The role of pulvinar in information transmission through the visual cortical hierarchy

Nelson Cortes (1) , Carl van Vreeswijk (1)

1 NPSM - Neurophysique et physiologie du système moteur

A hierarchical model of ten cortical areas is examinated to explain representation of stimulus contrast. In each area, the activity is integrated and then non-linearly transmitted feedforwardly to the next area. This arrangement of interactions creates a gradient from simple to complex visual patterns as one move from lower to higher cortical levels. In the model, each unit describes the neural activity represented by a fring-rate model where the gain controls the steepness of the input-output relation. A Gaussian random input is applied to level one as the distribution was modified in width, similar to variation in contrast in visual stimuli. The output activity ratio between a lower and higher contrast is analysed for the last level to observe sensitivity to a contrast and contrast invariant tuning. For a purely feedforward system, in a small gain range, the output of the last area presents aproximate contrast invariance, whereas the sensitivity to contrast was poor. To analyse other cortico-cortical interactions, positive and negative feedback connections from higher to lower cortical areas are then incorporated. In this case, neither the sentivity to contrast nor the contrast invariance of the tuning were improved. To account for an alternative visual processing pathway, non-reciprocal connections from and to a paralel pulvinar like-structure of nine areas are coupled to the system. Compared to the pure feedforward model, cortico-pulvino-cortical output presents more sensitivity to contrast and keep similar values of the tuning contrast invariance.

Mots clés: Biophysical models; Computational vision; Information coding; System dynamics

Domaines: Neurosciences [q-bio.NC]; Neurosciences

Nelson Cortes, Carl van Vreeswijk. The role of pulvinar in information transmission through the visual cortical hierarchy. Cinquième conférence plénière française de Neurosciences Computationnelles, "Neurocomp'10", Aug 2010, Lyon, France. ⟨hal-00553446⟩