TAROT: Towards Essentially Domain-Invariant Robustness with Theoretical Justification

2025-05-10

AI Summary

This work addresses robust domain adaptation under adversarial attacks, proposing the first theoretically grounded framework for domain-invariant robust learning. To overcome the lack of theoretical guarantees and the absence of a unified notion of essential domain invariance in existing methods, we derive the first generalization upper bound on target-domain robust risk and introduce a novel robust domain divergence metric, establishing its theoretical connection to domain-invariant representations. Based on this bound, we design TAROT—a unified algorithm that jointly optimizes robust feature alignment, domain-invariant representation learning, and adversarial training. Extensive experiments on benchmarks including DomainNet demonstrate significant improvements over state-of-the-art methods: average robust accuracy increases by 4.2%, while cross-domain generalization capability and model scalability are simultaneously enhanced.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Responsible Web: Algorithmic accountability and transparency on the web
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Robust domain adaptation against adversarial attacks is a critical research area that aims to develop models capable of maintaining consistent performance across diverse and challenging domains. In this paper, we derive a new generalization bound for robust risk on the target domain using a novel divergence measure specifically designed for robust domain adaptation. Building upon this, we propose a new algorithm named TAROT, which is designed to enhance both domain adaptability and robustness. Through extensive experiments, TAROT not only surpasses state-of-the-art methods in accuracy and robustness but also significantly enhances domain generalization and scalability by effectively learning domain-invariant features. In particular, TAROT achieves superior performance on the challenging DomainNet dataset, demonstrating its ability to learn domain-invariant representations that generalize well across different domains, including unseen ones. These results highlight the broader applicability of our approach in real-world domain adaptation scenarios.

Problem

• Enhancing domain adaptability and robustness against adversarial attacks
• Developing models for consistent performance across diverse domains
• Learning domain-invariant features for better generalization and scalability

Innovation

• Novel divergence measure for robust domain adaptation
• TAROT algorithm enhances domain adaptability and robustness
• Learns domain-invariant features for superior generalization