Non-parametric spiking neural network modelling of the eye-movement response to enforced controlled accelerations

The main objective of this study is to present a non-parametric model of the electrophysiological interplay between eye response (non-voluntary movements) and sensed head acceleration in the otholith system. The model is based on a class of spiking differential neural network with time dependent learning laws. These laws are developed using a class of control Lyapunov functions that takes weights as parameters that can enforce the origin that is a practical equilibrium point of the identification error. The network topology draws upon the Izhikevich representation of the artificial neuron activity. Each of the artificial neurons uses a model with fixed parameters that represents the evolution of the bioinspired artificial neuron. The modeling strategy consists in implementing an experimental system that collects data on three-axes translational acceleration as well as angular velocities from volunteers. An eye tracker device has collected information on eye movements which have been correlated to the head dynamic movement. The modeling process has been proven to be efficient considering the nature of the information provided by the experimental system. The benefits of using Izikevich artificial neurons have been evaluated by comparing the developed identifier with the modeling results obtained with the help of a traditional neural network that used sigmoidal neuron representation. The least mean square error for Izikevich-based identifier is 73 percent smaller due to the biologically-inspired nature of this activation function as an approximated model of the vestibulo-ocular reflex.