Adversarial Dropout Regularization

TL;DR

This paper introduces Adversarial Dropout Regularization (ADR), a novel method for unsupervised domain adaptation that improves feature discriminativeness by replacing the critic with a dropout-based discriminator, leading to better target domain performance.

Contribution

The paper proposes ADR, a new adversarial regularization technique that enhances feature discrimination in domain adaptation by using dropout to identify non-discriminative features.

Findings

Significant improvement over state-of-the-art in image classification

Effective in semantic segmentation tasks

Applicable to semi-supervised learning with GANs

Abstract

We present a method for transferring neural representations from label-rich source domains to unlabeled target domains. Recent adversarial methods proposed for this task learn to align features across domains by fooling a special domain critic network. However, a drawback of this approach is that the critic simply labels the generated features as in-domain or not, without considering the boundaries between classes. This can lead to ambiguous features being generated near class boundaries, reducing target classification accuracy. We propose a novel approach, Adversarial Dropout Regularization (ADR), to encourage the generator to output more discriminative features for the target domain. Our key idea is to replace the critic with one that detects non-discriminative features, using dropout on the classifier network. The generator then learns to avoid these areas of the feature space and thus creates better features. We apply our ADR approach to the problem of unsupervised domain adaptation for image classification and semantic segmentation tasks, and demonstrate significant improvement over the state of the art. We also show that our approach can be used to train Generative Adversarial Networks for semi-supervised learning.