Improving boosting performance with a local combination of learners

This work explores the possibility of improving the performance of Real Adaboost ensemble classifiers by replacing their standard linear combination of learners by a gating scheme. This more powerful fusion method is defined following the epoch-by-epoch construction of boosting ensembles. Preliminary experimental results support the potential of this new approach.

I. Introduction

To combine Neural Networks (NNs) is more effective than trying to solve difficult problems by using a big size single NN. Not only an easier design and better accuracy can be obtained, but also a clearer understanding of how the resulting machine works is possible. In recent years, there have been several proposals to build ensembles of learning machines [1], [2]; Mixture of Experts (MoE) and Boosting are among the most relevant.