ydc123 / mup_attack

MUP-Attack

This repository is the official repository for our paper "Generating Adversarial Examples with Better Transferability via Masking Unimportant Parameters of Surrogate Model". This repository provides codes for performing MUP-based adversarial attack methods.

Running Commands

You can run the following command to perform the MUP-MIM attack method, using an Inception-v3 model as a surrogate model.

python attack.py -a inceptionv3 --attack_method MIFGSM --pruning_mode dynamic --p 0.15

Please refer to attack.py for more details.

Citation

If you benefit from our work in your research, please consider to cite the following paper:

@inproceedings{yang2023generating,
  author={Yang, Dingcheng and Yu, Wenjian and Xiao, Zihao and Luo, Jiaqi},
  booktitle={2023 International Joint Conference on Neural Networks (IJCNN)}, 
  title={Generating Adversarial Examples with Better Transferability via Masking Unimportant Parameters of Surrogate Model}, 
  year={2023},
  volume={},
  number={},
  pages={01-08},
  doi={10.1109/IJCNN54540.2023.10191679}      
}