Our society and its critical infrastructures increasingly rely on complex networked control systems consisting of many inexpensive and yet powerful devices able to communicate, sense, and learn to interact with their environment. These devices are often connected through an omnipresent and global communication and computing infrastructure, allowing them to gather, process and learn from real-time data. Machine learning (ML) is a key enabler for these future networked systems and is a technology that over the past few years has blossomed from a promising but immature set of tools into techniques able to achieve human-level performance in a wide area of tasks. In a near future, ML is expected to be incorporated in a great variety of complex networked systems, including smart grids, intelligent transport systems, industrial processes and smart cities. There, learning algorithms, integrated to classical feedback and supervisory control loops, will help interpreting measurements from sensors, identifying and refining models of the physical processes, and in turn making wiser control decisions. ML-aided networked control systems have an enormous potential. Yet, they also offer unprecedented opportunities for attackers, and may become a reality only if we can address their vulnerabilities.
At this background, the objective of the project is to develop novel mathematical and computational tools for the fundamental understanding and engineering design of cyber-secure ML-aided networked control systems. We will focus on a new approach aiming at securing these systems globally, rather than addressing the security of each of their components individually, as our goal is to preserve the nominal overall behavior of the systems under attack. Our work will hence contribute to the development of tools towards unified defense strategies. Results from our theoretical and methodological activities will be fed directly into a unique demonstration environment, namely, the KTH Live-In Lab, which is a new innovation platform for research and development within smart building technologies located at the KTH main campus. Experiments and demonstration activities will be carried out at the 300 sqm innovation area that is exempt from building permit regulations, providing unique possibilities for testing and evaluations under realistic conditions. Via direct collaboration with ABB, the City of Stockholm, and Akademiska Hus, and through our own world-leading experience at the research front of networked control systems, machine learning, communication systems, and infrastructure systems, our team is one of very few world-wide capable of predicting the opportunities and challenges offered by future generation cyber-secure industrial control systems.
For this project, we have formed a team of five complementary research groups. The research leaders are centrally positioned in the international control, security, learning, and networking communities, manifested for instance by highly cited papers, regularly invited plenary presentations, scientific awards, strong international networks, and experiences of running large multi-disciplinary research projects. The project will develop new and unique competences for Sweden, of crucial importance for the long-term competitiveness of smart city, transport, communication, and vehicular industry. The key novel idea of the proposed research is the introduction of unified approach to secure ML-aided networked control systems. We claim that nobody has yet proposed or undertaken such a bold approach.
[1] J. Hespanha et al., A survey of recent results in networked control systems. Proc. of IEEE, vol 95, 2007.
[2] M. Barreno et al., The security of machine learning. Machine Learning 2010.