Experimental Analysis of a Hybrid Reservoir Computing Technique

Abstract

Recently a new Neural Network model named Reservoir with Random Static Projections (R2SP) was introduced in the literature. The method belongs to the popular family of Reservoir Computing (RC) models. The R2SP method is a combination of the RC models and Extreme Learning Machines (ELMs). In this article, we analyse the accuracy of a variation of the R2SP that consists of using Radial Basis Functions (RBF) projections instead of ELMs. We evaluate the proposed variation on two simulated benchmark problems obtaining promising results with respect to other RC models.