The Parallelization of a Knowledge Discovery System with Hypergraph Representation

Abstract. Knowledge discovery is a time-consuming and space intensive endeavor. By distributing such an endeavor, we can diminish both time and space. System INDEDpronounced indeed" is an inductive implementation that performs rule discovery using the techniques of inductive logic programming and accumulates and handles knowledge using a deductive nonmonotonic reasoning engine. We present four schemes of transforming this large serial inductive logic programming ILP knowledge-based discovery system into a distributed ILP discovery system running on a Beowulf cluster. We also present our data partitioning algorithm based on locality used to accomplish the data decomposition used in the scenarios.

• Type: Conference
• Year: 2000
• Where: IPPS
• Authors: Jennifer Seitzer, James P. Buckley, Yi Pan, Lee A. Adams
• Keywords: Discovery System, Distributed And Parallel Computing, Inductive Logic Programming, IPPS 2000, Space Intensive Endeavor