Logical Case Memory Systems: Foundations and Learning Issues Forschungsbericht Autor(en) / Author(s)

The focus of this paper is on the introduction of a quite general type of case-based reasoning systems called logical case memory systems. The development of the underlying concepts has been driven by investigations in certain problems of case-based learning. Therefore, the present development of the target concepts is accompanied by an in-depth discussion of related learning problems. Logical case memory systems provide some formal framework for the investigation and for the application of structural similarity concepts. Those concepts have some crucial advantage over traditional numerical similarity concepts: The result of determining a new case's similarity to some formerly experienced case can be directly taken as a basis for performing case adaptation. Essentially, every logical case memory system consists of two constituents, some partially ordered case base and some partially ordered set of predicates. Cases are terms, in a logical sense. Given some problem case, every predicate determines some view at the case base. Thus, the partial ordering of predicates is specifying some similarity concept in use. Within this formal framework, learning similarity concepts turns out to be learning of partial orderings. This insight allows for the derivation of some basic results about the power and limitations of case-based learning. It provides a rm basis for a wide spectrum of further investigations. The research reported here has been partially supported by the German Research Fund (DFG) within the research project Ind{Cbl under contract no. Ja 566/3{3.