Reinforcement Learning and Dynamic Programming

Sequential decision-making under uncertainty is the unifying thread here: how an agent should act when outcomes are stochastic and the world's dynamics may be unknown, formalized through Markov decision processes and the Bellman equations. You'll move from exact dynamic programming (value and policy iteration) into model-free territory — Monte Carlo, TD, Q-learning, policy gradients, actor-critic — with two quizzes, a midterm, a final, and a substantial implementation project where you actually code these algorithms. It's a graduate-level course that leans on probability and optimization, and it's the natural gateway into control, robotics, and modern ML research where RL underpins everything from recommender systems to robot learning.