Groupe d’études et de recherche en analyse des décisions


Efficient Aggregation of a Class of Large Scale Markov Desicion Processes Through the Use of a Census Process


The problem of controlling a population process, where each individual is a copy of the same fundamental Markov-core process, can be aggregated by considering a census process. As often in aggregation procedures, the advantage of state-space reduction can be lessened by the computational time necessary for the characterization of the reduced Markov chain. This paper is devoted to the presentation of efficient algorithms for generating census states and computing optimal strategies on the aggregate process.

, 21 pages