Understanding the Intrinsic Robustness of Image Distributions using Conditional Generative Models

Abstract

Starting with Gilmer et al. (2018), several works have demonstrated the inevitability of adversarial examples based on different assumptions about the underlying input probability space. It remains unclear, however, whether these results apply to natural image distributions. In this work, we assume the underlying data distribution is captured by some conditional generative model, and prove intrinsic robustness bounds for a general class of classifiers, which solves an open problem in Fawzi et al. (2018). Building upon the state-of-the-art conditional generative models, we study the intrinsic robustness of two common image benchmarks under ${\ell }_2$ perturbations, and show the existence of a large gap between the robustness limits implied by our theory and the adversarial robustness achieved by current state-of-the-art robust models. Code for all our experiments is available at https://github.com/xiaozhanguva/Intrinsic-Rob.

arXiv comment14 pages, 2 figures, 5 tables, AISTATS final paper reformatted for readability

References in corpus (10)

• Conditional Generative Adversarial Nets
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Conditional Image Synthesis With Auxiliary Classifier GANs
• Provable defenses against adversarial examples via the convex outer adversarial polytope
• Adversarial Attacks and Defences: A Survey
• Evaluating Robustness of Neural Networks with Mixed Integer Programming
• On the Convergence and Robustness of Adversarial Training
• On Adaptive Attacks to Adversarial Example Defenses
• Lower Bounds on Adversarial Robustness from Optimal Transport
• Generalized No Free Lunch Theorem for Adversarial Robustness

Cited by in corpus (1)

• A Closer Look at Accuracy vs. Robustness