{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:X5R3JTR4JBUIKUINLZUQCLYAVK","short_pith_number":"pith:X5R3JTR4","schema_version":"1.0","canonical_sha256":"bf63b4ce3c486885510d5e69012f00aa8384bbf88b7c85a6cb17abb05c8d4ac1","source":{"kind":"arxiv","id":"2605.01663","version":2},"attestation_state":"computed","paper":{"title":"Towards Efficient and Expressive Offline RL via Flow-Anchored Noise-conditioned Q-Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FAN achieves state-of-the-art offline RL performance using only a single flow-policy iteration and one Gaussian noise sample.","cross_cats":["cs.RO"],"primary_cat":"cs.LG","authors_text":"Dohyeong Kim, Eshan Balachandar, Keshav Pingali, Sungyoung Lee, Zelal Su Mustafaoglu","submitted_at":"2026-05-03T01:32:11Z","abstract_excerpt":"We propose Flow-Anchored Noise-conditioned Q-Learning (FAN), a highly efficient and high-performing offline reinforcement learning (RL) algorithm. Recent work has shown that expressive flow policies and distributional critics improve offline RL performance, but at a high computational cost. Specifically, flow policies require iterative sampling to produce a single action, and distributional critics require computation over multiple samples (e.g., quantiles) to estimate value. To address these inefficiencies while maintaining high performance, we introduce FAN. Our method employs a behavior reg"},"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":"2605.01663","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-03T01:32:11Z","cross_cats_sorted":["cs.RO"],"title_canon_sha256":"9163953be47922487f85a8383db863696a4148abbc3f1ffda5c63994e733a055","abstract_canon_sha256":"d45041ee201ca8c1a6bbe7d5376547b3aa0d0d90cd496f8364d355579af4f9ab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:11.368788Z","signature_b64":"AyZfooStJO9a14R8VCuwRbxMcRmiX7XqCMGcFVSIkbVKW36eGxTB755npKysF1B4phmDOVHsrngDaTMYX8o4CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf63b4ce3c486885510d5e69012f00aa8384bbf88b7c85a6cb17abb05c8d4ac1","last_reissued_at":"2026-05-29T01:05:11.368025Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:11.368025Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Efficient and Expressive Offline RL via Flow-Anchored Noise-conditioned Q-Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FAN achieves state-of-the-art offline RL performance using only a single flow-policy iteration and one Gaussian noise sample.","cross_cats":["cs.RO"],"primary_cat":"cs.LG","authors_text":"Dohyeong Kim, Eshan Balachandar, Keshav Pingali, Sungyoung Lee, Zelal Su Mustafaoglu","submitted_at":"2026-05-03T01:32:11Z","abstract_excerpt":"We propose Flow-Anchored Noise-conditioned Q-Learning (FAN), a highly efficient and high-performing offline reinforcement learning (RL) algorithm. Recent work has shown that expressive flow policies and distributional critics improve offline RL performance, but at a high computational cost. Specifically, flow policies require iterative sampling to produce a single action, and distributional critics require computation over multiple samples (e.g., quantiles) to estimate value. To address these inefficiencies while maintaining high performance, we introduce FAN. Our method employs a behavior reg"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our theoretical analysis of convergence and performance bounds demonstrates that these simplifications not only improve efficiency but also lead to superior task performance. Experiments on robotic manipulation and locomotion tasks demonstrate that FAN achieves state-of-the-art performance while significantly reducing both training and inference runtimes.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a single flow-policy iteration plus one Gaussian noise sample for the distributional critic is sufficient to preserve both the expressivity of full iterative flows and the accuracy of multi-sample critics without introducing bias that the behavior-regularization term cannot correct.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FAN achieves state-of-the-art offline RL performance on robotic tasks by anchoring flow policies and using single-sample noise-conditioned Q-learning, with proven convergence and reduced runtimes.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"FAN achieves state-of-the-art offline RL performance using only a single flow-policy iteration and one Gaussian noise sample.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b195a35fa0760693cb5967884294db4574282337badea60acf2ff498e338fa0e"},"source":{"id":"2605.01663","kind":"arxiv","version":2},"verdict":{"id":"9af6d24a-c1d6-4a6b-b05d-5df2ce6fa894","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T16:20:37.119102Z","strongest_claim":"Our theoretical analysis of convergence and performance bounds demonstrates that these simplifications not only improve efficiency but also lead to superior task performance. Experiments on robotic manipulation and locomotion tasks demonstrate that FAN achieves state-of-the-art performance while significantly reducing both training and inference runtimes.","one_line_summary":"FAN achieves state-of-the-art offline RL performance on robotic tasks by anchoring flow policies and using single-sample noise-conditioned Q-learning, with proven convergence and reduced runtimes.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a single flow-policy iteration plus one Gaussian noise sample for the distributional critic is sufficient to preserve both the expressivity of full iterative flows and the accuracy of multi-sample critics without introducing bias that the behavior-regularization term cannot correct.","pith_extraction_headline":"FAN achieves state-of-the-art offline RL performance using only a single flow-policy iteration and one Gaussian noise sample."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.01663/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T17:38:34.430715Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T05:01:23.561282Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T17:06:47.706456Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"063471dc615972128316ef49c620a90335b38be83431545bcdd6b0e8d931ad77"},"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":"2605.01663","created_at":"2026-05-29T01:05:11.368135+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.01663v2","created_at":"2026-05-29T01:05:11.368135+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.01663","created_at":"2026-05-29T01:05:11.368135+00:00"},{"alias_kind":"pith_short_12","alias_value":"X5R3JTR4JBUI","created_at":"2026-05-29T01:05:11.368135+00:00"},{"alias_kind":"pith_short_16","alias_value":"X5R3JTR4JBUIKUIN","created_at":"2026-05-29T01:05:11.368135+00:00"},{"alias_kind":"pith_short_8","alias_value":"X5R3JTR4","created_at":"2026-05-29T01:05:11.368135+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK","json":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK.json","graph_json":"https://pith.science/api/pith-number/X5R3JTR4JBUIKUINLZUQCLYAVK/graph.json","events_json":"https://pith.science/api/pith-number/X5R3JTR4JBUIKUINLZUQCLYAVK/events.json","paper":"https://pith.science/paper/X5R3JTR4"},"agent_actions":{"view_html":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK","download_json":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK.json","view_paper":"https://pith.science/paper/X5R3JTR4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.01663&json=true","fetch_graph":"https://pith.science/api/pith-number/X5R3JTR4JBUIKUINLZUQCLYAVK/graph.json","fetch_events":"https://pith.science/api/pith-number/X5R3JTR4JBUIKUINLZUQCLYAVK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK/action/storage_attestation","attest_author":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK/action/author_attestation","sign_citation":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK/action/citation_signature","submit_replication":"https://pith.science/pith/X5R3JTR4JBUIKUINLZUQCLYAVK/action/replication_record"}},"created_at":"2026-05-29T01:05:11.368135+00:00","updated_at":"2026-05-29T01:05:11.368135+00:00"}