The code is repository for "DEFEAT: Decoupled Feature Attack Across Deep Neural Networks" (Neural Networks).

python 3.6
tensorflow 1.14

Run DEFEAT to generate adversarial examples: main.py.

We attack four normally trained models to generate adversarial examples, and test the transferability against defense models.

Layer Transferability

Standalone Experiment

Ensemble Experiment

If you find this project is useful for your research, please consider citing:

@article{huang2022defeat,
  title={DEFEAT: Decoupled feature attack across deep neural networks},
  author={Huang, Lifeng and Gao, Chengying and Liu, Ning},
  journal={Neural Networks},
  volume={156},
  pages={13--28},
  year={2022},
  publisher={Elsevier}
}