Reinforcement Learning: An Introduction and Key Results

Explore reinforcement learning in this 51-minute talk by Erwann LePennec at the Institut des Hautes Etudes Scientifiques (IHES). Delve into the art of learning to act in an environment observed only through interactions. Begin with an introduction to the underlying probabilistic model, Markov Decision Process, and learn how to develop effective policies for both known and unknown models. Examine the impact of solution parametrization and the importance of understanding stochastic approximation. Discover practical applications of reinforcement learning, including faster issue detection in ultrasound exams, improved MDP solving through better approximation, and enhancing RL robustness while controlling sample complexity. The talk covers key topics such as sequences of decisions, dynamic programming, reinforcements, approximation techniques, practical implementation, and robustness considerations.