Reinforcement Learning (RL) is the branch of machine learning concerned with the development of agents that learn to make decisions by trial-and-error through interactions with an environment. Learning is based on rewards or punishments attached to certain behaviors, and the goal is to optimize a measure of cumulative reward over time. As such, greedily optimizing reward in the short term is less important than developing a policy for actions aiming towards a long-term goal (e.g. win a game).