Reinforcement Learning and Dynamic Programming

Sequential decision-making under uncertainty is the central question here: how should an agent act when outcomes are stochastic and the world's dynamics are only partially known? You'll move from the clean Markov decision process world, where Bellman equations and dynamic programming give you exact answers, into the messier model-free regime of Q-learning, temporal difference, and policy gradients, with multi-armed bandits framing the exploration-exploitation tradeoff. Two quizzes, a midterm, a final, and an implementation project anchor the work; the math leans on probability and linear algebra, and what you build here is the foundation for control, robotics, and modern ML research.