Forecasting coherence resonance in a stochastic Fitzhugh-Nagumo neuron model using reservoir computing

We delve into the intriguing realm of reservoir computing to predict the intricate dynamics of a stochastic FitzHugh-Nagumo neuron model subjected to external noise. Through innovative reservoir design and training, we unveil the remarkable capacity of a reservoir computer to forecast the behavior of this stochastic system across a wide spectrum of noise intensity variations. Notably, our reservoir computer astutely replicates the intricate phenomenon of coherence resonance in the stochastic FitzHugh-Nagumo neuron, signifying the superior modeling capabilities of this approach. A detailed examination of the microscopic dynamics within the reservoir's hidden layer reveals the emergence of distinct neuronal clusters, each displaying unique behaviors. Certain neurons within the reservoir are adept at faithfully reproducing the dynamical traits of neuron, particularly the spike generation mechanism. In contrast, the remaining neurons within the reservoir seem to emulate stochastic influences with remarkable precision, accurately capturing the moments of spike generation in the neuron under the sway of noise. This innovative reservoir design proves to be highly effective across a diverse range of noise control parameters, faithfully replicating the essential characteristics of original stochastic FitzHugh-Nagumo neuron. These findings illuminate the potential of reservoir computing to model and predict the dynamics of complex stochastic systems, showcasing its adaptability and versatility in understanding and simulating natural phenomena.