Masked Autoencoders are PDE Learners

arXiv:2403.17728v1 Announce Type: new
Abstract: Neural solvers for partial differential equations (PDEs) have great potential, yet their practicality is currently limited by their generalizability. PDEs evolve over broad scales and exhibit diverse behaviors; predicting these phenomena will require learning representations across a wide variety of inputs, which may encompass different coefficients, geometries, or equations. As a step towards generalizable PDE modeling, we adapt masked pretraining for PDEs. Through self-supervised learning across PDEs, masked autoencoders can learn useful latent representations for …

abstract arxiv autoencoders cs.lg differential diverse inputs type will