Parallel Induction Algorithms for Large Samples

ILP, in short, induces a hypothesis H that explains target relations from their positive and negative examples E with background knowledge B. QUINLAN's FOIL [1], known as one of the most successful ILP systems, requires a set of all examples T R belonging to relation R in B. FOIL starts with the left-hand side of the clause and specializes it keeping a training set T that starts from E to evaluate literals. While a clause C covers negative examples, FOIL adds a literal L R using R to C, T and T R are joined to T 0 consisting of tuples covered by C ^ L R . Positive examples covered by the found clause are removed from E, and FOIL returns the loop with the new training set.