TODO:
- Update target model
- Increase LR for generator and decrease LR for disc (currently the discriminator is training too fast)
- Change loss function(WLoss)
a Pytorch implementation of the paper "Generating Adversarial Examples with Adversarial Networks" (advGAN).
python3 train_target_model.py
python3 main.py
python3 test_adversarial_examples.py
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.