Human-level control through deep reinforcement…

DQN is remembered for learning Atari games from pixels. Read it for the less obvious result: why naive deep Q-learning chases its own moving data and targets, and how shuffled replay plus a delayed target network made that feedback loop trainable across 49 games.

Reading focus: Why a neural value function changes its own data distribution and bootstrap labels as it learns. How experience replay and a delayed target network stabilize different parts of the same feedback loop. How to read the five-game component ablation and the 49-game benchmark without turning either into a general superhuman-agent claim.

Nature 2015. Volodymyr Mnih et al.. 32 min read, medium difficulty.