Search papers, labs, and topics across Lattice.
The paper addresses the challenge of posterior sampling in scientific inverse problems using Flow Matching (FM) priors, where existing methods lack exploration of the source variable. They propose D-Flow SGLD, a source-space posterior sampling method that combines differentiable source inference with preconditioned stochastic gradient Langevin dynamics to enable scalable exploration of the source posterior. Benchmarking on 2D posteriors, chaotic systems, and turbulence reconstruction demonstrates D-Flow SGLD's effectiveness in balancing measurement assimilation, posterior diversity, and physics fidelity.
D-Flow SGLD unlocks scalable posterior sampling for scientific inverse problems using Flow Matching, achieving a better trade-off between measurement assimilation, posterior diversity, and physics fidelity than existing methods.
Data assimilation and scientific inverse problems require reconstructing high-dimensional physical states from sparse and noisy observations, ideally with uncertainty-aware posterior samples that remain faithful to learned priors and governing physics. While training-free conditional generation is well developed for diffusion models, corresponding conditioning and posterior sampling strategies for Flow Matching (FM) priors remain comparatively under-explored, especially on scientific benchmarks where fidelity must be assessed beyond measurement misfit. In this work, we study training-free conditional generation for scientific inverse problems under FM priors and organize existing inference-time strategies by where measurement information is injected: (i) guided transport dynamics that perturb sampling trajectories using likelihood information, and (ii) source-distribution inference that performs posterior inference over the source variable while keeping the learned transport fixed. Building on the latter, we propose D-Flow SGLD, a source-space posterior sampling method that augments differentiable source inference with preconditioned stochastic gradient Langevin dynamics, enabling scalable exploration of the source posterior induced by new measurement operators without retraining the prior or modifying the learned FM dynamics. We benchmark representative methods from both families on a hierarchy of problems: 2D toy posteriors, chaotic Kuramoto-Sivashinsky trajectories, and wall-bounded turbulence reconstruction. Across these settings, we quantify trade-offs among measurement assimilation, posterior diversity, and physics/statistics fidelity, and establish D-Flow SGLD as a practical FM-compatible posterior sampler for scientific inverse problems.
