On the Mechanisms of Adversarial Data Augmentation for Robust and Adaptive Transfer Learning

TL;DR

This paper investigates how adversarial perturbations can serve as constructive data augmentation to improve robust transfer learning across domains with distribution shift. Using an information-theoretic lens based on the Information Bottleneck, it shows that adversarial data augmentation compresses domain-specific information while promoting domain-invariant representations and flatter loss landscapes. The authors propose a unified framework that integrates ADA with consistency regularization and domain alignment, and validate it on benchmarks including VisDA, DomainNet, and Office-Home under unsupervised and few-shot adaptation, reporting consistent gains. The work reframes adversarial signals as principled regularizers for cross-domain generalization and outlines future directions for multi-modal, temporal, and physics-informed perturbations.

Abstract

Transfer learning across domains with distribution shift remains a fundamental challenge in building robust and adaptable machine learning systems. While adversarial perturbations are traditionally viewed as threats that expose model vulnerabilities, recent studies suggest that they can also serve as constructive tools for data augmentation. In this work, we systematically investigate the role of adversarial data augmentation (ADA) in enhancing both robustness and adaptivity in transfer learning settings. We analyze how adversarial examples, when used strategically during training, improve domain generalization by enriching decision boundaries and reducing overfitting to source-domain-specific features. We further propose a unified framework that integrates ADA with consistency regularization and domain-invariant representation learning. Extensive experiments across multiple benchmark datasets -- including VisDA, DomainNet, and Office-Home -- demonstrate that our method consistently improves target-domain performance under both unsupervised and few-shot domain adaptation settings. Our results highlight a constructive perspective of adversarial learning, transforming perturbation from a destructive attack into a regularizing force for cross-domain transferability.