Abstract
Inductive Logic Programming (ILP) [1] systems are general purpose learners that have had significant success on solving a number of relational problems, particularly from the biological domain [2,3,4,5]. However, the standard compression guided top-clown search algorithm implemented in state of the art ILP systems like Progol [6] and Aleph [7] is riot ideal for the Head Output Connected (HOC) class of learning problems. HOC is the broad class of predicates that have at least one output variable in the target concept. There are many relevant learning problems of this class such as arbitrary arithmetic functions and list manipulation predicates which axe useful in areas such as automated software verification[8]. In this paper we present a special purpose ILP system to efficiently learn HOC predicates.