Using reservoir computers to distinguish chaotic signals

TL;DR

This paper demonstrates that reservoir computers can effectively classify different Sprott chaotic systems without requiring embedding, and explores how parameters like data points, reservoir size, and noise affect performance.

Contribution

It introduces the use of reservoir computers for identifying Sprott systems and provides guidance on parameter selection for this classification task.

Findings

Reservoir computers can classify Sprott systems without embedding.

Performance depends on number of data points, reservoir nodes, and noise levels.

Guidelines for choosing reservoir parameters are provided.

Abstract

Several recent papers have shown that reservoir computers are useful for analyzing and predicting dynamical systems. Reservoir computers have also been shown to be useful for various classification problems. In this work, a reservoir computer is used to identify one out of the 19 different Sprott systems. An advantage of reservoir computers for this problem is that no embedding is necessary. Some guidance on choosing the reservoir computer parameters is given. The dependance on number of points, number of reservoir nodes and noise in identifying the Sprott systems is explored.