Reinforcement learning
Also: RL, deep reinforcement learning, deep RL
An agent observes a state, takes an action, and receives a reward and a new state from the environment; its goal is a policy that maximizes the expected sum of discounted future rewards. The problem is usually formalized as a Markov decision process.
G_t = r_{t+1} + γ r_{t+2} + γ² r_{t+3} + …
Example
In a gridworld the agent gets +1 for reaching the goal cell and −1 for falling into a pit; after many episodes it learns which way to step from every cell.