The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #94135

Title:Madlab; Masking and Multiple Bug Diagnosis
Date: 1994
Abstract:Multiple bug modelling involves the construction of procedural student models in which there are more than one bug. Previous work on multiple bug modelling has identified multiple bug interactions as a challenge for diagnostic algorithms (Burton, 1982). The term bug masking is used here to define the situation where there is no simple relationship between the presence of an atomic bug in a multiple bug hypothesis and the performance of a multiple bug hypothesis. By viewing multiple bug diagnosis as optimisation problem a measure of bug masking is constructed and standard optimisation algorithms are applied to multiple bug diagnosis. The Masking and Diagnosis Lab (MADLab) is a domain independent toolkit for exploring bug masking in multiple bug modelling. The MADLab system is applied to the domains of Prolog unification, Prolog control and subtraction. Experiments conducted with student data in the domains of Prolog unification and Prolog control provide evidence about the causes of bug masking and the difficulty of multiple bug diagnosis. An assessment of the experimental results points the way to future work on the application of the MADLab approach to further domains, on improved measures of bug masking, and on the effect of bug order in multiple bug diagnosis.

[Search These Pages] [DAI Home Page] [Comment]