Generating Adversarial Perturbation with Root Mean Square Gradient

TL;DR

This paper proposes a method for generating adversarial perturbations in image classification by utilizing the root mean square of gradients to craft more effective adversarial examples.

Contribution

It introduces a novel approach that leverages root mean square gradient calculations to improve adversarial perturbation generation.

Findings

Enhanced adversarial attack effectiveness

Improved robustness of generated perturbations

Potential applications in adversarial defense strategies

Abstract

We focus our attention on the problem of generating adversarial perturbations based on the gradient in image classification domain