Reinforcement Learning

The course will introduce reinforcement learning as an approximate dynamic programming problem. We will consider exact versions of value and policy iteration, followed by approximations based on gradient methods, temporal difference based methods, and last but not least, simulation based methods like Q-learning.

1. Students will be able to implement and analyze exact dynamic programming (value and policy iteration) algorithms
2. Students will be able to describe and implement approximations in value and policy space
3. Students will be able to describe and implement algorithms for model-free situations
4. Students will be able to execute projects using commonly available RL libraries