Some circumscriptional thoughts on SBL

Abstract

We illustrate how some of the Similitude Based Learning (SBL) paradigms can be reformulated using some logical formalisms as circumscription, predicate completion and the close world assumption. Our approach shows that is possible to use these logical tools in order to obtain a formal unified vision of SBL paradigms, and it also suggests some kind of improvements on the current implementations. We introduce a hitherto unmentioned direct link between machine learning and circumscription. We believe that this new framework and the obtained results are of practical interest.