Four-Layer Distance Metric and Distance-based Kernel Functions for Inductive Logic Programming

Nirattaya Khamsemanan, Cholwich Nattee, Masayuki Numao

Abstract

Inductive Logic Programming (ILP) is a field of study focusing developingmachine learning algorithms using logic programming to describe examples andhypotheses. This makes ILP techniques capable to deal with relational data,i.e. non-vector data. To learn from ILP data, an algorithm must be able tohandle non-linear data. Hypotheses generated from ILP techniques are in form ofHorn clauses, which can be interpreted by human. This is a benefit overconventional learning algorithms that generate black-box hypotheses orclassification models. Nevertheless, learning algorithms used by ILP techniquesare based on covering algorithms. It requires high comput