Additive Logistic Regression Applied to Retina Modelling

The accurate modelling of the human visual system, particularly of the retina, would be a great achievement and a big step in the development of visual prostheses. Several methods and algorithms have been proposed to accomplish such a difficult task, mainly to what concerns the adaptation and nonlinear mechanisms of the retina. This paper presents the results obtained by the employment of additive logistic regression techniques to model the nonlinear block of a canonical Linear-Nonlinear-Poisson retina model, considering the spike triggering process from a statistical point of view, complemented with the PCA of the stimuli covariance matrix. The displayed results were obtained by modelling real retina data using different forms for the nonlinear block and are assessed with different error measures.