Speaker: Dr. Dan Lizotte, Assistant Professor in the David R. Cheriton School of Computer Science, University of Waterloo
Title: Possible Futures: Complexity in Sequential Decision-Making
Date: Tuesday, January 22nd 2013
Category: Seminar

Abstract:

Effective sequential decision-making hinges on our ability to consider many possible futures. Identifying which futures to consider requires knowledge of how the environment changes in response to actions taken by the decision-maker, and knowledge of how the decision-maker will react to the changing environment. Often, one makes simplifying assumptions about the objective of the decision-maker in order to ensure that computing optimal actions remains tractable. (For example, the decision-maker is often assumed to be rational.) Unfortunately, such assumptions are often unreasonable in practice. We will discuss how assuming a more realistic but more complex decision-maker increases the computational complexity of finding good actions, and we will discuss techniques for controlling this computational complexity. Our motivating application is clinical decision support, where many competing objectives make the assumption of rationality unreasonable.

Speaker Biography:

Daniel Lizotte is an Assistant Professor in the David R. Cheriton School of Computer Science at that University of Waterloo. He is interested in the areas of machine learning, reinforcement learning, and statistics, particularly as they apply to sequential decision-making problems in health informatics. Dr. Lizotte aims to adapt and improve current techniques so they can be applied to new sources of sequential medical data, with the end goal of providing doctors with the best available evidence for non-myopic decision making.

Tagged with:
 
  • Facebook
  • Twitter