ASK-Attack and ASK-Defense
- Python: 3.7
- PyTorch: 1.5.0
ASK provides a novel Adversarial Soft kNN Attack, which has superior attack efficiency and accuracy degradation relative to previous kNN-based attacks. ASK also provides an Adversarial Soft kNN Defense that can improve the robustness of both kNN-based classifiers and DNNs.
ask_attack.py and ask_train.py correspond to attack and defense respectively. Run run_attack.ipynb to test.
If you use this code, please cite the following reference
@article{wang2021ask,
title={ASK: Adversarial Soft k-Nearest Neighbor Attack and Defense},
author={Wang, Ren and Chen, Tianqi and Yao, Philip and Liu, Sijia and Rajapakse, Indika and Hero, Alfred},
journal={arXiv preprint arXiv: 2106.14300},
year={2021}
}