Machine learning security attacks and defense approaches for emerging cyber physical applications: A comprehensive survey
The cyber physical systems integrate the sensing, computation, control and networking processes into physical objects and infrastructure, which are connected through the Internet to execute a common task. Cyber physical systems can be applied in various applications, like healthcare, transportation, industrial production, environment and sustainability, and security and surveillance. However, the tight coupling of cyber systems with physical systems introduce challenges in addressing stability, security, efficiency and reliability. The machine learning (ML) security is the inclusion of cyber security mechanism to provide protection to the machine learning models against various cyber attacks. The ML models work through the traditional training and testing approaches. However, execution of such kind of approaches may not function effectively in case if a system is connected to the Internet. As online hackers can exploit deployed security mechanisms and poison the data. This data is then taken as the input by the ML models. In this article, we provide the details of various machine learning security attacks in cyber physical systems. We then discuss some defense mechanisms to protect against these attacks. We also present a threat model of ML security mechanisms deployed in cyber systems. Furthermore, we discuss various issues and challenges of ML security mechanisms deployed in cyber systems. Finally, we provide a detailed comparative study on performance of the ML models under the influence of various ML attacks in cyber physical systems.
Authentication, Cyber physical systems (CPS), Intrusion detection, Machine learning (ML) security, Privacy and security
J. Sing, M. Wazid, A.K. Das, V. Chamola and M. Guizani, "Machine learning security attacks and defense approaches for emerging cyber physical applications: A comprehensive survey", in Computer Communications, vol. 192 (1), pp. 316-331, Aug 2022, doi: 10.1016/j.comcom.2022.06.012