Multi-Agent Decision Modeling

Stephen Lim Tung Min

Abstract

In recent years, multi-player decision problems are becoming a highly active research area in the form of multi-agent systems in artificial intelligence, with a wide range of applications. In multi-player decision problems, there are multiple decision makers who operate in an uncertain environment and can interact with other decision makers. The knowledge of these decision makers can be rather incomplete in regards to the actual decisions made by their opponents or even the probabilistic outcome of chance variables.

Decision theory has been widely used to analyse the set of optimal decisions that should be undertaken for a decision maker in a single-player setting. This theory has also commonly been called. Game Theory, on the other hand, has been widely used to analyse the decision outcomes in a multiple player setting, where the decision makers now not only makes decisions against nature, but also against other intelligent opponents.

This thesis therefore seeks to merge certain concepts in decision theory with game theory in order to exploit both their advantages and their advances in decision making. These concepts include the Perfect Bayesian Equilibrium, game trees and backward induction in game theory, and influence diagrams, decision trees and Multi-agent Influence Diagrams in decision theory.