What happens when you combine Q-learning with deep neural networks - and why did it let AI master Atari?

The answer

Q-learning, developed by Chris Watkins in 1989, is an algorithm that learns the value of taking each action in each state without needing a model of the environment. In 2013, DeepMind combined Q-learning with deep neural networks to create the Deep Q-Network (DQN), which learned to play 49 different Atari 2600 games from raw pixel input alone - achieving superhuman performance in 29 of them. The breakthrough paper, published in Nature in February 2015, introduced experience replay and target networks to stabilize training. DQN demonstrated that a single general-purpose agent could master diverse tasks, reigniting interest in reinforcement learning.