Novel Reservoir Computing Approach for the Detection of Chaos
Résumé
Reservoir Computing (RC) is a novel framework for data computation extending from recurrent neural networks. The high dimensionality and flexibility of such processing paradigm make it well worthy for high complex systems analysis. Conceptor-driven Network (ConDN) is a RC-based paradigm whose unique structure is well appropriate for modelling and analysing multiple-input systems. In this paper, we present a ConDN approach for chaos detection in systems. Some known parametrizable systems exhibiting chaotic and non-chaotic behaviours have been analysed. By only observing the selected output results (trace matrix) after training of the ConDN method, the systems can be classified as chaotic or non-chaotic. In addition, the robustness of this method against white noise is also investigated.