{"paper":{"title":"Flow Matching with Uncertainty Quantification and Guidance","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"UA-Flow adds per-sample uncertainty prediction to flow matching and uses it for guided sampling that improves image generation quality over baselines.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Juyeop Han, Lukas Lao Beyer, Sertac Karaman","submitted_at":"2026-02-10T22:03:13Z","abstract_excerpt":"Despite the remarkable success of sampling-based generative models such as flow matching, they can still produce samples of inconsistent or degraded quality. To assess sample reliability and generate higher-quality outputs, we propose uncertainty-aware flow matching (UA-Flow), a lightweight extension of flow matching that predicts the velocity field together with heteroscedastic uncertainty. UA-Flow estimates per-sample uncertainty by propagating velocity uncertainty through the flow dynamics. These uncertainty estimates act as a reliability signal for individual samples, and we further use th"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"UA-Flow produces uncertainty signals more highly correlated with sample fidelity than baseline methods, and uncertainty-guided sampling further improves generation quality.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That propagating the predicted heteroscedastic velocity uncertainty through the flow ODE yields a calibrated per-sample reliability signal that can be directly used for guidance without introducing new biases or requiring additional calibration data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"UA-Flow adds per-sample uncertainty prediction to flow matching and uses it for guided sampling that improves image generation quality over baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"34992cffed99fe94ff42ca5009eb35cec80905a64ef5242df1ab060d9d7221fc"},"source":{"id":"2602.10326","kind":"arxiv","version":2},"verdict":{"id":"18c38307-4b42-4cf7-9fab-0ded5d880d4d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T02:01:22.821249Z","strongest_claim":"UA-Flow produces uncertainty signals more highly correlated with sample fidelity than baseline methods, and uncertainty-guided sampling further improves generation quality.","one_line_summary":"UA-Flow adds per-sample uncertainty prediction to flow matching and uses it for guided sampling that improves image generation quality over baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That propagating the predicted heteroscedastic velocity uncertainty through the flow ODE yields a calibrated per-sample reliability signal that can be directly used for guidance without introducing new biases or requiring additional calibration data.","pith_extraction_headline":""},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8110dec389c22c1106897dc7d6caea7b806631f2fc2f95708409fb0261f33bc2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}