Risk Averse Dynamic Programming for MDP's with Unbounded Cost and Infinite Horizon
Department of Applied Mathematics
Date / Time: 17.03.2017 / 13.00
Abstract. We use Markov risk measures to formulate a risk averse version of the total cost problem for a controlled Markov process. We derive conditions for a system to have finite risk over an infinite time horizon, and derive risk-averse dynamic programming equations satisfied by the optimal policy. We further ropose an epsilon-optimal approximation to the optimal policy.