Macroscopic dynamics prediction by reservoir computing

Abstract

Nowadays, the problem of forecasting complex signals is significant and has many applications in real life. One of such applications is the prediction of neurophysiological signals, like EEG. Such signals are macroscopic signals of a group of neurons, and the connections between them adapt in time. Here, we investigate the possibility of forecasting the dynamics of the modulated adaptive network, which topology changes in time, using Reservoir Computing (RC). We show that the dynamics of the signal is chaotic, and RC cannot predict it, but reconstruction of the phase space by adding the delays improves the quality of the signal’s prediction.