Theory revision via prior operationalization

Abstract

Research in machine learning often focuses either on inductive learning - learning from experience with minimal reliance on prior theory - or, more recently, on explanation-based learning- deducing general descriptions from theories with minimal reliance on experience. Theory revision unites these two concerns: one must revise one's theory in the light of experience, but one must simultaneously use information implicit in the theory in order to guide the revision process. This paper focuses on the second step of a unified three-step method for solving theory revision problems for certain classes of empirical theories. Test results for the first two phases of this approach are reported.