Learning from multi-source data

Abstract

This paper proposes an efficient method to learn from multi source data with an Inductive Logic Programming method. The method is based on two steps. The first one consists in learning rules independently from each source. In the second step the learned rules are used to bias a new learning process from the aggregated data. We validate this method on cardiac data obtained from electrocardiograms or arterial blood pressure measures. Our method is compared to a single step learning on aggregated data. © Springer-Verlag Berlin Heidelberg 2004.