Value-function (VF) approximation is a central problem in Reinforcement
Learning (RL). Classical non-parametric VF estimation suffers from the curse of
dimensionality. As a result, parsimonious parametric models have been adopted
to approximate VFs in high-dimensional spaces, with most