Attacks on the neural network and defense methods

TL;DR

This paper examines various attack methods on neural networks trained on audio data, evaluates defense strategies like Art-IBM and advertorch, and presents accuracy metrics under attack conditions.

Contribution

It provides a comparative analysis of attack techniques and defense methods specifically for audio-based neural networks, including experimental accuracy results.

Findings

FGSM, PGD, and CW attacks significantly reduce accuracy.

Defense methods like Art-IBM and advertorch improve robustness.

Experimental results demonstrate effectiveness of certain defenses.

Abstract

This article will discuss the use of attacks on a neural network trained on audio data, as well as possible methods of protection against these attacks. FGSM, PGD and CW attacks, as well as data poisoning, will be considered. Within the framework of protection, Art-IBM and advertorch libraries will be considered. The obtained accuracy metrics within the framework of attack applications are presented