In this paper we are drafting the outline of a framework for a Multi Agent System (MAS) for the support of Collaborative Design in the architectural domain. The system we are proposing makes use of Machine Learning (ML) techniques to infer personalized knowledge from observing a users’ action in a generic working environment using standard tools such as CAD packages. We introduce and discuss possible strategies to combine Concept Modelling (CM)-based approaches using existing ontologies with statistical analysis of action sequences within a domain specific application. In a later step, Agent technologies will be used to gather additional related information from external resources such as examples of similar problems on the users hard disk, from corresponding work of team-members within an intranet or from advises of expert from different knowledge domains, themselves represented by agents. As users deny or reward resulting proposals offered by the agent(s) through an interface the system will be enhanced over time using methods like Reinforced Learning.
|Title of host publication
|Developments in Design & Decision Support Systems in Architecture and Urban Planning
|J.P. Leeuwen, van, H.J.P. Timmermans
|Eindhoven University of Technology
|Published - 2004