Increasing shape-bias in deep neural networks has been shown to improve robustness to common corruptions and noise. In this paper we analyze the adversarial robustness of texture and shape-biased models to Universal Adversarial Perturbations (UAPs). We use UAPs to evaluate the robustness of DNN models with varying degrees of shape-based training. We find that shape-biased models do not markedly improve adversarial robustness, and we show that ensembles of texture and shape-biased models can improve universal adversarial robustness while maintaining strong performance.
@article{arxiv.1911.10364,
title = {Universal Adversarial Robustness of Texture and Shape-Biased Models},
author = {Kenneth T. Co and Luis Muñoz-González and Leslie Kanthan and Ben Glocker and Emil C. Lupu},
journal= {arXiv preprint arXiv:1911.10364},
year = {2021}
}
Comments
In Proceedings of the 28th IEEE International Conference on Image Processing (ICIP 2021), code available at: https://github.com/kenny-co/sgd-uap-torch