Self-modeling databases

The Carper system, which uses inductive learning to check database consistency, even in poorly understood domains, is described. The application of Carper to the Xcon expert system database is discussed. It is shown that Carper can detect five general error types in Xcon: using value naming conventions inconsistently, assigning legal but incorrect values to attributes, omitting obscure but necessary attribute values, assigning values to attributes that should be left undefined, and failing to update attribute values when dependent attribute values change.<<ETX>>