Reinforcement learning trains an agent to choose actions by learning from rewards and consequences.
Reinforcement learning is a machine-learning method where an agent learns by interacting with an environment.
the decision-making system
the world it interacts with
what the agent chooses to do
feedback about the result
Goal: learn a policy that produces better long-term reward.
Learning a game by playing it
You try a move.
The game gives you a result.
Good decisions may lead to rewards.
Bad decisions may lead to poor outcomes.
Over many attempts, you learn which actions tend to work better.
State: Current board
Action: Choose move
Reward: Win / lose / score
State: Robot position
Action: Move
Reward: Progress toward goal
State: Current context
Action: Choose an option
Reward: Measured outcome
Reinforcement learning can be used in some recommendation and decision systems; not all recommendation systems are reinforcement learning.
Supervised: Correct answer supplied for examples.
RL: Feedback comes from interaction and rewards.
Unsupervised: Finds structure in unlabeled data.
RL: Learns actions through rewards.
Reward: A feedback signal.
Correct answer: A specific target label.
Explore next
Reinforcement learning teaches an agent which actions work better through interaction and reward.
A video explanation will be added when a real source is available.