English

Controlled Markov Chains with AVaR Criteria for Unbounded Costs

Probability 2015-11-18 v4 Optimization and Control

Abstract

In this paper, we consider the control problem with the Average-Value-at-Risk (AVaR) criteria of the possibly unbounded L1L^{1}-costs in infinite horizon on a Markov Decision Process (MDP). With a suitable state aggregation and by choosing a priori a global variable ss heuristically, we show that there exist optimal policies for the infinite horizon problem. To our knowledge, this is the first work of deriving dynamic programming equations with L1L^1-unbounded costs via AVaR-operator.

Keywords

Cite

@article{arxiv.1501.02518,
  title  = {Controlled Markov Chains with AVaR Criteria for Unbounded Costs},
  author = {Kerem Ugurlu},
  journal= {arXiv preprint arXiv:1501.02518},
  year   = {2015}
}