TAF PUBLISHING

This paper focuses on a machine learning problem that is especially relevant for the automation of graph-based conceptual design. In this field, graph-rewriting systems can be used to facilitate design automation. The present work deals with the automatic induction of production rules for graph-rewriting systems from example design graphs. An approach to this is presented and empirically evaluated in the context of prototypical applications for functional design. The approach suffices the requirement that the learned rules have to fit an existing rule set, but the inspection of existing rule definitions is prohibited. The results suggest that the proposed approach is able to achieve reasonable, but sometimes unexpected learning results.