mesh-based-learning

HI-MGN introduces a hierarchical multiscale graph neural network to improve long-range communication in mesh-based physics simulations, enhancing accuracy while reducing training time and memory usage compared to existing methods.