Self-adapting agent organisations

Autonomic systems, capable of self-management, are being advocated as a solution to the problem of maintaining modern, large, complex computing systems. Given this, we believe self-organising multi-agent systems provide a convenient paradigm to develop these autonomic systems because such self-organising systems can arrange and re-arrange their structure autonomously, without any external control, in order to adapt to changing requirements and environmental conditions. Furthermore, such systems need to be decentralised, so that they are robust against failures; again, this characteristic fits with the multi-agent paradigm. With this motivation, this thesis explores the area of self-organisation in agent systems, and particularly focuses on the decentralised structural adaptation of agent organisations.<br/><br/>In more detail, self-organisation has been generated in agent systems using various approaches like stigmergy, reinforcement mechanisms, cooperative actions of agents and reward based mechanisms for selfish agents. However, none of these are directly applicable to agent organisations because they cannot be incorporated into deliberative agents working towards organisational goals. The few adaptation mechanisms that are applicable are either centralised or are based on restricted settings and also ignore the resources being used by the adaptation process. Here, we particularly focus on such problem solving agent organisations because they provide a suitable representation for autonomic systems. We investigate and develop mechanisms to incorporate decentralised structural adaptation in organisations to improve their performance.<br/><br/>More specifically still, we provide a generic framework for representing problem solving agent organisations. This serves as the platform on which we investigate approaches for structural adaptation. Following this, we demonstrate a robust, decentralised adaptation method that enables the agents to modify the organisational structure. As the method is based on self organisation principles, the agents use only their local views to change their structural relations to achieve a better allocation of tasks in the organisation. Particularly, the agents reason about when and how to adapt using only their history of interactions as guidance. We empirically show that, in a wide range of closed, open, static and dynamic scenarios, the performance of organisations using our method is close (70?90%) to that of an idealised centralised allocation method and is considerably better (10?45%) than the current state of the art decentralised approaches.

Name dc:type.qualificationname

Ph.D.

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Southampton

Year dc:date.issued

2009

Author dc:creator

• Kota, Ramachandra

Advisors dc:contributor.advisor

• Jennings, Nicholas
• Gibbins, Nicholas