In this paper we consider the open problem how to unify version-space representations. We present a first solution to this problem, namely a new version-space representation called adaptable boundary sets (ABSs). We show that a version space can have a space of ABSs representations. We demonstrate that this space includes the boundary-set representation and the instance-based boundary-set representation; i.e., the ABSs unify these two representations.
We consider the task of learning ABSs as a task of identifying a proper representation within the space of ABSs depending on the applicability requirements given. This is demonstrated in a series of examples where ABSs are used to overcome the complexity problem of the boundary sets.
machine learning - concept learning - version spaces - boundary sets - instance-based boundary sets