Abstract
The discovery of adversarial examples revealed one of the most basic vulnerabilities of deep neural networks. Among the variety of techniques introduced to tackle this inherent weakness, adversarial training was shown to be the most common and efficient strategy to achieve robustness. It is usually done by balancing the robust and natural losses. In this work, we aim to achieve better trade-off between robust and natural performances by enforcing a domain invariant feature representation. We present a new adversarial training method, called Domain Invariant Adversarial Learning (DIAL) that learns a feature representation which is both robust and domain invariant. DIAL uses a variant of Domain Adversarial Neural Network (DANN) on the natural domain and its corresponding adversarial domain. In a case where the source domain consists of natural examples and the target domain is the adversarially perturbed examples, our method learns a feature representation constrained not to discriminate between the natural and adversarial examples, and can therefore achieve better representation. We demonstrate our advantage by improving both robustness and natural accuracy compared to current state-of-the-art adversarial training methods.
Abstract (translated)
URL
https://arxiv.org/abs/2104.00322
