Feature Selection for the Stochastic Integrate and Fire Model

This paper presents a novel training method for estimating the parameters of integrate and fire retina models. The presented model is described by a set of linear and nonlinear filters, described by basis functions and Taylor polynomials, respectively. This allows for the identification of a set of features which can be used for reproducing retina responses. A Bayesian-Laplace feature selection is proposed to choose which features can be eliminated. Thus, we are able to achieve a model using a reduced set of parameters. Experimental results show that the proposed algorithm is able to remove non-important features while still accurately reproducing retina responses.