The code for paper titled "PINNACLE: PINN Adaptive ColLocation and Experimental points selection". The paper has been accepted as a spotlight paper at ICLR 2024 (see https://openreview.net/forum?id=GzNaCp6Vcg), and as a oral presentation at ICML 2024 AI4Science Workshop (see https://openreview.net/forum?id=BbyOaDBuWy).
All the algorithm codes are kept in pinnacle_code/deepxde_al_patch. The code are based off the deepxde package, patched in order to add in collocation point selection as required in our tests. The code has also been heavily adjusted to be compatible with Jax.
Example notebooks are in pinnacle_code/*.ipynb. The various notebooks contain examples to run our modules and methods to set up the training process.
Test scripts used for our experiment can be found in pinnacle_code/al_pinn*.sh. They are wrapper scripts for the test cases used, and they call other python scripts that do the experiment setups.
Run pip install -r requirements.txt to install the relevant packages. The scripts and notebooks can be used in the directory pinnacle_code/ and will do imports accordingly.
Some physics simulation dataset need to be obtained from PDEBench. Refer to https://github.com/pdebench/PDEBench on how the dataset can be downloaded. Once installed their directory can be fed into the test scripts or to the test set loader. Our code includes a module which reads PDEBench data files so the pdebench package itself does not need to be installed.
@inproceedings{pinnacle,
title={PINNACLE: PINN Adaptive ColLocation and Experimental points selection},
author={Lau, Gregory Kang Ruey and Hemachandra, Apivich and Ng, See-Kiong and Low, Bryan Kian Hsiang},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024}
}