pyinn 0.1.1

Interpolating Neural Networks

Interpolating Neural Network

This is the github repo for the paper "Interpolating neural network (INN): A novel unification of machine learning and interpolation theory".

INN is a lightweight yet precise network architecture that can replace MLPs for data training, partial differential equation (PDE) solving, and parameter calibration. The key features of INNs are:

• Less trainable parameters than MLP without sacrificing accuracy
• Faster and proven convergent behavior
• Fully differntiable and GPU-optimized

Installation

Create a conda environment from the given environment.yaml file and activate it:

conda env create -f environment.yaml
conda activate pyinn-env

Install JAX

• See jax installation instructions. Depending on your hardware, you may install the CPU or GPU version of JAX. Both will work, while GPU version usually gives better performance.
• For CPU only (Linux/macOS/Windows), one can simply install JAX using:

pip install -U jax

• For GPU (NVIDIA, CUDA 12)

pip install -U "jax[cuda12]"

• For TPU (Google Cloud TPU VM)

pip install -U "jax[tpu]" -f https://storage.googleapis.com/jax-releases/libtpu_releases.html

Install Optax (optimization library of JAX)

pip install optax

Then there are two options to continue:

Option 1

Clone the repository:

git clone https://github.com/hachanook/pyinn.git
cd pyinn

and install the package locally:

pip install -e .

Option 2

Install the package from the PyPI release directly:

pip install pyinn

Quick tests

python ./pyinn/main.py

License

This project is licensed under the GNU General Public License v3 - see the LICENSE for details.

Citations

If you found this library useful in academic or industry work, we appreciate your support if you consider 1) starring the project on Github, and 2) citing relevant papers:

@article{park2024engineering,
  title={Engineering software 2.0 by interpolating neural networks: unifying training, solving, and calibration},
  author={Park, Chanwook and Saha, Sourav and Guo, Jiachen and Zhang, Hantao and Xie, Xiaoyu and Bessa, Miguel A and Qian, Dong and Chen, Wei and Wagner, Gregory J and Cao, Jian and others},
  journal={arXiv preprint arXiv:2404.10296},
  year={2024}
}