Speaker: Dr Yoni Nazarathy, School of Mathematics and Physics, University of Queensland
Title: Trying to make the right decision when you can’t see everything
When: 4:45 for 5pm to 6pm, Tuesday September 8, 2015. All welcome, including non-members.
Where: Room 314/6, O Block, Queensland University of Technology, Gardens Point
Abstract: In many automatic or human controlled systems, decisions need to be constantly made. For example, in different contexts: medical staff needs to be assigned to different locations, funds need to be distributed, computing power needs to be smartly utilized, bandwidth needs to allocated or fuel of a vessel needs to be efficiently used. Many aspects of control theory deal with these types of problems, often modelling the system as a deterministic or stochastic Markovian evolution over some state space. In these cases, once model parameters are roughly known, finding sensible or even optimal control policies is a key issue. Nevertheless, in many situations the system state is not fully observed, or alternatively is observed with some measurement noise. In this case, sensible (or optimal) controllers need to take such state uncertainty into account. This talk surveys some of the methods used in this type of situation. I’ll start by describing the classic Luenberger observer and quickly move onto Kalman filtering and describe how these methods integrate well with Linear Quadratic Optimal Control in systems with linear dynamics. I’ll then move onto Markovian Restless Bandit Problems combined with partially observable Markov decision processes. In such cases, problems often become quickly intractable, but certain relaxations allow to solve problems in an approximately optimal manner. Key concepts here are the belief state, Bellman’s equation and indexes such as the Gittins or more generally the Whittle Index. Most of the talk is aimed as a tutorial about key methods associated with state uncertainty and optimal control. A small portion of the talk will present published original work on examples in this field highlighting future directions as well.
Bio: Dr. Yoni Nazarathy got his PhD from the University of Haifa in Israel. Prior to his PhD studies he was working as a software developer, group leader and system engineer in the field of wireless network algorithms. His research lies in the field of stochastic operations research, control, statistics, scheduling and queueing theory. After his PhD studies he spent time as a research fellow in the Netherlands, then worked as a Lecturer at Swinburne University of Technology in Melbourne where he also carried out statistical consulting for industry. Now he is a Senior Lecturer at the University of Queensland where his research is funded by two ARC grants during the years 2013-2015. At the University of Queensland he serves as the Mathematics Honours Studies coordinator and also as an Industry Liaison. Within the wider scientific community he serves as an associate editor of the journal “Stochastic Models” and regularly co-organizes national and international scientific events. Within the local broad community, he participates in the Mathematicians in School program, often trying to challenge ten years olds with concepts such as the square root of minus one.
|Time:||4:45 pm - 6:00 pm|