Technical Report: DCC-2003-04

On avoiding redundancy in Inductive Logic Programming systems

Nuno Fonseca(1), Vitor Santos Costa(2), Fernando Silva(1), Rui Camacho(3)

(1) DCC-FC & LIACC, Universidade do Porto
R. do Campo Alegre 823, 4150-180 Porto, Portugal

(2) COPPE/Sistemas, Universidade Federal do Rio de Janeiro
Centro de Tecnologia, Bloco H-319, Cx. Postal 68511 Rio de Janeiro, Brasil
(3) Faculdade de Engenharia & LIACC, Universidade do Porto
Rua Dr. Roberto Frias, s/n 4200-465 Porto, Portugal

November 2003


Inductive Logic Programming (ILP) is a subfield of Machine Learning that provides an excellent framework for learning in multi-relational domains and inducing first-order clausal theories. ILP systems perform a search through very large hypothesis spaces containing redundant hypotheses. The generation of redundant hypotheses may prevent the systems from finding good models and increase the time to induce them. In this paper we propose a classification of hypotheses redundancy. We show how expert knowledge can be provided to an ILP system to avoid the generation of redundant hypotheses. Preliminary results suggest that the the number of hypotheses generated and execution time are substancially reduced when using expert knowledge to avoid the generation of redundant hypotheses.