Q-learning

Q-learning is a <a href="/facts/Reinforcement_learning/NrgPPS0Q">reinforcement learning</a> algorithm that trains an <a href="/facts/Intelligent_agent/8C2Q93Vw">agent</a> to assign values to its possible actions based on its current <a href="/facts/State_(computer_science)/4Yu8xXHS">state</a>, without requiring a model of the environment (<a href="/facts/Model-free_(reinforcement_learning)/sPfGYqR7">model-free</a>). It can handle problems with <a href="/facts/Stochastic_matrix/2cRWAqC3">stochastic transitions</a> and rewards without requiring adaptations.
For example, in a grid maze, an agent learns to reach an exit worth 10 points. At a junction, Q-learning might assign a higher value to moving right than left if right gets to the exit faster, improving this choice by trying both directions over time.
For any finite <a href="/facts/Markov_decision_process/cVlWIDrw">Markov decision process</a>, Q-learning finds an optimal policy in the sense of maximizing the expected value of the total reward over any and all successive steps, starting from the current state. Q-learning can identify an optimal <a href="/facts/Action_selection/PQPAvJvl">action-selection</a> policy for any given finite Markov decision process, given infinite exploration time and a partly random policy. 
"Q" refers to the function that the algorithm computes: the expected reward—that is, the quality—of an action taken in a given state.

Q-learning open-in-new

Q-learning