Reservoir computing system using discrete memristor for chaotic temporal signal processing

Abstract

Reservoir computing (RC) is a highly efficient neural network for processing temporal signals, primarily due to its significantly lower training cost compared to standard recurrent neural networks. In this work, a novel discrete memristor (DM) model is investigated and a simple two-dimensional chaotic map based on the DM model is presented, in which complex dynamics are simulated. By utilizing this DM-based map as a reservoir, a dynamic DM-based RC system is constructed, and the performance is verified through nonlinear regression and time-series prediction tasks. Our system achieves a high accuracy rate of 99.99 % in the nonlinear recognitions, as well as a low root mean square error of 0.0974 in the time-series prediction of the Logistic map. This work may pave the way for the future development of high-efficiency memristor-based RC systems to handle more complex temporal tasks.