A multi-regional neural network model for conscious perception
Date:
Oxford University, Mathematical Insitute
Models of Consciousness 4 (MoC4) Conference 2023
Oxford, UK
by the Oxford Mathematics of Consciousness and Applications Network (OMCAN)
video of the talk in youtube
Abstract
A multi-regional neural network model for conscious perception Most neural network models that have been employed to study conscious perception are non-biophysically realistic local circuit models. Nevertheless, evidence suggests that the neural correlates of conscious perception are distributed across the cortex. I present a large-scale, multi-regional, anatomically and physiologically constrained neural network model of the macaque cortex simulating Binocular Rivalry (BR)- a standard experimental paradigm for consciousness study. The model is based on a previously studied neural network model for working memory, which we have expanded to account for subjects’ behavior and neural activity in a detection task. BR emerges when dichoptically viewing dissimilar images induces perceptual alternations dissociating the sensory stimulation from the conscious content. In the multi-regional model, each area is represented by a local circuit capable to reproduce the observed neural activity within one cortical area. There are 30 areas of the macaque cortex, hierarchically organized from visual to prefrontal, and wired according to anatomical inter-areal connectivity data. The model reproduces the neural encoding of conscious content along the cortical hierarchy, consistent with electrophysiological recordings. The resulting neural dynamics results from a multi-area attractor rather than on locally generated ones. Furthermore, simulated lesions and targeted intracranial electrical stimulations test existing predictions of different theories.