{"paper":{"title":"Flow Matching for Offline Reinforcement Learning with Discrete Actions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Flow matching with continuous-time Markov chains recovers the optimal policy for offline RL with discrete actions under idealized conditions.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Fairoz Nower Khan, Haibo Yang, Nabuat Zaman Nahim, Peizhong Ju, Ruiquan Huang","submitted_at":"2026-02-05T19:13:44Z","abstract_excerpt":"Generative policies based on diffusion models and flow matching have shown strong promise for offline reinforcement learning (RL), but their applicability remains largely confined to continuous action spaces. To address a broader range of offline RL settings, we extend flow matching to a general framework that supports discrete action spaces with multiple objectives. Specifically, we replace continuous flows with continuous-time Markov chains, trained using a Q-weighted flow matching objective. We then extend our design to multi-agent settings, mitigating the exponential growth of joint action"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We theoretically show that, under idealized conditions, optimizing this objective recovers the optimal policy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The idealized conditions required for the theoretical recovery of the optimal policy via the Q-weighted flow matching objective on continuous-time Markov chains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Flow matching is adapted to discrete actions via continuous-time Markov chains and Q-weighted objectives, recovering optimal policies under idealized conditions and outperforming baselines in multi-agent and multi-objective offline RL experiments.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Flow matching with continuous-time Markov chains recovers the optimal policy for offline RL with discrete actions under idealized conditions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"257b3c19fed51e739de024701a9845a8ff131855ca4c1e688c555be4955e9c8b"},"source":{"id":"2602.06138","kind":"arxiv","version":2},"verdict":{"id":"a0619a25-7229-4b5d-abfb-ad74b8166353","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T06:36:26.134774Z","strongest_claim":"We theoretically show that, under idealized conditions, optimizing this objective recovers the optimal policy.","one_line_summary":"Flow matching is adapted to discrete actions via continuous-time Markov chains and Q-weighted objectives, recovering optimal policies under idealized conditions and outperforming baselines in multi-agent and multi-objective offline RL experiments.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The idealized conditions required for the theoretical recovery of the optimal policy via the Q-weighted flow matching objective on continuous-time Markov chains.","pith_extraction_headline":"Flow matching with continuous-time Markov chains recovers the optimal policy for offline RL with discrete actions under idealized conditions."},"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"}