Supervisor

Pr. Brigitte Jaumard
Concordia Institute Information Engineering - OCN
Concordia University, Montréal, Canada
Web Page: http://users.encs.concordia.ca/~bjaumard/
Mail: bjaumard.at.ciise.concordia.ca
Phone: + 1 (514) 848-2424 #5380
Fax: + 1 (514) 848-3171

Abstract

Barter systems are exchange systems among agents which are quite attractive for the management of peer-to-peer systems. Each agent, whether he is human or artificial, owns his preferences over the whole set of resources. It enforces him to initiate transactions with other agents in order to increase his own welfare. The classical assumption under which a multi-agent system is populated by selfish agents is more plausible than any hypothesis on a general cooperation among agents. Automated Mechanism Design (AMD) seeks to find, using algorithms, the optimal/best possible rules of interaction among selfish and rational agents in order to get the best resource allocation. AMD can be used to efficiently model a multi-agent resource allocation problem as an optimization problem. A difficulty lies in the size of the optimization problem: there is a huge number of variables and constraints, even for AMD instances of relatively small size. We study how to adapt the column generation techniques in order to solve the linear programming formulation of the automated mechanism design problem. Several problems arising in peer-to-peer networks can be easily modeled using a multi-agent system or a resource allocation problem such as the video streaming applications or the grid computing. We plan to study resource allocation problem using automated mechanism design which is associated with video streaming deliveries.