Nonlinear Dynamics of Reservoir Computing: Theory, Realization and Application

Reservoir computing leverages the response of driven dynamic systems for data-driven modeling, time-series prediction, classification, and control. In the past, recurrent neural networks have primarily served as dynamic reservoirs, but recent developments have extended this general approach to a variety of substrates, including passive and active optical systems, electro-optical systems, magnetic systems, memristors, spin wave systems, swarms, skyrmion textures, biological tissue, and quantum systems. A common goal of all these investigations is to improve our theoretical understanding of various types of reservoir computing and to achieve fast and energy-efficient hardware implementations. The performance of this type of machine learning method strongly depends on the dynamic properties of the driven reservoir system for a given input, where generalized synchronization plays an important role in achieving a reproducible and reliable dynamic response.

This Special Topic reviews recent advances in the field of reservoir computing with contributions focusing mainly on the close connections between reservoir computing and dynamical systems theory.

Topics covered include, but are not limited to:

Guest Editors

Andreas Amann, University College Cork, Ireland

Kathy Lüdge, Technische Universität Ilmenau, Germany

Ulrich Parlitz, Max Planck Institute for Dynamics and Self-Organization, Germany

Michael Small, The University of Western Australia, Australia