A Weighted Markov Decision Process
The paper deals with a discrete time Markov decision process with finite state and action spaces. The authors consider a reward criterion which is a convex combination of the expected average rewards per unit time and expected total discounted rewards over an infinite time horizon. The existence of an ultimately deterministic \(\varepsilon\)-optimal policy is proved. A policy is called ultimately deterministic if it is nonrandomized Markov and it is deterministic (or, in other words, nonradomized stationary) from some deterministic epoch onward. A simple algorithm for computation of an ultimately deterministic \(\varepsilon\)- optimal policy is provided. Reviewer's remark: The paper does not refer to the reviewer's paper, ``Controlled Markov processes with arbitrary numerical criteria [Theory Probab. Appl. 27, 486-503 (1982; Zbl 0515.90076)] where a convex combination of the linear combination of the expected total and upper limit of average rewards was considered (section 4.6) as an illustration of the general methods developed in the paper. For models with Borel state and action spaces, the existence of a nonrandomized Markov \(\varepsilon\)-optimal policy for a given initial distribution was proved. This result implies the existence of an ultimately deterministic \(\varepsilon\)-optimal policy for the model considered in the reviewed paper.
- scientific article; zbMATH DE number 700091
- Minimising a threshold probability in discounted Markov decision processes
- A variance minimization problem for a Markov decision process
- Exponential Convergence in Undiscounted Continuous-Time Markov Decision Chains
- Mean-variance criteria in an undiscounted Markov decision process
- On maximizing the average time at a goal
- Constrained Undiscounted Stochastic Dynamic Programming
- Semi-infinite Markov decision processes
- Value Iteration in a Class of Communicating Markov Decision Chains with the Average Cost Criterion
- scientific article; zbMATH DE number 1786117
- Weighted discounted Markov decision processes with perturbation
- Semi-infinite Markov decision processes
- Semi-Infinite Weighted Markov Decision Processes
- Weighted reward criteria in Competitive Markov Decision Processes
- Markov Decision Models with Weighted Discounted Criteria
- Controlled Markov processes on the infinite planning horizon: Weighted and overtaking cost criteria
- Multi-objective discounted continuous-time Markov decision processes
This page was built for publication: A Weighted Markov Decision Process
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4032595)
