Using Case-Base Data to Learn Adaptation Knowledge for Design

Abstract

One advantage of Case-Based Reasoning (CBR) is the relative ease of constructing and maintaining CBR systems, especially as a number of commercial CBR tools are available. However, there are areas of CBR that current tools have not yet addressed. One of these is easing or automating the acquisition of adaptation knowledge. Since tasks like design or planning typically require a significant amount of adaptation, CBR systems for these tasks still do not fully benefit from CBR’s promise of reducing the development effort. To address this, we have developed several “knowledge-light” methods for learning adaptation knowledge from the cases in the case-base. These methods perform substitutional adaptation, for both nominal and numerical values, and are suitable for decomposable design problems, in particular formulation and configuration. Tests performed on a tablet formulation domain show promising results. The automatic adaptation methods we present can easily be incorporated in general-purpose CBR tools, thus further contributing to reducing the cost of CBR systems.