Distributional Smoothing with Virtual Adversarial Training

TL;DR

This paper introduces Virtual Adversarial Training (VAT), a regularization technique that enhances model smoothness by measuring local distributional robustness, applicable to semi-supervised learning with low computational cost.

Contribution

The paper proposes LDS-based regularization called VAT, which determines adversarial directions from the model distribution alone, improving semi-supervised learning performance efficiently.

Findings

VAT outperforms most existing methods on MNIST, SVHN, and NORB datasets.

VAT achieves competitive results without using label information for adversarial direction.

The computational cost of VAT is low, requiring only a few forward and backward passes.

Abstract

We propose local distributional smoothness (LDS), a new notion of smoothness for statistical model that can be used as a regularization term to promote the smoothness of the model distribution. We named the LDS based regularization as virtual adversarial training (VAT). The LDS of a model at an input datapoint is defined as the KL-divergence based robustness of the model distribution against local perturbation around the datapoint. VAT resembles adversarial training, but distinguishes itself in that it determines the adversarial direction from the model distribution alone without using the label information, making it applicable to semi-supervised learning. The computational cost for VAT is relatively low. For neural network, the approximated gradient of the LDS can be computed with no more than three pairs of forward and back propagations. When we applied our technique to supervised and semi-supervised learning for the MNIST dataset, it outperformed all the training methods other than the current state of the art method, which is based on a highly advanced generative model. We also applied our method to SVHN and NORB, and confirmed our method's superior performance over the current state of the art semi-supervised method applied to these datasets.