TopLog: ILP Using a Logic Program Declarative Bias

This paper introduces a new Inductive Logic Programming (ILP) framework called Top Directed Hypothesis Derivation (TDHD). In this framework each hypothesised clause must be derivable from a given logic program called top theory (>). The top theory can be viewed as a declarative bias which defines the hypothesis space. This replaces the metalogical mode statements which are used in many ILP systems. Firstly we present a theoretical framework for TDHD and show that a standard SLD derivation can be used to efficiently derive hypotheses from >. Secondly, we present a prototype implementation of TDHD within a new ILP system called TopLog. Thirdly, we show that the accuracy and efficiency of TopLog, on several benchmark datasets, is comparable with the accuracy and efficiency of a state of the art ILP system like Aleph.