Title: How Privacy Can Improve Security Against Data-Driven Attackers
Time, Date, Location: 11:00, Friday, 18 November, 2022, Brian Anderson Building Seminar Room
Abstract: In this talk, we will discuss methods that can be used as a first line of defence from data-driven attacks on Cyber-Physical Systems (CPS). First, we consider how measurement leaks about a linear, discrete, time-invariant system’s dynamics helps an adversary to calculate attacks that are difficult to detect. It turns out that these undetectable attacks can be generated with small amounts of input-output data using Willems’ Fundamental Lemma. Insight into the attack generation shows that security may be obtained through sudden changes in the system. In the second part of the talk, we will consider how injected measurement noise makes these sudden events difficult to estimate for an adversary. In particular, we use the Hammersley-Chapman-Robbins bound to provide lower bounds on the estimation error variance of when the change has occurred, which is difficult to do with traditional methods such as Cramér-Rao due to the time-step being discrete. We also present extensions of to our analysis to nonlinear systems and to simultaneous estimation of the amplitude and the time-step of the change using the Barankin bound. The latter result enables us to design open-loop control sequences that reduce information leakage. Finally, we use numerical simulations to verify our results in a smart-home setting, and to show how a forensic analysis can be made.
Biography: Rijad Alisic is a Ph.D. student with the Division of Decision and Control Systems at KTH Royal Institute of Technology in Stockholm under the supervision of Henrik Sandberg and co-supervisor Karl Henrik Johansson. He received his M.Sc degree in Engineering Physics from Lund University in 2018, with a specialization in automatic control. In 2017, he was part of the Summer Undergraduate Research Fellowship (SURF) program at California Institute of Technology in Pasadena, and in 2016, he was part of the Summer Research School with the Department of Experimental Medical Sciences at Lund University. His current research interests include privacy, security, forensics, and control of cyber-physical systems