For large state or action spaces, reinforcement learning algorithms can be designed to use generalizing learners such as neural networks) to represent the value function, the Q-function and/or the policy.