Induction Over the Unexplained: Integrated Learning of Concepts with Both Explainable and Conventional Aspects

ABSTRACT This paper presents a new approach to combining explanation-based and empirical learning called Induction Over the Unexplained (IOU). Unlike other approaches to integrated learning, which use one method to focus the other or provide it with information, IOU uses each method to learn a different part of the final concept definition. It is therefore suited for learning concepts with both explainable and unexplainable aspects. An initial nonincremental feature-based implementation of IOU is presented together with an example illustrating IOU's advantage over a purely empirical or analytical system and over other integrated learning systems such as IOE.