Explanation-based learning with diagnostic models

Abstract

The author discusses an approach to identifying and correcting errors in diagnostic models using explanation based learning. The approach uses a model of the system to be diagnosed that may have missing information about the relationships between tests and possible diagnoses. In particular, he uses a structural model or information flow model to guide diagnosis. When misdiagnosis occurs, the model is used to determine how to search for the actual fault through additional testing. When the fault is identified, an explanation is constructed from the original misdiagnosis and the model is modified to compensate for the incorrect behavior of the system.<>