Deep Q-networks were first introduced by DeepMind in their paper Human-level control through deep reinforcement learning, published in the British scientific journal Nature, and now commonly referred to as the Nature Paper. The goal of Deep Q-learning was to create an AI agent that could learn a policy from high-dimensional inputs such as video games. In our case, we'll want to construct a Deep Q-network that can advance through basic tasks and then towards harder tasks.
Deep Q-networks approximate Q-values instead of calculating them individually with Q tables, and they do this by using artificial neural networks as a value approximator. The input of the network will be a stack of preprocessed frames, and the ...
