Prototype-Based Reasoning: An Integrated Approach to Solving Large Novel Problems

Two important computational approaches to problem solving are model-based reasoning (MBR) and case-based reasoning (CBR). MBR, since it reasons from first principles, is especially suited for solving novel problems. CBR, since it reasons from previous experience, is especially suited for solving frequently encountered problems. However, large novel problems pose difficulties for both approaches-MBR rapidly grows intractable and CBR fails to find a relevant previous case. In this paper we describe an approach called prototypebased reasoning that integrates both approaches to solve such problems. Prototype-based reasoning treats a large novel problem as a novel combination of several familiar subproblems. It uses CBR to find and solve the subproblems, formulates a new problem by combining these individual solutions, and uses MBR to solve this new problem. We demonstrate the effectiveness of this method on several examples involving the causal simulation of complex electronic circuits.