Exploiting Vulnerabilities in Deep Neural Networks: Adversarial and Fault-Injection Attacks

From tiny pacemaker chips to aircraft collision avoidance systems, the\nstate-of-the-art Cyber-Physical Systems (CPS) have increasingly started to rely\non Deep Neural Networks (DNNs). However, as concluded in various studies, DNNs\nare highly susceptible to security threats, including adversarial attacks. In\nthis paper, we first discuss different vulnerabilities that can be exploited\nfor generating security attacks for neural network-based systems. We then\nprovide an overview of existing adversarial and fault-injection-based attacks\non DNNs. We also present a brief analysis to highlight different challenges in\nthe practical implementation of adversarial attacks. Finally, we also discuss\nvarious prospective ways to develop robust DNN-based systems that are resilient\nto adversarial and fault-injection attacks.\n