BEGIN:VCALENDAR
METHOD:PUBLISH
PRODID:-//Apple Computer\, Inc//iCal 1.0//EN
X-WR-CALNAME;VALUE=TEXT:USC
VERSION:2.0
BEGIN:VEVENT
DESCRIPTION:Speaker: Yasser Shoukry, Postdoctoral Scholar, UCLA/UC Berkeley /UPenn
Talk Title: Secure State Estimation For Cyber Physical Systems Under Sensor Attacks: A Satisfiability Modulo Theory Approach
Abstract: Motivated by the need to secure critical infrastructure against sensor attacks, in this talk I will focus on a problem known as "secure state estimation". It consists of estimating the state of a dynamical system when a subset of its sensors is arbitrarily corrupted by an adversary. Although of critical importance, this problem is combinatorial in nature since the subset of attacked sensors in unknown. Previous work in this area can be classified into two broad categories. The first category is based on numerical optimization techniques. These techniques are well suited to handle the continuous part of the problem, estimating the real-valued variable describing the state, if the combinatorial part of the problem has been solved. The second category is based on Boolean reasoning, which is well suited to handle the combinatorial part of the problem, if the continuous part of the problem has been solved. However, since we need to simultaneously solve the combinatorial and the continuous part of the secure state estimation problem, the existing approaches result in algorithms with worst case exponential time complexity.\n
\n
In this talk, I will present a novel and efficient algorithm for the secure state estimation problem that uses the lazy SMT approach in order to combine the power of both SAT solving as well as convex optimization. While SAT solving is used to perform the combinatorial search, convex optimization techniques are used to reason more efficiently about the real-valued state of the system and/or generating theory lemmas explaining conflicts in the combinatorial search. We show that by splitting the reasoning between the two domains (Booleans and Reals) and intermixing a powerful tool from each domain, we obtain a new suite of tools that scales more favorably compared to the previous techniques. I will start by discussing the simplest case when the underlying dynamics are linear, sensors are perfect (noiseless), and only data collected over a finite window is considered. I will then move forward by showing several extensions to handle noisy measurements, recursive implementations (data over infinite windows) and nonlinear dynamics.\n
Biography: Yasser Shoukry is a Postdoctoral Scholar at the EECS Department at UC Berkeley, the EE Department at UCLA and the ESE Department at UPenn. He received the Ph.D. in Electrical Engineering from UCLA in 2015 where he was affiliated with both the Cyber-Physical Systems Lab as well as the Networked and Embedded Systems Lab. Before joining UCLA, he spent four years as an R&D engineer in the industry of automotive embedded systems. His research interests include the design and implementation of secure- and privacy- aware cyber-physical systems by drawing on tools from embedded systems, control and optimization theory, and formal methods. \n
\n
Dr. Shoukry is the recipient of the Best Paper Award from the International Conference on Cyber-Physical Systems (ICCPS) in 2016. He is also the recipient of the UCLA EE Distinguished PhD Dissertation Award in 2016, the UCLA Chancellor's prize in 2011 and 2012, UCLA EE Graduate Division Fellowship in 2011 and 2012, and the UCLA EE Preliminary Exam Fellowship in 2012. In 2015, Dr. Shoukry led the UCLA/Caltech/CMU team to win the first place in the NSF Early Career Investigators (NSF-ECI) research challenge. His team represented the NSF-ECI in the NIST Global Cities Technology Challenge, an initiative designed to advance the deployment of Internet of Things (IoT) technologies within a smart city.\n
Host: Paul Bogdan
SEQUENCE:5
DTSTART:20160912T140000
LOCATION: 248
DTSTAMP:20160912T140000
SUMMARY:EE 598 Cyber-Physical Systems Seminar Series
UID:EC9439B1-FF65-11D6-9973-003065F99D04
DTEND:20160912T150000
END:VEVENT
END:VCALENDAR