Abstract: There are full of uncertainties in today’s autonomous systems. Some of the uncertainties are caused by potential deadline misses, namely timing disturbances, due to the complex environment. Also with the growing interest on data-driven approaches, more uncertainties arise due to applying data-driven controllers. In this talk, I would like to introduce our recent works on safety verification of autonomous systems with either timing disturbances or neural-network controllers. For the systems with timing disturbance, we consider the model of (m,K) weakly-hard systems and give the sufficient condition to ensure the system safety. For the neural-network controlled systems, we abstract feedforward neural network controllers with Bernstein polynomials and propose two methods to estimate the approximation error bound. Combining with forward reachability analysis, our approach can compute overapproximated reachable sets of neural-network controlled systems with precise error bounds.
 
Bio: Dr. Chao Huang is a postdoc working with Prof. Qi Zhu, at the Department of Electrical and Computer Engineering (ECE) in Northwestern University. Dr. Huang received a Ph.D. and a B.E. in CS from Nanjing University in 2018 and 2011 respectively. His research interests include synthesis and verification of cyber-physical systems (CPS), Internet of Things, embedded and real-time systems. His papers were accepted by the conferences including IJCAI, FM, EMSOFT, HSCC etc. He served IJCAI 19 as a PC member.