Stochastic integrate-and-fire model for the retina

Prostheses are an efficient way of alleviating some of the handicaps suffered by the disabled. One of the most prominent impairments which would greatly benefit from the existence of visual prosthesis is blindness. Several models and training algorithms have been proposed to reach such aim. This paper presents a stochastic model for the retina and introduces a training method for fitting the model to real data. The model is based on an integrate-and-fire scheme under additive white noise. A gradient ascent training method is used to maximize the probability of occurrence of spike events at a given set of time stamps. The model is trained using real data and the results are evaluated by using different error measures. The quality and the validity of the whole process is discussed based on that analysis.