Learning naive Bayes models for multiple-instance learning with label proportions

Abstract

This paper deals with the problem of multi-instance learning when label proportions are provided. In this classification problem, the instances of the dataset are divided into disjoint groups, where there is no certainty about the labels associated with individual samples. However, in each group the number of instances that belong to each class is known. We propose several versions of an EM-algorithm that learns naive Bayes models to deal with the exposed problem. The proposed algorithms are evaluated on synthetic and real datasets, and compared with state-of-the-art approaches. The obtained results show a competitive behaviour of our proposals. © 2011 Springer-Verlag.