Playing Atari Games with DQNs

Training bots to play Atari games at super-human levels with Deep-Q-Networks

DQNs play Atari games at super-human levels.

Playing Atari with DQNs

Using Deep-Q-Networks (DQNs) I created bots that can play almost every Atari game at super human levels.

DQNs use reinforcement learning, and use neural networks as a function approximator. An experience replay memory is used to prevent the network from diverging, and deals with non-stationary data distributions and highly correlated data.

In just 3 hours of training on Google Colab my DQN achevied super-human performance in pong.

Check out the full code on Github, or the interactive notebook on Google Colab.