This is the official code for "Theoretical Understanding of Learning from Adversarial Perturbations" [S. Kumano et al., ICLR24].

All generated data, including artificial datasets, model parameters, and adversarial examples, can be downloaded from here.

Setup

docker-compose -f "docker/docker-compose.yaml" up -d --build 

Run

bash/artificial.sh <gpu_id: int>
bash/train_natural.sh <gpu_id: int>
bash/create.sh <gpu_id: int>
bash/train_perturbation.sh <gpu_id: int>