Skill Refinement through Competence Feedback

Learningis generally performedin twostages, knowledge acquisition and skill refinement. Developments within machinelearning havetended to concentrate on knowledgeacquisition as opposedto skill refinement. In this paper wedevelopmechanismsfor skill refinementwithinthe context of lazy learning by incorporating competencefeedback into the exemplar base. Theextent of competencefeedbackdependson the richness of the data collected duringtask execution. Wepresent twotechniquesfor competencefeedback, exception spaces and knowledgeintensive exception spaces (KINS).Thetechniquesdiffer in the extent of competencefeedbackand the resulting degree of skill refinement.Previousdefinitions of exception spaces and KINSare extendedandthe resulting improvementis evaluated using six data sets. A genetic algorithmis utilised to optimisethe definitions of the exceptionspaces and KINSfor each exemplarin the exemplarbase. Wealso providea visualisation of KINSand exceptionspaceswith an aimto exemplifyhowthese ...

• Type: Conference
• Year: 2001
• Where: FLAIRS
• Authors: David W. Patterson, Sarabjot S. Anand, John G. Hughes
• Keywords: Artificial Intelligence, Exception Spaces, FLAIRS 2001, Opposedto Skill Refinement, Skill Refinement