This repository contains the code of "Self-Ensemble Adversarial Training for Improved Robustness" published at ICASSP 2024.
The development environment is:
- Python (3.10)
- Pytorch (1.13)
- CUDA
- Torchvision (0.14)
- Torchattacks (3.3)
To train a ResNet-18 on CIFAR-10 using PGD-10:
python meat.py
To train a WideResNet-34-10 on CIFAR-10 using PGD-10:
python meat.py --arch 'WRN'