Adversarial Training against Location-Optimized Adversarial Patches
This work addresses a practical security issue for deep learning systems by enhancing defense against adversarial patches, which are easier to deploy in the physical world compared to imperceptible adversarial examples.
The paper tackled the problem of adversarial patches, which are visible rectangular patches that cause misclassification in image classifiers, by developing a method to optimize patch locations and applying adversarial training to improve robustness on CIFAR10 and GTSRB datasets, achieving significant robustness gains without reducing accuracy.
Deep neural networks have been shown to be susceptible to adversarial examples -- small, imperceptible changes constructed to cause mis-classification in otherwise highly accurate image classifiers. As a practical alternative, recent work proposed so-called adversarial patches: clearly visible, but adversarially crafted rectangular patches in images. These patches can easily be printed and applied in the physical world. While defenses against imperceptible adversarial examples have been studied extensively, robustness against adversarial patches is poorly understood. In this work, we first devise a practical approach to obtain adversarial patches while actively optimizing their location within the image. Then, we apply adversarial training on these location-optimized adversarial patches and demonstrate significantly improved robustness on CIFAR10 and GTSRB. Additionally, in contrast to adversarial training on imperceptible adversarial examples, our adversarial patch training does not reduce accuracy.