Attribute-Guided Adversarial Training for Robustness to Natural Perturbations

Authors

  • Tejas Gokhale Arizona State University
  • Rushil Anirudh Lawrence Livermore National Laboratory
  • Bhavya Kailkhura Lawrence Livermore National Laboratory
  • Jayaraman J. Thiagarajan Lawrence Livermore National Laboratory
  • Chitta Baral Arizona State University
  • Yezhou Yang Arizona State University

Keywords:

Adversarial Learning & Robustness

Abstract

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.

Downloads

Published

2021-05-18

How to Cite

Gokhale, T., Anirudh, R., Kailkhura, B., J. Thiagarajan, J., Baral, C., & Yang, Y. (2021). Attribute-Guided Adversarial Training for Robustness to Natural Perturbations. Proceedings of the AAAI Conference on Artificial Intelligence, 35(9), 7574-7582. Retrieved from https://ojs.aaai.org/index.php/AAAI/article/view/16927

Issue

Section

AAAI Technical Track on Machine Learning II