The Geometry of the Space of Value Functions in POMDPs via Interval Matrix Enclosures
Date:
In this talk I described using interval matrix enclosures to characterize the geometry of the space of value functions in reinforcement learning problems, especially POMDPs.