16.5 Deep Reinforcement Learning

When the state space is huge — like raw pixels from a game screen — you cannot store a Q-value for every state in a table. Deep Reinforcement Learning replaces the table with a neural network that approximates the Q-values or the policy directly. The Deep Q-Network (DQN) that learned to play Atari games from pixels is the landmark example.

Picture teaching a system to control traffic lights across all of Bole using live camera feeds. The number of possible road situations is astronomical, so a neural network learns to map camera images to good signaling decisions instead of a lookup table.