TensorFlow 1.x Deep Learning Cookbook

Game of Atari using Deep Q Networks

A deep Q network (DQN) is a combination of Q learning with convolutional neural networks (CNNs), first proposed by Mnih and others in 2013 (https://arxiv.org/pdf/1312.5602.pdf). The CNN network, due to its ability to extract spatial information, is able to learn successful control policies from raw pixel data. We have already played with CNNs in Chapter 4, Convolutional Neural Networks, and so we start directly with the recipe here.

This recipe is based on the original DQN paper, Playing Atari with Deep Reinforcement Learning by DeepMind. In the paper, they used a concept called experience replay, which involved randomly sampling the previous game moves (state, action reward, next state).