The paper presents experiments in learning in a multi-agent design system. Inductive techniques are used to construct models of other agents' behaviors, by relating design requirements to design proposals. The models are used to anticipate future design proposals and reduce the amount of interaction needed to find solutions for parametric design problems. 1. Introduction This paper present
... [Show full abstract] experiments in learning in a multi-agent design system. We approach non-routine parametric design problems by using small, specialized, knowledge-based design agents called Single Function Agents (SiFAs). The size of the agents makes them very `competent' in their decisions, but imposes an overhead on the design system due to the complexity of the task of reaching an agreement. The number of interactions between agents can be significantly reduced if the agents learn to predict the responses from the other agents with which they interact. We first describe what SiFAs are and how th...