Computing with transient states generated by delay-coupled systems

Abstract

Inspired by the way the brain process information, the recently introduced paradigm of reservoir computing has been proven to be very successful in solving certain tasks that are inherently difficult for traditional computers. Reservoir computing based on delay-coupled systems has been shown to perform as well as traditional reservoir-computing methods allowing to replace the large networks by a small number of delayed-coupled dynamical elements. Its simplest manifestation, a single nonlinear node subject to delayed feedback, provides excellent performance in certain tasks as spoken digit recognition and time series predictions.