Adversarial and Random Transformations for Robust Domain Adaptation and Generalization

TL;DR

This paper demonstrates that combining simple random data augmentation with a differentiable adversarial augmentation method significantly improves domain adaptation and generalization performance, robustness, and outperforms existing methods.

Contribution

It introduces a differentiable adversarial augmentation technique using spatial transformer networks and shows that combining it with random augmentation achieves state-of-the-art results.

Findings

State-of-the-art results on multiple DA and DG benchmarks

Enhanced robustness to data corruption

Simple random augmentation combined with adversarial methods outperforms complex search-based approaches

Abstract

Data augmentation has been widely used to improve generalization in training deep neural networks. Recent works show that using worst-case transformations or adversarial augmentation strategies can significantly improve the accuracy and robustness. However, due to the non-differentiable properties of image transformations, searching algorithms such as reinforcement learning or evolution strategy have to be applied, which are not computationally practical for large scale problems. In this work, we show that by simply applying consistency training with random data augmentation, state-of-the-art results on domain adaptation (DA) and generalization (DG) can be obtained. To further improve the accuracy and robustness with adversarial examples, we propose a differentiable adversarial data augmentation method based on spatial transformer networks (STN). The combined adversarial and random transformations based method outperforms the state-of-the-art on multiple DA and DG benchmark datasets. Besides, the proposed method shows desirable robustness to corruption, which is also validated on commonly used datasets.