Date and Time: Friday, August 28, 2020 at 1300 AEST
Speaker: Tim Molloy, University of Melbourne
Title: No Time to Learn: Robust Quickest Simultaneous Detection and Classification of Unknown Changes
Abstract: The problem of quickly detecting and simultaneously classifying (or isolating) unknown changes in stochastic processes arises naturally in many applications of control engineering and signal processing including fault detection and isolation, manoeuvring target tracking, and cognitive radio. Almost all theoretical results for quickest detection and isolation have however been obtained under the assumption that the pre- and post-change distributions (or statistical models of the observed process) are perfectly known – only in 2000 did Lai obtain some non-trivial theoretical results for computationally prohibitive generalized likelihood ratio algorithms. This talk will present recent novel theoretical results characterizing the performance of efficient recursive algorithms for quickly detecting and isolating unknown changes. By exploiting these characterizations along with information-theoretic relative entropy inequalities, it will be shown that efficient recursive algorithms can constitute minimax robust quickest change detection and isolation solutions with practical performance that rivals that of computationally prohibitive generalized likelihood ratio algorithms.
Bio: Tim received the B.E. and PhD degrees in 2010 and 2015 from the Queensland University of Technology (QUT), Australia. From 2016 to 2019 he was a Research Associate and then Advance Queensland Research Fellow at QUT. Since 2020 he has been a Research Fellow at the University of Melbourne on the AUSMURI project “Neuro-Autonomy: Neuroscience-Inspired Perception, Navigation, and Spatial Awareness for Autonomous Robots”. His research interests include inverse optimal control, dynamic game theory, and information-theoretic perception and inference for robots and autonomous systems. He was awarded a QUT University Medal (2010), a QUT Outstanding Doctoral Thesis Award (2016), and an Advance Queensland Early Career Research Fellowship (2017-2019) from the Queensland Government Department of Science, Information Technology and Innovation (DSITI) supported by Boeing Research & Technology Australia.
STAEOnline 