{"paper":{"title":"Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Many reward-based fine-tuning methods for diffusion and flow models reduce to a single score-matching objective against a value-guided target.","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Jeongjae Lee, Jeongsol Kim, Jinho Chang, Jong Chul Ye","submitted_at":"2026-04-19T12:47:52Z","abstract_excerpt":"Reward-based fine-tuning steers a pretrained diffusion or flow-based generative model toward higher-reward samples while remaining close to the pretrained model. Although existing methods are derived from different perspectives, we show that many can be written under a common framework, which we call reward score matching (RSM). Under this view, alignment becomes score matching against a value-guided target, and the main differences across methods reduce to the construction of the value-guidance estimator and the effective optimization strength across timesteps. This unification clarifies the "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Although existing methods are derived from different perspectives, we show that many can be written under a common framework, which we call reward score matching (RSM).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the primary distinctions among existing methods reduce to the construction of the value-guidance estimator and the effective optimization strength across timesteps, without material loss of generality or overlooked auxiliary mechanisms.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Reward Score Matching unifies reward-based fine-tuning for flow and diffusion models by recasting alignment as score matching to a value-guided target.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Many reward-based fine-tuning methods for diffusion and flow models reduce to a single score-matching objective against a value-guided target.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4a6e37e38fb316c83f3835af855adce2bbccf2923c582855d1481dae5a07005b"},"source":{"id":"2604.17415","kind":"arxiv","version":3},"verdict":{"id":"be8edd1a-49b9-46d3-ba94-a222182b8113","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T05:30:29.711746Z","strongest_claim":"Although existing methods are derived from different perspectives, we show that many can be written under a common framework, which we call reward score matching (RSM).","one_line_summary":"Reward Score Matching unifies reward-based fine-tuning for flow and diffusion models by recasting alignment as score matching to a value-guided target.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the primary distinctions among existing methods reduce to the construction of the value-guidance estimator and the effective optimization strength across timesteps, without material loss of generality or overlooked auxiliary mechanisms.","pith_extraction_headline":"Many reward-based fine-tuning methods for diffusion and flow models reduce to a single score-matching objective against a value-guided target."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.17415/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"}