Knowledge MaxxingReinforcement learning
Reinforcement learning trains a system to choose actions that maximize a cumulative reward signal, learned through trial and error rather than from labeled examples.
Prerequisite: Supervised learning
What matters
Because feedback comes as a reward for a sequence of actions rather than a single correct label per example, reinforcement learning suits problems where the right action depends on long-term consequences.
Example
A reinforcement learning agent playing a game learns which sequences of moves lead to a higher score, through repeated attempts rather than being shown correct moves.
Common misconception
Reinforcement learning does not require a human to specify the correct action at every step; it only requires a reward signal reflecting how well things went.
Recall question
What kind of feedback does reinforcement learning use instead of per-example labels?