Efficient Inductive Logic Programming Based on Particle Swarm Optimization

Abstract

Inductive Logic Programming (ILP) is an inductive reasoning method based on the first-order predicative logic. This technology is widely used for data mining using symbolic artificial intelligence. ILP searches for a suitable hypothesis that covers positive examples and uncovers negative examples. The searching process requires a lot of execution cost to interpret many given examples for practical problems. In this paper, we propose a new hypothesis search method using particle swarm optimization (PSO). PSO is a meta-heuristic algorithm based on behaviors of particles. In our approach, each particle repeatedly moves from a hypothesis to another hypothesis within a hypothesis space. At that time, some hypotheses are refined based on the value returned by a predefined evaluation function. Since PSO just searches a part of the hypothesis space, it contributes to the speed up of the execution of ILP. In order to demonstrate the effectiveness of our method, we have implemented it on Progol that is one of the ILP systems [6], and then we conducted numerical experiments. The results showed that our method reduced the hypothesis search time compared to another conventional Progol.