Abstract
Researchers in artificial intelligence and agent technologies are presented with a massive array of various technologies that they might use for their research projects. It is difficult for researchers to test their theories effectively in the field. It takes a great deal of time to develop the platform on which the newly created agent will be tested, with little or no time left for troubleshooting and the investigation of further solutions. Every time a new technique or agent is researched, the agent has to be redeveloped from the ground up. This makes it difficult for researchers to compare their own theories with previously developed components.
With the wide range of technologies and techniques available, there is no easy way to effectively make use of the various components, as each tool uses different technologies that cannot be combined easily.
This dissertation outlines the new plug-in oriented agent architecture (POAA) and describes the agents that use the POAA. POAA agents make extensive use of functional and controller-based plug-ins in order to extend the functionality and behaviour of the agent. The architecture was designed to facilitate machine learning and agent mobility techniques. POAA agents are created by mounting newly created dynamic plug-in components into the static structure of the agent. The static structure of the agent serves as the basis of agent functionality and as the controller for the agent’s life cycle. The static and dynamic components of the POAA agent interact with each other in order to perform the agent’s required tasks.
The use of plug-ins will greatly improve the effectiveness of researchers, as only a single, standard architecture will exist. Researchers only need design and develop the plug-in required for their specific agent to function as desired. This will also facilitate
the comparison of various tools and methods, as only the components being reviewed need to be interchanged to measure system performance.
The use of different plug-in architectures is also investigated. This includes deciding if the plug-in base will be configured at application run-time or at the time of application compilation.
This dissertation focuses on techniques that will facilitate machine learning and agent mobility. For these purposes, extensive use is made of the machine learning tool WEKA developed by University of Waikato in New Zealand [Wi00]. The use of Java in the prototype will also facilitate the cross platform capability of the proposed agents.
Prof. E.M. Ehlers