DPGBA: Rethinking Graph Backdoor Attacks: A Distribution-Preserving Perspective (KDD 2024)

ENVIRONMENTS

The packages can be installed by directly run the commands in install.sh by

bash install.sh

RUN

bash script/train_DPGBA.sh

NOTES

1. Set 'defense_mode=reconstruct' to introduce outlier detection; set 'defense_mode=none' for the case of no defense method.

2. When there are no defense methods, increasing the value of 'weight_target' for $\mathcal{L}_T$ and 'weight_targetclass' for $\mathcal{L}_E$ will enhance the attack performance.

3. When an outlier detection method is adopted, please also tune the parameter 'weight_ood' for $\mathcal{L}_D$ to achieve a balance between stealthiness and attack performance.

If you find this repo to be useful, please consider cite our paper. Thank you.

@inproceedings{zhang2024rethinking,
  title={Rethinking graph backdoor attacks: A distribution-preserving perspective},
  author={Zhang, Zhiwei and Lin, Minhua and Dai, Enyan and Wang, Suhang},
  booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  pages={4386--4397},
  year={2024}
}

The code is built on UGBA.