Learning approximate diagnosis

Abstract

In earlier work on incorporating explanation-based learning (EBL) in model-based diagnosis (MBD), a diagnostic architecture integrating EBL and MBD components was suggested. The authors relax the requirement on completeness and specificity of the diagnostic candidates. They allow the learning component to make errors in a training phase where it is given feedback on its actual performance. A method is described for trading off accuracy for efficiency. In this approach, most diagnosis problems are handled by the associational rules learned from previous problems. Model-based reasoning and learning are activated only when performance drops below a given threshold. Empirical results are presented on circuits with an increasing number of components illustrating how this approach scales up.<>