Adversarial Attacks with Time-Scale Representations

TL;DR

This paper introduces a wavelet-based adversarial attack framework that disrupts early convolutional layers in deep learning models more effectively than traditional time-domain methods, with potential implications for model robustness.

Contribution

The paper presents a novel wavelet space perturbation method for black-box adversarial attacks, including a theoretical dual mapping framework and empirical validation against defenses.

Findings

Wavelet-based perturbations outperform time-based attacks.

The method effectively disrupts early convolutional layers.

It remains effective against various defense mechanisms.

Abstract

We propose a novel framework for real-time black-box universal attacks which disrupts activations of early convolutional layers in deep learning models. Our hypothesis is that perturbations produced in the wavelet space disrupt early convolutional layers more effectively than perturbations performed in the time domain. The main challenge in adversarial attacks is to preserve low frequency image content while minimally changing the most meaningful high frequency content. To address this, we formulate an optimization problem using time-scale (wavelet) representations as a dual space in three steps. First, we project original images into orthonormal sub-spaces for low and high scales via wavelet coefficients. Second, we perturb wavelet coefficients for high scale projection using a generator network. Third, we generate new adversarial images by projecting back the original coefficients from the low scale and the perturbed coefficients from the high scale sub-space. We provide a theoretical framework that guarantees a dual mapping from time and time-scale domain representations. We compare our results with state-of-the-art black-box attacks from generative-based and gradient-based models. We also verify efficacy against multiple defense methods such as JPEG compression, Guided Denoiser and Comdefend. Our results show that wavelet-based perturbations consistently outperform time-based attacks thus providing new insights into vulnerabilities of deep learning models and could potentially lead to robust architectures or new defense and attack mechanisms by leveraging time-scale representations.