From RL to Deep RL

Summary

A key difference between RL and Deep RL is the use of a deep neural network. Think of the collection of value-action pairs that define what actions an agent should take in any situation as a function of the observations that the agent receives from its environment.

A neural network can be used to approximate this function because through its large quantity of parameters that can be “learned” through trial and error.

Additional References

Dettmers, Tim. "Deep Learning in a Nutshell: Reinforcement Learning." NVIDIA Developer Blog. 8 Sept. 2016. Web. 2 Feb. 2018. https://devblogs.nvidia.com/deep-learning-nutshell-reinforcement-learning/

Pixels to Actions

SOLUTION:

the deep neural network’s ability to take raw sensor data and directly choose an action directly for an end-to-end solution.