Analysis of the Effect of a Recurrent Neural Network Size on Modeling and Prediction Accuracy of a Stochastic FitzHugh–Nagumo Neuron

We study the performance of a reservoir computing-based network in dynamics forecast of a FitzHugh–Nagumo model driven by white noise versus reservoir size. We show that the accuracy of signal prediction of a model neuron strongly depends on the number of neurons in the reservoir. The most accurate prediction of both the dynamics itself and the simulation of the coherence resonance phenomenon is achieved when using 500 neurons.