Exploring the Landscape of Spatial Robustness
This work addresses the problem of spatial robustness for AI systems, highlighting it as a distinct and understudied setting compared to p-norm adversarial attacks.
The paper investigates the vulnerability of neural network classifiers to spatial perturbations like rotations and translations, finding that data augmentation provides limited robustness, but robust optimization and test-time input aggregation significantly improve it, with first-order methods failing to reliably find worst-case perturbations.
The study of adversarial robustness has so far largely focused on perturbations bound in p-norms. However, state-of-the-art models turn out to be also vulnerable to other, more natural classes of perturbations such as translations and rotations. In this work, we thoroughly investigate the vulnerability of neural network--based classifiers to rotations and translations. While data augmentation offers relatively small robustness, we use ideas from robust optimization and test-time input aggregation to significantly improve robustness. Finally we find that, in contrast to the p-norm case, first-order methods cannot reliably find worst-case perturbations. This highlights spatial robustness as a fundamentally different setting requiring additional study. Code available at https://github.com/MadryLab/adversarial_spatial and https://github.com/MadryLab/spatial-pytorch.