Dilated convolution neural operator for multiscale partial differential equations

Bo Xu, Lei Zhang

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: operator learning, multiscale PDE, dilated convolution, high frequency features, spectral bias

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Dilated convolution neural operator for multiscale PDEs.

Abstract: This paper presents a data-driven operator learning method for multiscale partial differential equations, where preserving high-frequency information is critical. We propose the Dilated Convolution Neural Operator (DCNO), which combines dilated convolution layers to effectively capture high-frequency features at a low computational cost, along with Fourier layers to handle smooth features. We conduct experiments to evaluate the performance of DCNO on various datasets, including the multiscale elliptic equation, its inverse problem, Navier-Stokes equation, and Helmholtz equation. DCNO stands out with significantly higher accuracy compared to existing neural operator techniques, and strikes an optimal balance between accuracy and computational cost.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5125