Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning

13 Apr 2017Takeru MiyatoShin-ichi MaedaMasanori KoyamaShin Ishii

We propose a new regularization method based on virtual adversarial loss: a new measure of local smoothness of the conditional label distribution given input. Virtual adversarial loss is defined as the robustness of the conditional label distribution around each input data point against local perturbation... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Semi-Supervised Image Classification CIFAR-10, 4000 Labels VAT+EntMin Accuracy 89.45 # 5
Semi-Supervised Image Classification CIFAR-10, 4000 Labels VAT Accuracy 88.64 # 6
Semi-Supervised Image Classification SVHN, 1000 labels VAT Accuracy 94.58 # 6
Semi-Supervised Image Classification SVHN, 1000 labels VAT+EntMin Accuracy 96.14 # 3