CBR for modeling complex systems

Abstract

This paper describes how CBR can be used to compare, reuse, and adapt inductive models that represent complex systems. Complex systems are not well understood and therefore require models for their manipulation and understanding. We propose an approach to address the challenges for using CBR in this context, which relate to finding similar inductive models (solutions) to represent similar complex systems (problems). The purpose is to improve the modeling task by considering the quality of different models to represent a system based on the similarity to a system that was successfully modeled. The revised and confirmed suitability of a model can become additional evidence of similarity between two complex systems, resulting in an increased understanding of a domain. This use of CBR supports tasks (e.g., diagnosis, prediction) that inductive or mathematical models alone cannot perform. We validate our approach by modeling software systems, and illustrate its potential significance for biological systems. © Springer-Verlag Berlin Heidelberg 2005.