Wednesday, May 4, 2022 - 1:00pm to 2:00pm
Zoom: https://oregonstate.zoom.us/j/93591935144?pwd=YjZaSjBYS0NmNUtjQzBEdzhPeDZ5UT09

Speaker Information

Roni Khardon
Professor
Department of Computer Science
Indiana University, Bloomington

Abstract

Planning to control an agent in a probabilistic environment requires reasoning about actions, their outcomes and their effects on the overall utility of the agent. The talk will emphasize the role of approximate probabilistic inference as a tool for understanding and developing new planning algorithms. I will first discuss planning through a differentiable symbolic form of belief propagation, and our corresponding system SOGBOFA that yields state of the art performance in a range of problems with large combinatorial action spaces. I will then discuss a more general framework that combines inference style and approximation style to derive old and new planning algorithms and show some preliminary results on their empirical performance. Time permitting I will show how these ideas can be extended for control in continuous spaces, with applications for reinforcement learning and robotics.

Speaker Bio

Roni Khardon is a professor in the Department of Computer Science at Indiana University, Bloomington. He holds a Ph.D. in Computer Science from Harvard University, and M.Sc. and B.Sc. degrees from the Technion. Prior to moving to Bloomington he held faculty positions at the University of Edinburgh (1997-2000) and at Tufts University (2000-2018).

His research interests are in developing agents that can learn from data, build representations of their world, use such knowledge for reasoning and decision making, and act in their environment so as to optimize their objectives. His recent work spans topics in AI (probabilistic planning, knowledge representation), machine learning (graphical models, approximate inference, computational learning theory) and the connections between these areas.