Universal adversarial perturbations

TL;DR

This paper demonstrates the existence of small, universal perturbations that can fool deep neural networks across many images, revealing vulnerabilities and security concerns in current classifiers.

Contribution

It introduces a systematic algorithm for computing universal adversarial perturbations and shows their high transferability across different neural networks.

Findings

Universal perturbations cause misclassification with high probability.

Deep neural networks are highly vulnerable to these perturbations.

Universal perturbations generalize well across different models.

Abstract

Given a state-of-the-art deep neural network classifier, we show the existence of a universal (image-agnostic) and very small perturbation vector that causes natural images to be misclassified with high probability. We propose a systematic algorithm for computing universal perturbations, and show that state-of-the-art deep neural networks are highly vulnerable to such perturbations, albeit being quasi-imperceptible to the human eye. We further empirically analyze these universal perturbations and show, in particular, that they generalize very well across neural networks. The surprising existence of universal perturbations reveals important geometric correlations among the high-dimensional decision boundary of classifiers. It further outlines potential security breaches with the existence of single directions in the input space that adversaries can possibly exploit to break a classifier on most natural images.