WideResNet implementation on MNIST dataset. FGSM and PGD adversarial attacks on standard training, PGD adversarial training, and Feature Scattering adversarial training.
For standard training and PGD adversarial training use
bash run.sh
It automatically executes main.py with additional arguments like no. of iteration, epsilon value, max iterations for attack and step size in each attack. After training the model it executes FGSM and PGD attacks on it.
For feature scattering based adversarial training use
bash fs_run.sh
Does the same previous one but implements feature scattering based adversarial training.