A Data-Parallel Version of Aleph

This is to present work on modifying the Aleph ILP system so that it evaluates the hypothesised clauses in parallel by distributing the data-set among the nodes of a parallel or distributed machine. The paper briefly discusses the MPI specification and the extension of Yap Prolog with an interface to MPI libraries. It then proceeds to describe an implementation of data-parallel clause evaluation for Aleph through this interface. The paper concludes by testing the data-parallel Aleph on artificially constructed data-sets. Inductive Logic Programming (ILP) is the Machine Learning discipline that deals with the induction of Logic Programmes from examples. Despite the constant advances in computing hardware, ILP remains a computationally expensive application with learning sessions taking many hours or even days to complete, even on the most powerful processors. A four-day experiment that had to be repeated all over again because of some minor bug in the background knowledge or a powerful server brought to the thrashing point by a memory-hungry ILP experiment has to be accepted as a fact of life. This makes ILP algorithms good candidates for parallel or distributed computing, so that multiple CPUs or workstations can share the computational and data-storage load of an ILP experiment. This paper describes an extension of the Yap Prolog compiler with an interface to MPI libraries (Section 1.2) and an adaptation of the Aleph ILP system so that it can evaluate hypothesised clauses in parallel through this interface (Section 2). Section 3 tests Aleph/MPI and draws conclusions.