Speaker: Tim Ireland, Quantitative Researcher at Rotella Capital Management (Mon. 04/15)
- 3:00 PM to 4:00 PM
- COLH 444
How to make good decisions: a statistical introduction to reinforcement learning
Abstract: Making good decisions is difficult. Teaching machines to make good decisions, especially in situations where people struggle to do so, is extremely difficult. The standard approaches to machine learning that one might use to classify images or generate synthetic text generalize rather poorly to decision-making problems, where there is often inherent uncertainty in the environment and the goal is difficult to evaluate or sparsely distributed. Reinforcement learning gives us a different way to look at problems, and suggests a new class of machine learning algorithms for approaching these challenges. In this talk we'll examine the fundamental ideas behind reinforcement learning, starting with foundations in probability and statistics. We'll then use those ideas to discuss state-of-the-art results in artificial intelligence.