liyuan Xu, Nando de Freitas, Tom Le Paine, Yutian Chen","submitted_at":"2021-05-21T06:22:34Z","abstract_excerpt":"We show that the popular reinforcement learning (RL) strategy of estimating the state-action value (Q-function) by minimizing the mean squared Bellman error leads to a regression problem with confounding, the inputs and output noise being correlated. Hence, direct minimization of the Bellman error can result in significantly biased Q-function estimates. We explain why fixing the target Q-network in Deep Q-Networks and Fitted Q Evaluation provides a way of overcoming this confounding, thus shedding new light on this popular but not well understood trick in the deep RL literature. An alternative"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.10148","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/2105.10148/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":"2105.10148","created_at":"2026-07-05T05:20:58.270130+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.10148v2","created_at":"2026-07-05T05:20:58.270130+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.10148","created_at":"2026-07-05T05:20:58.270130+00:00"},{"alias_kind":"pith_short_12","alias_value":"AIGVCECJDA3S","created_at":"2026-07-05T05:20:58.270130+00:00"},{"alias_kind":"pith_short_16","alias_value":"AIGVCECJDA3SHJGR","created_at":"2026-07-05T05:20:58.270130+00:00"},{"alias_kind":"pith_short_8","alias_value":"AIGVCECJ","created_at":"2026-07-05T05:20:58.270130+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.17309","citing_title":"Confounded Causal Imitation Learning with Instrumental Variables","ref_index":2021,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R","json":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R.json","graph_json":"https://pith.science/api/pith-number/AIGVCECJDA3SHJGRMXCCYAXP6R/graph.json","events_json":"https://pith.science/api/pith-number/AIGVCECJDA3SHJGRMXCCYAXP6R/events.json","paper":"https://pith.science/paper/AIGVCECJ"},"agent_actions":{"view_html":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R","download_json":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R.json","view_paper":"https://pith.science/paper/AIGVCECJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.10148&json=true","fetch_graph":"https://pith.science/api/pith-number/AIGVCECJDA3SHJGRMXCCYAXP6R/graph.json","fetch_events":"https://pith.science/api/pith-number/AIGVCECJDA3SHJGRMXCCYAXP6R/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R/action/storage_attestation","attest_author":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R/action/author_attestation","sign_citation":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R/action/citation_signature","submit_replication":"https://pith.science/pith/AIGVCECJDA3SHJGRMXCCYAXP6R/action/replication_record"}},"created_at":"2026-07-05T05:20:58.270130+00:00","updated_at":"2026-07-05T05:20:58.270130+00:00"}