Distributed Genetic Programming with Mobile Agents

This paper presents a system for accessing the potential of Genetic Programming [Koza 1992] by exploiting the availability of computer networks to distribute the computational load over a large number of machines. The goal is to utilise the CPU and memory resources of Intranet systems, to facilitate GP operations. In addition the use of a large-scale parallel network of machines allows a demetic grouping of subpopulations to be created, in order to enhance the genetic diversity within a GP system. While several existing schemes for parallelisation of GP exist, (e.g. Andre & Koza 1996) this work demonstrates how Java based mobile agents are ideally suited to this task. An intelligent agent infrastructure has been created, which assists a user by automating the deployment and monitoring of a parallel GP system using mobile agents. The major benefits realised are greater cost efficiency in utilising available computer power and ease of use.