Population coding of motion patterns in the early visual system

Abstract

Using extracellular recordings and computational modeling, we study the responses of a population of turtle (Pseudemys scripta elegans) retinal ganglion cells to different motion patterns. The onset of motion of a bright bar is signaled by a rise of the population activity that occurs within less than 100 ms. Correspondingly, more complex stimulus movement patterns are reflected by rapid variations of the firing rate of the retinal ganglion cell population. This behavior is reproduced by a computational model that generates ganglion cell activity from the spatio-temporal stimulus pattern using a Wiener model complemented by a non-linear contrast gain control feedback loop responsible for the sharp transients in response to motion onset. This study demonstrates that contrast gain control strongly influences the temporal course of retinal population activity, and thereby plays a major role in the formation of a population code for stimulus movement patterns.

Computational Models of Neural Retina

“…A more complex block-structured architecture of the retina is shown in Fig. 2 (Wilke et al 2001). In this model, a spatiotemporal light stimulus s(r,t) is transformed through linear convolution into a neural activation signal u(r,t) according to u r,t ð Þ ¼ is then multiplied by a local modulation factor g(r,t) between 0 and 1, scaled, and rectified to produce the output local RGC firing rate f(r,t):…”

Visual neuroprosthesis: a non invasive system for stimulating the cortex

“…It is composed by two main modules connected in series. The early layers, based on models previously proposed in [25], is responsible for processing the visual signal and for its conversion into a spike rate. This rate is then taken as input by the neuromorphic pulse coding (NPC) module and translated into the actual spike events by using a simplified version of the integrate-and-fire spiking neuron [39].…”

“…The early layers block, responsible for spatio-temporal filtering, is based on research published in [25] and extended to the chromatic domain by considering independent filters for each basic color component. The first filtering element of the early layers is an edge detector composed by a set of two Gaussian spatial filters per color channel [see Fig.…”

“…Another block is required to convert this firing rate into spike trains. In 2001, a new model was proposed that introduced a nonlinear feedback loop at the output of the temporal filter [25], performing a contrast gain control (CGC). The loop is composed by a temporal low-pass filter followed by a nonlinear function and feedback through a multiplier, introducing a modulation factor in the response.…”

“…Based on this study the model with the CGC [25] was adopted in the CORTIVIS project. Particular aspects of the retina organization are considered in this work.…”