Towards a hybrid model of first-order theory refinement

Abstract

The representation and learning of a first-order theory using neural networks is still an open problem.We define a propositional theory refinement system which uses min and max as its activation functions, and extend it to the first-order case. In this extension, the basic compu-tational element of the network is a node capable of performing complex symbolic processing. Some issues related to learning in this hybrid model are discussed.