Hands-On Reinforcement Learning with Python

Chapter 8

1. Deep Q Network (DQN) is a neural network used for approximating the Q function.
2. Experience replay is used to remove the correlations between the agent's experience.
3. When we use the same network for predicting target value and predicted value there will lot of divergence so we use separate target network.
4. Because of the max operator DQN overestimates Q value.
5. By having two separate Q functions each learning independently double DQN avoids overestimating Q values.
6. Experiences are priorities based on TD error in prioritized experience replay.
7. Dueling DQN estimating the Q value precisely by breaking the Q function computation into value function and advantage function.