Attribute-Guided Adversarial Training for Robustness to Natural Perturbations
This work aims to improve the robustness of deep neural networks against naturally occurring, real-world perturbations for practitioners deploying models in diverse environments, representing an incremental step in adversarial robustness research.
This paper addresses the problem of making deep neural networks robust to natural perturbations, such as object-related shifts, geometric transformations, and image corruptions, without direct access to test domain data. The authors propose an attribute-guided adversarial training method that generates samples to maximize classifier exposure to attribute space, resulting in robustness gains on MNIST, CIFAR-10, and a CLEVR variant.
While existing work in robust deep learning has focused on small pixel-level norm-based perturbations, this may not account for perturbations encountered in several real-world settings. In many such cases although test data might not be available, broad specifications about the types of perturbations (such as an unknown degree of rotation) may be known. We consider a setup where robustness is expected over an unseen test domain that is not i.i.d. but deviates from the training domain. While this deviation may not be exactly known, its broad characterization is specified a priori, in terms of attributes. We propose an adversarial training approach which learns to generate new samples so as to maximize exposure of the classifier to the attributes-space, without having access to the data from the test domain. Our adversarial training solves a min-max optimization problem, with the inner maximization generating adversarial perturbations, and the outer minimization finding model parameters by optimizing the loss on adversarial perturbations generated from the inner maximization. We demonstrate the applicability of our approach on three types of naturally occurring perturbations -- object-related shifts, geometric transformations, and common image corruptions. Our approach enables deep neural networks to be robust against a wide range of naturally occurring perturbations. We demonstrate the usefulness of the proposed approach by showing the robustness gains of deep neural networks trained using our adversarial training on MNIST, CIFAR-10, and a new variant of the CLEVR dataset.