Inductive Learning of Normal Clauses

In this paper, we are interested in the induction of normal clauses. Several semantics have been proposed for normal programs, we consider here the well-founded semantics, based on a three-valued logics. We have generalized to this semantic the classical constraint: all the positive examples must be covered by the learned program and all the negative ones must be rejected. In the case of inductive learning, this constraint seems too strong and we have deened a weaker criteria: we require that a positive (resp. negative) example is not considered as False (resp. True) by the learned program. This study has been applied to the system FOIL. Some positive examples that were covered by the program during the learning process were not in the well-founded semantics. and we deene biases to solve this problem.