Machine Learning for Modelling and Decision Making in Complex Physical Domains - Mark Crowley, Assistant Professor, Waterloo
AI & Fundamentals
DATE: Mon, November 4, 2019 - 3:30 pm
LOCATION: MacLeod (MCLD) - 242, 2356 Main Mall, Vancouver, BC
My lab at the University of Waterloo carries out work on a variety of topics within Artificial Intelligence and Machine Learning with a focus on using real world problems to discover computational challenges for modelling of uncertainty, dealing with large or streaming datasets, learning predictive models and enabling decision making. In this talk I will provide a brief overview of a few ongoing projects where these challenges arise from different sources. In combustion modelling for energy production and engine design, standard practices are to simplify complex turbulence and other physical dynamics down to simple lookup tables. We have introduced a new approach for using supervised learning to create more a powerful and compact way of doing this. In physical chemistry, automated material design presents a computational challenge arising from the combinatorial size inherent in a multi-step process of continuous actions each relying on a detailed physics model. I will describe a new project we have looking at using Deep Reinforcement Learning to automate parts of this problem. In multi-agent domains where there are many human decision makers and where decision support systems or full automation are desirable, computational challenges can arise from communication needs as many individuals interact as well as their needs for communication or their relation to each other as they seek to optimize their goals. We have been developing several approaches for addressing this as a Multi-Agent Reinforcement Learning (MARL) problem in domains such as autonomous driving and management of forest fires.
Mark Crowley is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Waterloo. He received his PhD from the University of British Columbia in 2011 and did a postdoc at Oregon State University researching problems in the field of Computational Sustainability. He is a member of the Waterloo Artificial Intelligence Institute and a Faculty Research Fellow at Element AI in Montreal. His research seeks dependable and transparent ways to augment human decision making in complex domains in the presence of spatial structure, large scale streaming data, and uncertainty. His focus is on developing new algorithms within the fields of Reinforcement Learning, Deep Learning, Manifold Learning and Ensemble Methods. Dr. Crowley often works in collaboration with researchers and policy makers in diverse fields such as sustainable forest management, physics and chemistry, autonomous driving and medical imaging.
Professor David Poole, Computer Science Department, UBC