Securing DNN for smart vehicles: an overview of adversarial attacks, defenses, and frameworks

Abstract

AbstractRecently, many applications have begun to employ deep neural networks (DNN), such as image recognition and safety-critical applications, for more accurate results. One of the most important critical applications of DNNs is in smart autonomous vehicles. The operative principles of autonomous vehicles depend heavily on their ability to collect data from the environment via integrated sensors, then employ DNN classification to interpret them and make operative decisions. The security and the reliability of DNNs raise many challenges and concerns for researchers. One of those challenges currently in the research domain is the threat of adversarial attacks on DNNs. In this survey, we present state-of-the-art research on DNN frameworks, adversarial attacks, and defenses. We discuss each work along with its advantages and limitations and present our thoughts on and future directions for adversarial attacks and defenses.