This repo contains my implementation for adversarial attacks coding pratice in the Trustworthy Machine Learning course in Peking University directed by Yisen Wang, spring 2021. I implement 3 classical adversarial attack methods in Pytorch:
Then I pratice both whitebox and blackbox attacks on neural networks trained for CIFAR10 and MNIST classification. The target model for CIFAR10 is PreActResNet18 and the model for MNIST is a small CNN. For more details, please see the report.