chapter six

6 Stabilizing value-based deep reinforcement learning method

 

In this chapter:

  • You'll improve on the methods you learned in the previous chapter by making them more stable and therefore less prone to divergence.
  • You'll explore advanced value-based deep reinforcement learning methods, and the many components that make value-based methods better.
  • You'll implement more complex exploration strategies and flexible loss functions with function approximation.
  • You'll solve the cart-pole environment in a fewer number of samples, and with more reliable and consistent results.

"There are times I am happy. There are times I am sad. But I always try to separate emotion from the need to reach for something stronger, deeper. And then no matter the emotion, I can reach for a stability that helps me accomplish what is the goal."

Troy Polamalu , A former American football strong safety, Samoan descent

6.1   DQN: Making reinforcement learning more like supervised learning

Common problems in value-based deep reinforcement learning

It's important we are clear and understand the two most common problems that consistenly show up in value-based deep reinforcement learning.

6.2   Double DQN: Mitigating the overestimation of approximate action-value functions

6.3   Summary