Virtual Adversarial Training

Pytorch implementation of "Virtual Adversarial Training: a Regularization Method for Supervised and Semi-Supervised Learning" http://arxiv.org/abs/1704.03976

For reproducing semi-supervised learning results for SVHN with VAT loss:

python main.py --dataroot=<dataroot> --dataset=svhn --method=vat

For reproducing semi-supervised learning results for CIFAR10 with VAT loss:

python main.py --dataroot=<dataroot> --dataset=cifar10 --method=vat --num_epochs=500 --epoch_decay_start=460 --epsilon=10.0 --top_bn=False

For reproducing semi-supervised learning results for SVHN with VAT loss + Entropy loss:

python main.py --dataroot=<dataroot> --dataset=svhn --method=vatent

For reproducing semi-supervised learning results for CIFAR10 with VAT loss + Entropy loss:

python main.py --dataroot=<dataroot> --dataset=cifar10 --method=vatent --num_epochs=500 --epoch_decay_start=460 --epsilon=10.0 --top_bn=False