This repository contains the official implementation of Taylor Mode Neural Operators (TMNO) as presented in the paper:
Taylor Mode Neural Operators: Enhancing Computational Efficiency in Physics-Informed Neural Operators
Anonymous, NeurIPS 2024 Machine Learning and the Physical Sciences Workshop.
OpenReview Link
Taylor Mode Neural Operators introduce a novel application of Taylor-mode Automatic Differentiation (AD) to efficiently compute high-order derivatives in Physics-Informed Neural Operators (PINOs). TMNO demonstrates significant computational efficiency in high-order derivative calculations by propagating Taylor coefficients directly through neural operator architectures. The approach is validated on DeepONet and Fourier Neural Operators (FNOs), achieving:
- Up to an order-of-magnitude speed-up for DeepONet.
- An eightfold acceleration for FNOs.
The implementation of TMNO applied to Fourier Neural Operators is publicly available in this repository. Refer to the code and examples in the fno/ folder for detailed instructions and experiments.
The code for TMNO applied to DeepONet and the extension to 3D cases is available upon request. If interested, please contact us through the repository's issue tracker or via the email provided in the contact section.
If you use this repository or its contents in your research, please cite the paper as follows:
@inproceedings{ anonymous2024taylor, title={Taylor Mode Neural Operators: Enhancing Computational Efficiency in Physics-Informed Neural Operators}, author={Anonymous}, booktitle={Machine Learning and the Physical Sciences Workshop @ NeurIPS 2024}, year={2024}, url={https://openreview.net/forum?id=BvA24ROnJ0} }
For inquiries about the code or additional features, please open an issue or reach out via email (refer to the contact information in the repository).