A FAST DEEP STACKED NETWORK USING EXTREME LEARNING MACHINE

BRUNO LÉGORA SOUZA DA SILVA; FERNANDO  KENTARO INABA; PATRICK MARQUES CIARELLI

doi:10.20906/CBA2022/692

BRUNO LÉGORA SOUZA DA SILVA Universidade Federal do Espírito Santo
FERNANDO KENTARO INABA Universidade Federal do Espírito Santo
PATRICK MARQUES CIARELLI Universidade Federal do Espírito Santo

DOI: https://doi.org/10.20906/CBA2022/692

Keywords: Extreme Learning Machine, Deep Stacked Network, Big data, Classification

Abstract

Access to large amounts of data is becoming more common, as well as the use of methods based on "deep" learning to obtain better results. However, using those techniques can lead to long training times. To deal with this problem, the Deep Stacked Network (DSN) was proposed, where several small modules are stacked to increase the model eciency. However, this architecture suers from some problems that increase its training time and the amount of memory required to store it. To deal with some of these problems, we propose in this paper a fast algorithm to train a DSN using Extreme Learning Machine (ELM). Experiments performed on many classication datasets showed that the proposed method achieves similar accuracy when compared with other techniques, with the advantage of training the network in less time and storing fewer parameters.