Overview: Reinforcement Learning

Every model in this unit so far has learned from a fixed dataset of labelled or unlabelled examples. Reinforcement learning (RL) breaks that mould: an agent interacts with an environment over time, choosing actions, observing consequences, and receiving a scalar reward signal — with no supervisor ever telling it "here is the correct action." This is the unit's frontier closer, kept deliberately compact per the profile's length-flex note ("optional for a quant-focused cut"): the goal here is fluency with the Markov decision process (MDP) formalism and the two dominant algorithmic families — value-based methods (value iteration, policy iteration, Q-learning) and policy-gradient methods (REINFORCE) — plus a pointer to how deep learning scales both to high-dimensional problems.