Wuzhe Xu: Diffusion Models Meet Scientific Computing

Diffusion models have achieved remarkable success across various domains, yet their application in scientific computing remains underexplored. In the first part of this talk, I will introduce the Physics-Guided Diffusion Model, a surrogate model that achieves over tenfold speedup compared to traditional numerical solvers while maintaining high accuracy. In the second part, I will present a novel framework that leverages diffusion models to enhance data fidelity without requiring correspondence between low-fidelity and reference data during training. This approach is validated through both theoretical analysis and experimental results.