Summary
International Symposium on Nonlinear Theory and Its Applications
2022
Session Number:B3L-E
Session:
Number:B3L-E-01
Spatiotemporal Chaotic Characteristics with Multidimensional Inputs in Echo State Networks
Takahiro Iinuma , Sou Nobukawa,
pp.351-354
Publication Date:12/12/2022
Online ISSN:2188-5079
DOI:10.34385/proc.71.B3L-E-01
PDF download (4.8MB)
Summary:
An echo state network (ESN) is a recurrent neural network model with higher learning efficiency. Recent various studies have attempted to input multidimensional data to ESN, although the performance of ESN under multidimensional inputs deteriorates. We focused on the synchrony of states of reservoir neurons and investigated the ESN characteristics under multidimensional inputs using the measure of mean correlation coefficient of pairs among all reservoir neuron states. The results showed that in the case of high-dimensional inputs, maximum memory performance is achieved when the state exhibits less than zero MLE and minimum synchronization.