Mark12Ding / GNN-Practical-Attack

Code for Towards More Practical Adversarial Attacks on Graph Neural Networks (NeurIPS 2020)

Towards More Practical Adversarial Attacks on Graph Neural Networks

This repo provides the official implementations for the experiments described in the following paper:

Towards More Practical Adversarial Attacks on Graph Neural Networks

Jiaqi Ma*, Shuangrui Ding*, and Qiaozhu Mei. NeurIPS 2020.

(*: equal constribution)

A previous version of this paper is appeared with the title Black-Box Adversarial Attacks on Graph Neural Networks with Limited Node Access.

Update: The reader is encouraged to look at this repo, which is a follow-up work published in WSDM 2022. This follow-up work has an improved experiment setup as well as a full implementation of RWCS and GC-RWCS.

Requirements

• dgl 0.4.2
• torch 1.4.0
• networkx 2.3
• numpy 1.16.4

Run the code

Example command to run the code: python main.py --dataset cora --model JKNetMaxpool --threshold 0.1 --steps 4.

Cite

@inproceedings{ma2020practical,
  title={Towards More Practical Adversarial Attacks on Graph Neural Networks},
  author={Ma, Jiaqi and Ding, Shuangrui and Mei, Qiaozhu},
  booktitle={Advances in Neural Information Processing Systems},
  year={2020}
}