GAP++: Learning to generate target-conditioned adversarial examples

TL;DR

GAP++ introduces a target-conditioned adversarial attack framework that leverages both input images and target labels, achieving higher success rates and smaller perturbations compared to previous methods.

Contribution

The paper presents a novel framework for target-conditioned adversarial attacks that considers both input images and target labels, improving attack effectiveness.

Findings

Achieves higher fooling rates on MNIST and CIFAR10.

Generates smaller perturbations with high success.

Outperforms existing single-target attack models.

Abstract

Adversarial examples are perturbed inputs which can cause a serious threat for machine learning models. Finding these perturbations is such a hard task that we can only use the iterative methods to traverse. For computational efficiency, recent works use adversarial generative networks to model the distribution of both the universal or image-dependent perturbations directly. However, these methods generate perturbations only rely on input images. In this work, we propose a more general-purpose framework which infers target-conditioned perturbations dependent on both input image and target label. Different from previous single-target attack models, our model can conduct target-conditioned attacks by learning the relations of attack target and the semantics in image. Using extensive experiments on the datasets of MNIST and CIFAR10, we show that our method achieves superior performance with single target attack models and obtains high fooling rates with small perturbation norms.