MDP models can be solved in many ways. One of the methods to solve MDP is using Monte Carlo prediction, which helps in predicting value functions and control methods for further optimization of those value functions. This method is only for time-bound tasks or episodic tasks. The problem with this method is that if the environment is large or the episodes are long, the time taken to optimize the value functions takes a while as well. However, we won't be discussing Monte Carlo methods in this section. Instead, we shall look at a more interesting model-free learning technique that is actually a combination of Monte Carlo methods and dynamic programming. This technique is called temporal difference learning. This learning can be applied to non-episodic tasks as well and doesn't need any model information to be known in advance. Let's discuss this technique in detail with the help of an example.