AI
MintFlow: Minimal Trajectory Intervention for Constrained Flow Matching
Publication: cs.AI updates on arXiv.org
Author: Yesom Park, Kelvin Kan, Qifan Chen, Thomas Flynn, Hayden Schaeffer. Xihaier Luo
Source published: 05/10/2026, 04:00:00 UTC
Briefing created: 05/10/2026, 04:29:48 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 04:03:24 UTC.
Briefings
MintFlow is a training-free constrained sampling framework that enforces physical or measurement constraints by minimizing perturbations to an intermediate flow trajectory. It uses an adjoint formulation to derive a closed-form expression for the required perturbation, avoiding expensive iterative optimization. The method adapts the intervention time to balance perturbation magnitude with amplification by the remaining flow. Across generative vision and physical system modeling tasks, MintFlow achieves competitive constraint satisfaction while preserving the pretrained generative distribution substantially better than state-of-the-art constrained methods.
Key points
- MintFlow enforces constraints via minimal perturbation to an intermediate flow state rather than displacing samples from the pretrained distribution.
- An adjoint formulation provides a closed-form expression for the perturbation, eliminating the need for expensive iterative optimization.
Context and limitations
The study covers tasks in generative vision and physical system modeling. Results are presented as competitive performance compared to state-of-the-art methods, but specific quantitative metrics for each task are not detailed in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 05/10/2026, 04:29:48 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeib75zei4n7zmhzinsu74hgvpqawy7tibkvdb7ahrjsungzcpx2uj4