Loading…
Glossary term
A reinforcement learning method that learns the value of taking each action in each state.
Q-learning learns a table, or a neural network, Q(state, action) estimating how good each action is in each state. After each step it nudges that estimate toward the reward received plus the best value expected next.
Acting greedily on the learned Q values gives a good policy.
The Q-learning update
alpha, gamma = 0.1, 0.9
Q[s][a] += alpha * (reward + gamma * max(Q[s_next]) - Q[s][a])It is a foundational reinforcement learning algorithm, and its deep version learned to play Atari games.