Liam Paull - Universite de Montreal and Montreal Institute for Learning Algorithms
Dec. 1, 2017, 2:30 p.m. - Dec. 1, 2017, 3:30 p.m.
Hosted by: Joelle Pineau
Robots that operate in the physical world have finite resources and hard realtime constraints. For autonomous systems to be able to perform useful tasks, their algorithms must scale well with time, space, and complexity. In this talk I will present a framework for resource-constrained graphical inference. The approach optimally utilizes available computation, memory, and bandwidth to provide the best possible estimates of the state of the system given the limited resources. The flexibility of the framework is demonstrated in two diverse application domains: underwater cooperative localization and mapping with communication constraints, and collision-free navigation with computation and memory constraints. In both cases, we exploit task-specific structure to optimize task performance without exceeding the finite available resources. I will also then take some time to describe some new projects and directions that I’m interested to explore in the newly formed robotics group at Université de Montréal.
Liam Paull is an assistant professor at Université de Montréal in the Université de Montréal Département d'informatique et de recherche opérationnelle (DIRO), and an associate member of the Montreal Institute for Learning Algorithms (MILA). He received the B.Sc. degree in computer engineering from McGill University, Montreal, QC, Canada, in 2004 and the Ph.D. degree in electrical and computer engineering from University of New Brunswick, Fredericton, NB, Canada, in 2013, where he studied marine robotics. He was previously a Research Scientist and postdoctoral associate in the Computer Science and Artificial Intelligence Laboratory (CSAIL) at Massachusetts Institute of Technology where he led the Toyota Research Institute Funded “Parallel Autonomy” self-driving car project. His research interests include inference on resource-constrained and safety-critical robotic systems as well as bridging deep learning methods to develop novel representations for increased mobile robot autonomy.