Implementation in TensorFlow 2.0 of different examples put together by Yibo Yang, Paris Perdikaris on their original publication on Adversarial Uncertainty Quantification in Physics Informed Neural Networks.
Building upon previous work by Raissi et al on Physics-Informed Neural Networks, the use of latent variable models and adversarial inference allow for quantifying and propagating the uncertainty when solving or identifying Partial Differential Equations.
git clone https://github.com/pierremtb/UQPINNs-TF2.0- Yibo Yang, Paris Perdikaris, Adversarial uncertainty quantification in physics-informed neural networks, Journal of Computational Physics, 2019, ISSN 0021-9991, https://doi.org/10.1016/j.jcp.2019.05.027.
@article{yang2019adversarial,
title={Adversarial uncertainty quantification in physics-informed neural networks},
author={Yang, Yibo and Perdikaris, Paris},
journal={Journal of Computational Physics},
volume={394},
pages={136--152},
year={2019},
publisher={Elsevier}
}
MIT License
Copyright (c) 2019 Pierre Jacquier