We make use of a probabilistic model in order to formalize the basic assumption underlying case-based reasoning (CBR), suggesting that "similar problems have similar solutions." Taking this model as a point of departure, we propose a similarity-guided inference scheme in which case-based evidence is represented in form of belief functions over the set of solutions, and in which the combination of evidence derived from individual cases is considered in the context of information fusion. Our approach is meant to support the overall process of problem solving by estimating the quality of potential solutions. Besides, it reveals that probabilistic methods and related techniques from the field of reasoning under uncertainty provide a convenient framework in which parts of the CBR methodology can be formalized. This framework seems particularly suitable since it allows for taking the heuristic and, hence, uncertain character of case-based problem solving into account.
Keywords: Case-Based Reasoning, Uncertainty in AI
Citation: Eyke Hüllermeier: Similarity-based Inference as Evidential Reasoning. In W.Horn (ed.): ECAI2000, Proceedings of the 14th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2000, pp.55-59.