Learning Composite Concepts

This paper proposes a framework to learn concepts from di erent kinds of observations. We de ne a language to describe meta-concepts, that represent the sets of possible concepts that can be the result of learning given a set of observations. The kinds of observations that we havestudied are subsumption,membershipand part-of. We exemplify the framework by showing how composite concepts can be learned in a speci c description logic and we show that previous machine learning approaches in description logics can be reformulated in our framework.

• Type: Conference
• Year: 1998
• Where: DLOG
• Authors: Patrick Lambrix, Pierpaolo Larocchia
• Keywords: Automated Reasoning, Description Logics, Di Erent Kinds, DLOG 1998, Possible Concepts