Reinforcement Learning: Non-linear Approximation, GPUs, & Policy Gradient
Details
Last meeting we reviewed approximation techniques that form parameterized value functions and used the concept of generalized policy iteration to find an optimal solution. We focused on linear methods for which the gradient computation is particularly simple. This time we'll extend these techniques to non-linear approximation and see when GPU acceleration is helpful. Then we will apply the set of approximation options to policy gradient techniques to see what advantages they may have on the mountain car problem.
As usual you can find below links to the textbook, previous chapter notes, slides, and recordings of some of the previous meetings.
Useful Links:
Reinforcement Learning: An Introduction by Richard S. Sutton and Andrew G. Barto
Recordings of Previous Meetings
Short RL Tutorials
My exercise solutions and chapter notes
Kickoff Slides which contain other links
Video lectures from a similar course