Reinforcement Learning · intermediate · concept 122 of 176

🎮Q-Learning & Deep Q-Networks (DQN)

Q-Learning estimates the value of actions in each state. Deep Q-Networks (DQN) use neural networks to approximate Q-values, DeepMind's DQN played Atari games at superhuman level (2013). SARSA is Q-learning's on-policy sibling: it learns from the action actually taken rather than the greedy one.