Blind Multi-class Ensemble Learning with Dependent Classifiers

Panagiotis A. Traganitis, Georgios B. Giannakis

Abstract: In recent years, advances in pattern recognition and data analytics have spurred the development of a plethora of machine learning algorithms and tools. However, as each algorithm exhibits different behavior for different types of data, one is motivated to judiciously fuse multiple algorithms in order to find the “best” performing one, for a given dataset. Ensemble learning aims to create such a high-performance meta-learner, by combining the outputs from multiple algorithms. The present work introduces a simple blind scheme for learning from ensembles of classifiers. Blind refers to the combiner who has no knowledge of the ground-truth labels that each classifier has been trained on. While most current works presume that all classifiers are independent, this work introduces a scheme that can handle dependencies between classifiers. Preliminary tests on synthetic data showcase the potential of the proposed approach.