Enhancing Output Diversity Improves Conjugate Gradient-based Adversarial Attacks

TL;DR

This paper introduces ReACG, an improved adversarial attack method based on conjugate gradient, which increases output diversity by adjusting search parameters, leading to more effective attacks especially on complex ImageNet models.

Contribution

Proposes ReACG, a novel modification of ACG that automatically enhances output diversity by rescaling search components, improving attack success rates.

Findings

ReACG outperforms ACG in attack effectiveness.

Increasing the distance between search points boosts output diversity.

ReACG is particularly effective on ImageNet models with many classes.

Abstract

Deep neural networks are vulnerable to adversarial examples, and adversarial attacks that generate adversarial examples have been studied in this context. Existing studies imply that increasing the diversity of model outputs contributes to improving the attack performance. This study focuses on the Auto Conjugate Gradient (ACG) attack, which is inspired by the conjugate gradient method and has a high diversification performance. We hypothesized that increasing the distance between two consecutive search points would enhance the output diversity. To test our hypothesis, we propose Rescaling-ACG (ReACG), which automatically modifies the two components that significantly affect the distance between two consecutive search points, including the search direction and step size. ReACG showed higher attack performance than that of ACG, and is particularly effective for ImageNet models with several classification classes. Experimental results show that the distance between two consecutive search points enhances the output diversity and may help develop new potent attacks. The code is available at \url{https://github.com/yamamura-k/ReACG}