Uncertainty-aware Multi-fidelity Closure via Conditional Normalizing Flows

In three linesPaper proposing a multi-fidelity approach using conditional normalizing flows to improve reduced-order model (ROM) accuracy. Framework learns probabilistic mapping from low-fidelity to high-fidelity ROM coefficients with uncertainty quantification. Two strategies tested on 2D Navier-Stokes equations: direct and residual learning, with residual learning outperforming direct learning.

Summary generated by Claude — human-verified