LP-BFGS attack: An adversarial attack based on Hessian with limited pixels

Official implementation for

• LP-BFGS attack: An adversarial attack based on Hessian with limited pixels.

For any questions, contact (zhangjiebao2014@mail.ynu.edu.cn).

Requirements

1. Python
2. Pytorch
3. Torattacks >= 3.2.6
4. Torchvision
5. Pytorchcv
6. pytorch-minimize
7. captum, IntegratedGradient implementation by Pytorch

Preparations

• some datasets will be downloaded.

Attack

python attack_models.py

Citation

If you find this repo useful for your research, please consider citing the paper

@misc{lp-bfgs-attack,
  doi = {10.48550/arXiv.2210.15446},
  url = {https://arxiv.org/abs/2210.15446},
  author = {Zhang, Jiebao and Qian, Wenhua and Nie, Rencan and Cao, Jinde and Xu, Dan},
  title = {LP-BFGS attack: An adversarial attack based on the Hessian with limited pixels},
  publisher = {arXiv},
  year = {2022},
}