Enhance DNN Adversarial Robustness and Efficiency via Injecting Noise to Non-Essential Neurons

TL;DR

This paper proposes a novel non-uniform noise injection method targeting non-essential neurons in DNNs to improve adversarial robustness and computational efficiency, outperforming uniform noise strategies.

Contribution

It introduces a non-uniform noise injection algorithm that selectively targets non-essential neurons, enhancing robustness and efficiency simultaneously.

Findings

Improves robustness against various adversarial attacks

Reduces computational costs of DNNs

Effective across multiple architectures and datasets

Abstract

Deep Neural Networks (DNNs) have revolutionized a wide range of industries, from healthcare and finance to automotive, by offering unparalleled capabilities in data analysis and decision-making. Despite their transforming impact, DNNs face two critical challenges: the vulnerability to adversarial attacks and the increasing computational costs associated with more complex and larger models. In this paper, we introduce an effective method designed to simultaneously enhance adversarial robustness and execution efficiency. Unlike prior studies that enhance robustness via uniformly injecting noise, we introduce a non-uniform noise injection algorithm, strategically applied at each DNN layer to disrupt adversarial perturbations introduced in attacks. By employing approximation techniques, our approach identifies and protects essential neurons while strategically introducing noise into non-essential neurons. Our experimental results demonstrate that our method successfully enhances both robustness and efficiency across several attack scenarios, model architectures, and datasets.