Adaptive Ensemble Variants of Random Vector Functional Link Networks

Abstract

In this paper, we propose a novel adaptive ensemble variant of random vector functional link (RVFL) networks. Adaptive ensemble RVFL networks assign different weights to the sub-classifiers according to prediction performance of single RVFL network. Generic Adaptive Ensemble RVFL is composed of a series of unrelated, independent weak classifiers. We also employ our adaptive ensemble method to the deep random vector functional link (dRVFL). Each layer in dRVFL can be regarded as a sub-classifier. However, instead of training several models independently, the sub-classifiers of dRVFL can be obtained by training a single network once. © 2020, Springer Nature Switzerland AG.