The Prioritized Inductive Logic Programs

TL;DR

This paper explores the long-term behavior of inductive logic programs, introduces the concepts of convergence and limit-correctness, and proposes a new prioritized GOLEM system to ensure limit correctness in incremental learning.

Contribution

It defines limit behavior for inductive logic programs and introduces the prioritized GOLEM system to achieve limit correctness in online learning.

Findings

The GOLEM system is not limit-correct.

A new prioritized GOLEM system is proposed.

The prioritized GOLEM system ensures limit correctness.

Abstract

The limit behavior of inductive logic programs has not been explored, but when considering incremental or online inductive learning algorithms which usually run ongoingly, such behavior of the programs should be taken into account. An example is given to show that some inductive learning algorithm may not be correct in the long run if the limit behavior is not considered. An inductive logic program is convergent if given an increasing sequence of example sets, the program produces a corresponding sequence of the Horn logic programs which has the set-theoretic limit, and is limit-correct if the limit of the produced sequence of the Horn logic programs is correct with respect to the limit of the sequence of the example sets. It is shown that the GOLEM system is not limit-correct. Finally, a limit-correct inductive logic system, called the prioritized GOLEM system, is proposed as a solution.