A Survey of Attacks and Defenses for Deep Neural Networks

Abstract

This survey provides an overview of adversarial attacks and defenses for deep neural networks. We discuss the taxonomies of attacks on Machine learning systems and common algorithms for generating attacks. We also present a taxonomy of defense techniques for adversarial machine learning. Using the information in this paper, researchers can make an informed decision on creating secure models in machine learning. Based on the reviewed literature, we foresee promising paths for future research.