/advGAN_pytorch

a Pytorch implementation of the paper "Generating Adversarial Examples with Adversarial Networks" (advGAN).

Primary LanguagePython

Updated for CIFAR-10

TODO:

  • Update target model
  • Increase LR for generator and decrease LR for disc (currently the discriminator is training too fast)
  • Change loss function(WLoss)

Original Readme

advGAN_pytorch

a Pytorch implementation of the paper "Generating Adversarial Examples with Adversarial Networks" (advGAN).

training the target model

python3 train_target_model.py

training the advGAN

python3 main.py

testing adversarial examples

python3 test_adversarial_examples.py

results

attack success rate in the MNIST test set: 99%

NOTE: My implementation is a little different from the paper, because I add a clipping trick.