Feature Selection for Neuro-Dynamic Programming

Speaker : Sean Meyn
University of Florida
Date: 26/11/2013
Time: 2:00 pm - 3:00 pm
Location: LINCS Meeting Room 40


Neuro-Dynamic Programming encompasses techniques from both reinforcement learning and approximate dynamic programming. Feature selection refers to the choice of basis that defines the function class that is required in the application of these techniques. This talk reviews two popular approaches to neuro-dynamic programming, TD-learning and Q-learning. The main goal of this work is to demonstrate how insight from idealized models can be used as a guide for feature selection for these algorithms. Several approaches are surveyed, including fluid and diffusion models, and the application of idealized models arising from mean-field game approximations. The theory is illustrated with several examples.