Risk Sensitive Multiple Goal Stochastic Optimization, with application to Risk Sensitive Partially Observed Markov Decision Processes

Vaios Laschos, Robert Seidel, Klaus Obermayer

Published 2018-08-13Version 1

We study Risk Sensitive Partially Observable Markov Decision Processes (RSPOMDPs) where the performance index is either the expected utility corresponding to a function that can be written as a weighed sum of exponentials or the expected shortfall generated by functions of the hyperbolic sine type. In that direction we utilize methods for treating Risk Sensitive Multiple Goal Markov Decision Processes (RSMGMDPs) where the performance index is given by sums of expected utilities, where each utility is applied to a different running cost, but also methods from the theory of Risk Sensitive Constrained Markov Decision Processes (RSCMDPs).