Masked autoencoders enhance PDE solvers through pretraining, improving performance on downstream tasks and generalizing to diverse equations.
