الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 16 |  |  |
| | | from | 16 | | |
Abstract
Multi-Agent Systems is the branch of Artificial Intelligence that views intelligence and the emergence of intelligent behavior as the result of a complex structural arrangement of multiple independent autonomous interacting agents; in other words it is a social theory of intelligence. From a practical perspective, multi-agent systems aims at creating a system that inlcrconnccts separately developed agents, thus enabling the ensemble to function beyond the capabilities of any singular agent in the set-up. Multi-agent systems have gained much interest in recent years because of the increasing complexity, openness, and distributivity of current and future systems. Perhaps the best known example of such systems is the World Wide Web (WWW).
One important class of problems in multi-agent systems is that of planning or sequential decision making. Planning is the process of constructing an optimal policy with the objective of reaching some terminal goal state. The key aspect of multi-agent planning is coordinating the actions of the individual agents. There are three major approaches for coordination: communication, pre-imposed conventions, and learning.
This thesis addresses multi-agent planning through the learning approach. To facilitate this study a neu taxonomy of multi-agent systems is proposed that divides (hem into hio main classes, cooperative multi-agent systems and competitive multi-agent systems. The thesis focuses on planning in the cooperative setting with some extensions to the competitive selling. A new reinforcement learning-based algorithm is proposed for learning a joint optimal plan in cooperative multi-agent systems. This algorithm is based on the decomposition of the global planning problem into multiple local matrix games planning problems. The algorithm assumes that all agents are rational. Experimental studies on some grid games show the convergence of this algorithm to an optimal joint plan.