Clustering Algorithms in Echo State Networks

Abstract

In this work, we develop a new method of setting the input to reservoir and reservoir to reservoir weights in echo state machines. We use a clustering technique which we have previously developed as a pre-processing stage to set the reservoir parameters which at this stage are prototypes. We then use these prototypes as weights in the standard architecture while setting the reservoir to output weights in a standard manner. We show results on a variety of data sets in the literature which show that this method out-performs a standard random echo state machine.