Qu'est-ce que « Reinforcement learning, Markov decision process, policy, reward, cumulative reward » ?

Reinforcement learning learns how to act from the consequences of actions. A Markov decision process (MDP) is a set of states $x$, actions $u$, transition probabilities $\P(x_{t+1}\mid x_t, u_t)$ and a reward $g_{t+1}$ received after each action; the cumulative reward from time $t$ is $G_t = \sum_{k\ge0}\gamma^kg_{t+k+1}$, with a discount $\gamma\in(0,1]$ ($\gamma = 1$ on the finite horizons of this chapter). A policy $\pi(u\mid x)$ gives the probability of each action in each state.