Controllable Patching for Compute-Adaptive Surrogate Modeling of Partial Differential Equations

This repository will contain the code for implementing convolutional kernel (CKM) and stride (CSM) modulators.

CKM and CSM represent a class of general flexible patching and striding strategies which allows flexible patching in vision transformer based PDE surrogates. Across a range of challenging 2D and 3D PDE benchmarks, CKM and CSM are shown to dramatically improve the long term stability of video-like prediction tasks. Its plug-and-play design makes it broadly applicable across architectures—establishing a general foundation for compute-adaptive modeling in PDE surrogate tasks.