zknus / Robustness-of-Graph-Neural-Diffusion

On the Robustness of Graph Neural Diffusion to Topology Perturbations

This repository contains the implementation to reproduce the numerical experiments of the NEURIPS 2022 paper On the Robustness of Graph Neural Diffusion to Topology Perturbations

Running the experiments

The codes are written based on

• https://github.com/twitter-research/graph-neural-pde
• https://github.com/THUDM/grb

Experiments

For example to run

cd src
python run_GNN2.py 

Saved models are available in ./saved_models2.

Requirements

• scipy==1.5.2
• numpy==1.19.1
• torch==1.8.0
• networkx==2.5
• pandas~=1.2.3
• cogdl~=0.3.0.post1
• torch-cluster==1.5.9
• torch-geometric==1.7.0
• torch-scatter==2.0.6
• torch-sparse==0.6.9
• torch-spline-conv==1.2.1
• torchdiffeq==0.2.1

Citation

If you found our work useful in your research, please cite our paper at:

@INPROCEEDINGS{SonKanWan:C22,
		author      = {Yang Song and Qiyu Kang and Sijie Wang and Kai Zhao and Wee Peng Tay},
		title       = {On the Robustness of Graph Neural Diffusion to Topology Perturbations},
		booktitle   = {Advances in Neural Information Processing Systems (NeurIPS)},
		month       = {Nov.},
		year        = {2022},
		address     = {New Orleans, USA},
		}

(Also consider starring the project on GitHub.)