Guided Adversarial Attack for Evaluating and Enhancing Adversarial Defenses

TL;DR

This paper introduces a guided adversarial attack method that improves attack strength and efficiency, and a corresponding training approach that enhances defense robustness against adversarial attacks.

Contribution

It proposes GAMA, a guided attack using image mapping for stronger attacks, and GAT, a training method leveraging this attack for improved robustness.

Findings

GAMA outperforms existing attacks on multiple defenses.

GAT achieves state-of-the-art results among single-step defenses.

The relaxation term enhances attack efficacy and training efficiency.

Abstract

Advances in the development of adversarial attacks have been fundamental to the progress of adversarial defense research. Efficient and effective attacks are crucial for reliable evaluation of defenses, and also for developing robust models. Adversarial attacks are often generated by maximizing standard losses such as the cross-entropy loss or maximum-margin loss within a constraint set using Projected Gradient Descent (PGD). In this work, we introduce a relaxation term to the standard loss, that finds more suitable gradient-directions, increases attack efficacy and leads to more efficient adversarial training. We propose Guided Adversarial Margin Attack (GAMA), which utilizes function mapping of the clean image to guide the generation of adversaries, thereby resulting in stronger attacks. We evaluate our attack against multiple defenses and show improved performance when compared to existing attacks. Further, we propose Guided Adversarial Training (GAT), which achieves state-of-the-art performance amongst single-step defenses by utilizing the proposed relaxation term for both attack generation and training.