Incremental Refinement of Approximate Domain Theories

Ronen Feldman, Alberto Maria Segre, Moshe Koppel

Abstract: Abstract In this paper we present a framework for performing incremental correction of approximate domain theories. Approximate domain theories are domain theories which are incomplete and/or incorrect. Based on initial information, belief values are assigned to different subsets of each clause in the domain theory. These belief values provide bias towards the correct refinement of the domain theory. We provide an incremental algorithm that refines the domain theory after observing positive and negative exemplars. Our algorithm requires a smaller number of misclassified exemplars than other algorithms presented in the literature.