{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VPUYMYPBMGB2PBR76ZODRJ2PMC","short_pith_number":"pith:VPUYMYPB","schema_version":"1.0","canonical_sha256":"abe98661e16183a7863ff65c38a74f609acc3b4f5ba010438fe2c47f372b18ca","source":{"kind":"arxiv","id":"2605.15620","version":1},"attestation_state":"computed","paper":{"title":"Pessimistic Risk-Aware Policy Learning in Contextual Bandits","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Optimizing general Lipschitz risk criteria in offline contextual bandits incurs no additional statistical cost beyond expected-reward optimization.","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Xianyi Wu, Yilong Wan, Yuqiang Li","submitted_at":"2026-05-15T05:02:39Z","abstract_excerpt":"We study risk-aware offline policy learning, aiming to learn a decision rule from logged data that is optimal under general risk criteria. This problem is crucial in high-stakes domains where online interaction is infeasible and adverse outcomes must be carefully controlled. However, existing literature on offline contextual bandits either centers on expected-reward criteria or restricts risk considerations to policy evaluation instead of optimization. In this work, we propose a unified distributional framework for optimizing Lipschitz-continuous risk functionals, a broad class of risk measure"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15620","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2026-05-15T05:02:39Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1cebe2e302eaee8d624a7f67f11397e57cbc39183447f338bc17e23d1dab15de","abstract_canon_sha256":"50373642908d5d2856918dfd4d66f73ac2a09880e82adcaf66f556428726e16e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:01:08.570748Z","signature_b64":"tTFZXlpZYV0KC02UUuWo5/hLLF67tq5Tkk9uFLycaJ434ZV6HFb2AxESgHjKke4ETwBEa3RCGiOGsPUfS3mHCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abe98661e16183a7863ff65c38a74f609acc3b4f5ba010438fe2c47f372b18ca","last_reissued_at":"2026-05-20T00:01:08.569800Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:01:08.569800Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pessimistic Risk-Aware Policy Learning in Contextual Bandits","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Optimizing general Lipschitz risk criteria in offline contextual bandits incurs no additional statistical cost beyond expected-reward optimization.","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Xianyi Wu, Yilong Wan, Yuqiang Li","submitted_at":"2026-05-15T05:02:39Z","abstract_excerpt":"We study risk-aware offline policy learning, aiming to learn a decision rule from logged data that is optimal under general risk criteria. This problem is crucial in high-stakes domains where online interaction is infeasible and adverse outcomes must be carefully controlled. However, existing literature on offline contextual bandits either centers on expected-reward criteria or restricts risk considerations to policy evaluation instead of optimization. In this work, we propose a unified distributional framework for optimizing Lipschitz-continuous risk functionals, a broad class of risk measure"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By developing novel empirical concentration inequalities for importance sampling-based distributional estimators, our analysis derives data-dependent suboptimality bounds with an Õ(1/√n) rate, without relying on restrictive uniform overlap assumptions. This rate is minimax optimal and matches that of risk-neutral offline policy optimization, indicating that optimizing general Lipschitz risk criteria incurs no additional statistical cost relative to the expected-reward.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The risk functionals under consideration are Lipschitz continuous (invoked to unify mean-variance, entropic risk, CVaR, etc.), and the novel empirical concentration inequalities for the importance-sampling distributional estimators hold under the paper's data-dependent conditions rather than uniform overlap.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A distributional framework for optimizing Lipschitz risk functionals in offline contextual bandits yields data-dependent suboptimality bounds of Õ(1/√n) that match risk-neutral rates and are minimax optimal.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Optimizing general Lipschitz risk criteria in offline contextual bandits incurs no additional statistical cost beyond expected-reward optimization.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"67433709bdf798a67057a2615b33ecdfd5ec33a451c17b25f89ba927c230f43c"},"source":{"id":"2605.15620","kind":"arxiv","version":1},"verdict":{"id":"59a36701-8fab-46d8-a810-5197b5946ea0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T19:55:02.621438Z","strongest_claim":"By developing novel empirical concentration inequalities for importance sampling-based distributional estimators, our analysis derives data-dependent suboptimality bounds with an Õ(1/√n) rate, without relying on restrictive uniform overlap assumptions. This rate is minimax optimal and matches that of risk-neutral offline policy optimization, indicating that optimizing general Lipschitz risk criteria incurs no additional statistical cost relative to the expected-reward.","one_line_summary":"A distributional framework for optimizing Lipschitz risk functionals in offline contextual bandits yields data-dependent suboptimality bounds of Õ(1/√n) that match risk-neutral rates and are minimax optimal.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The risk functionals under consideration are Lipschitz continuous (invoked to unify mean-variance, entropic risk, CVaR, etc.), and the novel empirical concentration inequalities for the importance-sampling distributional estimators hold under the paper's data-dependent conditions rather than uniform overlap.","pith_extraction_headline":"Optimizing general Lipschitz risk criteria in offline contextual bandits incurs no additional statistical cost beyond expected-reward optimization."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15620/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T20:01:32.540667Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T20:01:19.280114Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T19:34:34.614121Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T17:41:56.037474Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"87f730bfe9c9a9b492df9a6dde8d0da079e57cb4f63095ba879a3247b323f483"},"references":{"count":48,"sample":[{"doi":"10.1111/1467-9965.00068","year":1999,"title":"Mathematical Finance , volume =","work_id":"ad02250a-d8c8-4778-8728-47023c29ce1e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Policy learning with observational data","work_id":"904aa1ed-af83-4258-8a65-18ef7e54e1ea","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Regret bounds for risk-sensitive reinforcement learning","work_id":"f88b16fa-83e0-49a1-bb5d-b1b2f8a69966","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Bellemare, Will Dabney, and R \\'e mi Munos","work_id":"01462f4b-76ea-4d24-8179-253f0c3ebeab","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1287/mnsc.2018.3253","year":2019,"title":"From predictive to prescriptive analytics","work_id":"6367a78c-7e2d-47a8-8ba7-fc75bf2341be","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":48,"snapshot_sha256":"1ad5bc460e8a33280dfe04a4bb4e05f7ef6acf0bf0dd461a17766df49f34f363","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"7829d69fc38279d2394ddd98d1ecb3e7d51e776e7a49e6f20f71f11409e22e38"},"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":"2605.15620","created_at":"2026-05-20T00:01:08.569972+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15620v1","created_at":"2026-05-20T00:01:08.569972+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15620","created_at":"2026-05-20T00:01:08.569972+00:00"},{"alias_kind":"pith_short_12","alias_value":"VPUYMYPBMGB2","created_at":"2026-05-20T00:01:08.569972+00:00"},{"alias_kind":"pith_short_16","alias_value":"VPUYMYPBMGB2PBR7","created_at":"2026-05-20T00:01:08.569972+00:00"},{"alias_kind":"pith_short_8","alias_value":"VPUYMYPB","created_at":"2026-05-20T00:01:08.569972+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":1,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC","json":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC.json","graph_json":"https://pith.science/api/pith-number/VPUYMYPBMGB2PBR76ZODRJ2PMC/graph.json","events_json":"https://pith.science/api/pith-number/VPUYMYPBMGB2PBR76ZODRJ2PMC/events.json","paper":"https://pith.science/paper/VPUYMYPB"},"agent_actions":{"view_html":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC","download_json":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC.json","view_paper":"https://pith.science/paper/VPUYMYPB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15620&json=true","fetch_graph":"https://pith.science/api/pith-number/VPUYMYPBMGB2PBR76ZODRJ2PMC/graph.json","fetch_events":"https://pith.science/api/pith-number/VPUYMYPBMGB2PBR76ZODRJ2PMC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC/action/storage_attestation","attest_author":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC/action/author_attestation","sign_citation":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC/action/citation_signature","submit_replication":"https://pith.science/pith/VPUYMYPBMGB2PBR76ZODRJ2PMC/action/replication_record"}},"created_at":"2026-05-20T00:01:08.569972+00:00","updated_at":"2026-05-20T00:01:08.569972+00:00"}