Modifying the Model Set During Diagnosis

Abstract

GDE presented a very clean approach to model-based diagnosis. Recent work by de Kleer & Williams has extended the original system to include multiple models for components. When many of the models for the components are instantiated from the beginning of a diagnosis session, however, the cost of doing predictions quickly explodes.

This paper presents an extension to gde-2, in which observations are explicitly maintained in a context. This allows models to be added or removed during the course of diagnosis, thus ameliorating the combinatoric problem.

The author thankfully acknowledge the support provided by SIEMENS for this research.