Machine Learning for CPS Security: Applications, Challenges and Recommendations

Machine Learning (ML) based approaches are becoming increasingly common for securing critical Cyber Physical Systems (CPS), such as electric power grid and water treatment plants. CPS is a combination of physical processes (e.g., water, electricity, etc.) and computing elements (e.g., computers, communication networks, etc.). ML techniques are a class of algorithms that learn mathematical relationships of a system from data. Applications of ML in securing CPS is commonly carried out on data from a real system. However, there are significant challenges in using ML algorithms as it is for security purposes. In this chapter, two case studies based on empirical applications of ML for the CPS security are presented. First is based on the idea of generating process invariants using ML and the second is based on system modeling to detect and isolate attacks. Further several challenges are pointed out and a few recommendations are provided.