Q-learning
lastmod 2024-07-15created 2024-07-153 edits
Q-learning is a model-free reinforcement learning technique. Q-learning can be used to find an optimal policy for a given finite Markov decision process. Q-learning learns the optimal policy by learning the Q function, a function that predicts the expected utility of performing a given action in a given state. A policy is a rule that indicates which action to perform in a given state. Once the Q function has been learned, the optimal policy can be derived by performing, in each state, the action that gives the highest Q. One advantage of Q-learning is that it can compare the expected values of the actions performed without a model of the given environment. In addition, Q-learning can be applied without special modification to environments where transitions occur stochastically or rewards are given stochastically. It has been proven that, for any finite MDP, Q-learning can learn an optimal policy that obtains the maximum reward from the current state.