On the Interaction of Information and Decisions in Dynamic Networked Systems.

Abstract

Efficient operation of modern dynamic networked systems, such as communication systems, queueing networks, power systems, and surveillance systems, can significantly improve our quality of life. The operation of a dynamic networked system involves series of decision making processes by many decision makers (DMs) who may or may not have the same information, and may or may not share the same objective.

The quality of each DM's decision depends on the quality of the information available for decision-making in the network. Since the network is dynamic, the information available to the DMs over time is a dynamic process that depends on the DMs' decision rules. Information affects decisions, and decisions influence information. This interaction between information and decisions in dynamic networks results in complex decision-making problems.

In this thesis, we study the impact of the information-decision interaction on system performance within the context of: (i) centralized stochastic control; (ii) decentralized stochastic control; and (iii) game theory. Specifically, within the context of centralized stochastic control, we study a multi-state channel sensing problem, and discover sets of conditions sufficient to guarantee the optimality of a myopic policy. Within the context of decentralized stochastic control, we consider a decentralized routing problem as well as a multiple access communication problem; we discover an optimal decentralized routing policy for the routing problem, and an efficient decentralized multiple access protocol. Within the context of game theory, we study a general model of dynamic stochastic games with asymmetric information; we introduce the concept of common information based perfect Bayesian equilibrium (CIB-PBE), and provide a sequential decomposition for the dynamic games that leads to an algorithm to determine CIB-PBE.