RobGC

Introduction

This repository is the implementation of paper: RobGC: Towards Robust Graph Condensation.

Data

The code is based on the graph condnesation method GGond. For cora, citeseer and pubmed, the code will directly download them; so no extra script is needed. For Ogbn-arxiv, we use the datasets provided by GraphSAINT. They are available on Google Drive link (alternatively, BaiduYun link (code: f1ao)). Rename the folder to ./data at the root directory. Note that the links are provided by GraphSAINT team.

Model training

To generate the noisy data and condense the graph:

$ bash ./scr/attack.sh
$ bash ./scr/run.sh

Aknowledgements

We express our gratitude for the contributions of GGond and DeepRobust in their remarkable work.