Reference: Nayak, P. Reasoning with Multiple Models. Knowledge Systems Laboratory, June, 1989.
Abstract: It is well accepted that model-based reasoning is a powerful tool to achieve robust problem-solving. However, often, the price that must be paid for this robustness is excessive computational cost. One approach to managing this excessive computational cost is to use multiple, approximate models of the domain. Given a problem, one must try and choose the simplest available model that provides acceptable answers for that problem. To reason successfully with multiple, approximate models of the domain, a number of issues need to be addressed. These include ways of picking the appropriate model for each problem, ways of detecting that the current model is inappropriate, and ways of finding and creating more appropriate models when the current model is found to be inappropriate. In this paper we will discuss these issues in some detail and provide some approaches to addressing them.
Full paper available as tar.