Skip to content

Boosting adversarial attack with AdaGrad, AdaDelta, RMSProp, Adam and more...

License

Notifications You must be signed in to change notification settings

jasonliuuu/AI-FGSM

Repository files navigation

SI-AI-FGSM

🚧 WIP

Boosting adversarial attack with AdaGrad, AdaDelta, RMSProp, Adam and more...

Requirements

  • Python 3.6.5
  • Tensorflow 1.12.0
  • Numpy 1.15.4
  • opencv-python 3.4.2.16
  • scipy 1.1.0

Experiments

Download the data and pretrained models

Running

  • python agi_fgsm.py to generate adversarial examples for inception_v3 using AGI-FGSM;
  • python ri_fgsm.py to generate adversarial examples for inception_v3 using RI-FGSM;
  • python ai_fgsm.py to generate adversarial examples for inception_v3 using AI-FGSM;
  • python si_agi_fgsm.py to generate adversarial examples for inception_v3 using SI-AGI-FGSM;
  • python si_ri_fgsm.py to generate adversarial examples for inception_v3 using SI-RI-FGSM;
  • python si_ai_fgsm.py to generate adversarial examples for inception_v3 using SI-AI-FGSM;
  • python si_agi_ti_dim.py to generate adversarial examples for inception_v3 using SI-AGI-TI-DIM;
  • python si_ri_ti_dim.py to generate adversarial examples for inception_v3 using SI-RI-TI-DIM;
  • python si_ai_ti_dim.py to generate adversarial examples for inception_v3 using SI-AI-TI-DIM;
  • python simple_eval.py: evaluate the attack success rate under 8 models including normal training models and adversarial training models.

Acknowledgements

Code refers to NAG attack

About

Boosting adversarial attack with AdaGrad, AdaDelta, RMSProp, Adam and more...

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages