Multi-task-Physics-informed-neural-networks

The concept of multi-task physics-informed neural networks was first proposed in https://arxiv.org/abs/2104.14320. Please visit this repository for the implementation.

----- Research codebase -----
The current directory corresponds to the Burgers' PDE discovery.

See /inverse_KdV for the Korteweg-de Vries (KdV) PDE discovery.

See /inverse_KS for the Kuramoto Sivashinsky (KS) PDE discovery.

See /inverse_NLS for the Non-linear Schrodinger (NLS) (complex-valued) PDE discovery.

See /inverse_qho for the Quantum Harmonic Oscillator (complex-valued) PDE discovery.

For a small (lower data samples) version of KdV and KS, check out /inverse_small_KdV and /inverse_small_KS.

This repo also include the ladder networks simple implementation in PyTorch (See ladder.py).

VAE & AutoEncoder implementation -> /vae_experiments (Credits go to https://github.com/hellojinwoo/TorchCoder for the LSTM autoencoder implementation.)

if you find this repo useful, please consider citing this!

Citation

@article{thanasutives2022noise,
  title={Noise-aware Physics-informed Machine Learning for Robust PDE Discovery},
  author={Thanasutives, Pongpisit and Morita, Takashi and Numao, Masayuki and Fukui, Ken-ichi},
  journal={arXiv preprint arXiv:2206.12901},
  year={2022}
}

@inproceedings{thanasutives2021adversarial,
  title={Adversarial multi-task learning enhanced physics-informed neural networks for solving partial differential equations},
  author={Thanasutives, Pongpisit and Numao, Masayuki and Fukui, Ken-ichi},
  booktitle={2021 International Joint Conference on Neural Networks (IJCNN)},
  pages={1--9},
  year={2021},
  organization={IEEE}
}