A decision theoretic perspective of multiple agent problem solving: Application to a resource allocation problem

Hejamadi Raghavendra Rao, Purdue University

Abstract

This thesis explores multiple agent decision making from an information theoretic perspective. In particular, we consider resource allocation among multiple agents in an environment of partial and incomplete preference information. The study examines the process of collecting partial information, about the utilities of individual decision makers. The information is then used as a basis for taking an optimal solution. The thesis posits the existence of a central policy maker who decides about allocating resources to the members of a team, such that the team welfare is maximized, subject to resource constraints. The policy maker bases his recommendations on information about an individual agents' preference function, that is gathered during a process of polling the various agents. The polling processes are based on a nonparametric revealed preference approach. One basic assumption that is used is that all individual agents are rational, i.e. they have preferences that are total, reflexive, and transitive. Thus the study synthesizes two major paradigms for use by the policy maker: (1) The sequential, or adaptive paradigm, which is used as a foundation for the request-conduct-reply cycle of experimentation and information gathering about preferences of a single agent. (2) The global or parallel paradigm that is used by the policy maker to poll across distributed agents or a subset thereof. The inherent complexity of the problem arises, when the policy maker has to operationalize the process of preference elicitation. This complexity is the result of information being partial, contaminated, and costly. Thus the tradeoff between the cost of information gathering, and the improvement in the decision produced is taken into account. In addition, the cost of information acquisition and the problem complexity have naturally brought up the issue of integrating approximation into the decision process. Therefore a key feature of the thesis is that the policy maker integrates information acquisition with decision making to emerge with a socially optimal resource allocation.

Degree

Ph.D.

Advisors

Moore, Purdue University.

Subject Area

Management

COinS