Java Deep Learning Cookbook

Configuring and training a DQN agent

DQN refers to an important class of reinforcement learning, called value learning. Here, we use a deep neural network to learn the optimal Q-value function. For every iteration, the network approximates Q-value and evaluates them against the Bellman equation in order to measure the agent accuracy. Q-value is supposed to be optimized while the agent makes movements in the world. So, how we configure the Q-learning process is important. In this recipe, we will configure DQN for a Malmo mission and train the agent to achieve the task.