Reinforcement Learning

Machine Learning (ML) methods, in the case of supervised ML, typically rely on a supervisor providing a knowledge base which is used by the ML algorithm to learn new patterns and make predictions. In the case of unsupervised learning, reinforcement learning differs in that in the former there is no input-output mapping whereas in the latter the input-output mapping still exists. In reinforcement learning there are concepts such as actions, rewards, agents and environments. For each action an agent takes in an environment it gets a reward in the form of a numerical value. Thus the objective of the agent is to maximise this reward as much as possible. Consequently, the agent can learn any behaviour using this logic.