Learning Models for Decentralized Decision Making

This paper presents a learning approach to the modeling of decentralized decision making problems. Minimal information is assumed to be available to the decision makers, motivating the use of simple known learning schemes for updating decisions. Models in which all decision makers act and update decisions synchronously are shown to lead to repeated strategic games. Sequential models, in which only one decision maker acts at a time, are also introduced and their relevance to certain decentralized control problems is indicated.