Universal Adversarial Perturbations

TLDR

The study demonstrates that a tiny, image‑agnostic perturbation can reliably fool state‑of‑the‑art neural network classifiers. The authors present a systematic algorithm that efficiently computes such universal perturbations, revealing the networks’ susceptibility. Empirical results show that these perturbations generalize across different networks, expose geometric correlations in decision boundaries, and highlight potential security risks.

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.