Fit Value functions with a function approximator (often a parametric one), as in Supervised learning. This reduces the memory and time requirements to learn them, and solve the Reinforcement learning problem.
video
Experience replay
DQN