Spaces of Theories with Ideal Refinement Operators

Refinement operators for theories avoid the problems related to the myopia of many relational learning algorithms based on the operators that refine single clauses. However, the non-existence of ideal refinement operators has been proven for the standard clausal search spaces based on 0-subsumption or logical implication, which scales up to the spaces of theories. By adopting different generalization models constrained by the assumption of object identity, we extend the theoretical results on the existence of ideal refinement operators for spaces of clauses to the case of spaces of theories. 1 Motivation In the investigation of the algorithms for relational learning, regarding induction as a refinement process allows to decouple search from heuristics. Therefore, the choice of the generalization model for a search space plays a key role since it affects both its algebraic structure and the definition of refinement operators for that space. Logical implication and 0-subsumption are the...

• Type: Conference
• Year: 2003
• Where: IJCAI
• Authors: Nicola Fanizzi, Stefano Ferilli, Nicola Di Mauro, Teresa Maria Altomare Basile
• Keywords: Generalization Model, IJCAI 2003, IJCAI 2007, Refinement Operators, Search Spaces