CSCI-B659 (Topics in Artificial Intelligence): Reinforcement Learning Longer title: Learning Planning and Acting in Complex Environments The course will explore agents that learn, plan and act in non-deterministic and complex environments. We will cover the main relevant results and approaches from Reinforcement Learning (RL), planning in Markov Decision Processes (MDP) and planning in Partially Observable MDPs (POMDP) including: tabular methods, methods using structured representations, methods using deep learning and method connecting planning/RL to probabilistic inference. We will discuss some theoretical results including approximation and convergence guarantees but will emphasize practical algorithms and their empirical performance. The course will start with a lecture based format to cover the foundations in these areas and then transition into a seminar format with student led presentation of recent papers from the literature. Prerequisites: At least one of B555, B556, B565, B551.