Constructing the extensional representation of an intensional domain theory in inductive logic programming

Abstract

Inductive Logic Programming — ILP — is a new paradigm in the field of Machine Learning, which adopts the language of logic programs as a description language for the expression of instances, hypothesis and domain theory. Learning logical definitions requires the exploration of a very large space of hypothesis descriptions and, consequently, restrictions should be imposed on the hypothesis space to make learning a feasible task. In this work we discuss ways of confining the ILP hypothesis space by restricting the domain theory description language. Specifically we focus on the notion of generative clauses, which enable the construction of a restricted form of Horn clause program. In fact, the use of generative logic programs as description language allows for automatic transformation of the intensional expression of a domain theory into its extensional expression, as required by many existing ILP systems. We present an ILP environment which takes as input an intensional domain theory and, after verifying that all clauses in the theory are generative, automatically constructs the extensional expression of this domain theory. We illustrate the use of this implemented environment through the ILP system GOLEM.KeywordsLogic ProgramDescription LanguageInductive LogicInductive Logic ProgrammingHorn ClauseThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.