Feature Selection for Value Function Approximation Using Bayesian Model Selection

Feature selection in reinforcement learning (RL), i.e. choosing basis functions such that useful approximations of the unkown value function can be obtained, is one of the main challenges in scaling R