Implementation of Reservoir Computers Using Simulated Electrostatically Linked Systems

Reservoir computing has become a domain of in-tense research during the last decade. It has emerged as an al-ternative approach to Artificial Neural Networks, demonstrating specific advantages. In this work, we explore the possibility of physical systems to be acting as reservoir computers. Specifically, we have simulated the performance of a reservoir consisting of a sphere containing 700 randomly distributed nodes with random charge (positive or negative), linked only by Coulomb forces, in the case of a typical nonlinear time series prediction. The time series utilized was the one of the number of sunspots starting in 1749. Results show that the proposed topological approach for the reservoir, even if not as efficient as other implementations, is able to predict with a reasonable accuracy the evolution of that series, establishing a proof of concept.