A framework and an analysis of current proposals for the case-based organization and representation of procedural knowledge
Roland J. Zito-Wolf, Richard Alterman · National Conference on Artificial Intelligence · 1993
Case-based refers to the class of memory-based problem solving methods which emphasize the adaptation of recalled solutions (explanations, diagnoses, plans) over the generation of solutions from first principles. CBR has become a popular methodology, resulting in a proliferation of case organization and representation proposals. The goal of this paper is to sort through some of these proposals. Using the formal models of procedure and case-based reasoning introduced in Zito-Wolf and Alterman (1992), we compare three current proposals for the organization of procedural case-bases: individual cases, microcases, and multi cases. We give a worst-case analysis that shows the advantages of the multi case in terms of case storage and retrieval costs. The model predicts that multi cases reduce case storage and retrieval costs as compared to the other two models. We then provide some empirical evidence from an implemented system that suggests that the trends observed in the formal model are also observable in case bases of practical size.