Towards Feature Space Adversarial Attack

TL;DR

This paper introduces a novel adversarial attack method that perturbs style-related features in images, producing more natural-looking adversarial examples and revealing limitations of current defenses.

Contribution

It presents a new attack focusing on style features rather than pixels, demonstrating its effectiveness and exposing vulnerabilities in existing defense methods.

Findings

Generates more natural adversarial images than previous methods

Style-based attack bypasses current pixel-space defenses

Style perturbations can cause misclassification with imperceptible changes

Abstract

We propose a new adversarial attack to Deep Neural Networks for image classification. Different from most existing attacks that directly perturb input pixels, our attack focuses on perturbing abstract features, more specifically, features that denote styles, including interpretable styles such as vivid colors and sharp outlines, and uninterpretable ones. It induces model misclassfication by injecting imperceptible style changes through an optimization procedure. We show that our attack can generate adversarial samples that are more natural-looking than the state-of-the-art unbounded attacks. The experiment also supports that existing pixel-space adversarial attack detection and defense techniques can hardly ensure robustness in the style related feature space.