This is a short summary of the AdverTorch project that I lead. It is a Python toolbox for adversarial robustness research. The primary functionalities are implemented in PyTorch. Specifically, AdverTorch contains modules for generating adversarial perturbations and defending against adversarial examples, also scripts for adversarial training.

The github repo is here.

The tech report is here.

We also have a blog post here on BorealisAI website.