From sequential to parallel Inductive Logic Programming

Abstract

Inductive Logic Programming (ILP) has achieved considerable success in a wide range of domains. It is recognized however that eciency is a major obstacle to the use of ILP systems in applications requiring large amounts of data. In this paper we address the problem of eciency in ILP in three steps: i) we survey speedup techniques proposed for sequential execution of ILP systems; ii) we survey dierent ways of parallelizing an ILP system and; ii) adapt and combine the sequential execution speedup techniques in the parallel implementations of an ILP system. We also propose a novel technique to partition the search space into independent sub-spaces that may be adequately searched in parallel.