Colored Noise Injection for Training Adversarially Robust Neural Networks

TL;DR

This paper introduces colored noise injection during adversarial training to enhance neural network robustness, outperforming previous white noise methods against various attacks on CIFAR datasets.

Contribution

It extends white Gaussian noise injection to colored noise, demonstrating improved adversarial robustness and providing extensive ablation analysis for optimal configurations.

Findings

Outperforms previous noise injection methods in adversarial accuracy

Effective against both white-box and black-box attacks

Validated on CIFAR-10 and CIFAR-100 datasets

Abstract

Even though deep learning has shown unmatched performance on various tasks, neural networks have been shown to be vulnerable to small adversarial perturbations of the input that lead to significant performance degradation. In this work we extend the idea of adding white Gaussian noise to the network weights and activations during adversarial training (PNI) to the injection of colored noise for defense against common white-box and black-box attacks. We show that our approach outperforms PNI and various previous approaches in terms of adversarial accuracy on CIFAR-10 and CIFAR-100 datasets. In addition, we provide an extensive ablation study of the proposed method justifying the chosen configurations.