We show a close relationship between the Expectation - Maximization (EM) algorithm and direct optimization algorithms such as gradient-based methods for parameter learning.

Abstract. Solving finite-horizon Markov Decision Processes with stationary policies is a computationally difficult problem. Our dynamic dual decomposition approach uses Lagrange duality to decouple this hard problem into a sequence of tractable sub-problems. The resulting procedure is a straightforward modification of standard non-stationary Markov Decision Process solvers and gives an upper-bound on the total expected reward. The empirical performance of the method suggests that not only is it a rapidly convergent algorithm, but that it also performs favourably compared to standard planning algorithms such as policy gradients and lower-bound procedures such as Expectation Maximisation.