Returns, policies, the Bellman equation and the deep Q-network that approximates it.
8h · 10 lessons · 1 challenge
Before this: Recommender Systems
Reinforcement learning replaces the labelled example with a reward, and almost everything follows from that one change. This module builds the formalism: the return and its discount, policies, the state-action value function, and the Bellman equation that ties them together.
Then the practical half: stochastic environments, continuous state spaces, deep Q-learning, and the exploration problem that every one of them has.
Ten arms, a thousand pulls and no second chance. Beat the optimal policy, or find out how far off you were.
Worth 400 XP.
