GREAD: Graph Neural Reaction-Diffusion Networks

Introduction

Please see our paper for more details: GREAD: Graph Neural Reaction-Diffusion Networks
We will update more information of the code soon.

Reaction-diffusion on a grid graph | Diffusion on a grid network :-------------------------:|:-------------------------: <img src="asset/gif/reaction_diffusion_on_grid_network.gif" width="250"> | <img src="asset/gif/diffusion_on_grid_network.gif" width="250">

An illustrative comparison between the diffusion equation and our proposed blurring-sharpening (reaction-diffusion) equation on a grid graph with one-dimensional node features. The diffusion equation causes the problem of oversmoothing while the reaction-diffusion seeks a balance between smoothing and sharpening. The diffusion equation is a special case of the reaction-diffusion equation when the sharpening term is zero.

Set environment

The environment can be set up using either environment.yml file or manually installing the dependencies.

Using an environment.yml file

conda env create -f environment.yml

Manually install

conda create -n gread python=3.9
conda activate gread
pip install torch==1.11.0+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install torch-scatter torch-sparse torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
pip install torchdiffeq ogb wandb deeprobust==0.2.4

Dataset and pre-processing

All data gets downloaded and preprocessed automatically and stored in data directory (which gets automatically created the first time one of the experiments is run).

How to run

To run each experiment, navigate into src. Then, run the following command:

python run_GNN.py --kwargs

where kwargs are specified in each individual run_GNN.py file.

You can also run the best hyperparameters for each dataset by adding --use_best_params flag. For example, to run the best hyperparameters for Squirrel dataset, run the following command:

python run_GNN.py --dataset=Squirrel --use_best_params

Citation

If you find this repository useful in your research, please cite our paper:

@inproceedings{choi2023gread,
  title={GREAD: Graph Neural Reaction-Diffusion Networks},
  author={Choi, Jeongwhan and Hong, Seoyoung and Park, Noseong and Cho, Sung-Bae},
  booktitle={ICML},
  year={2023}
}

GREAD

Install / Use

README