Rubén Moreno-Bote (1), John Rinzel (1,2) and Nava Rubin (1)

 
   (1) Center for Neural Science, New York University, New York, NY 10003-6621, USA.

 
   (2) Courant Institute of Mathematical Sciences, New York University, New York.

 
Abstract

 
  When a stimulus supports more than one possible interpretation, perception alternates in a haphazard manner between them.  What causes the bi-stable perceptual switches remains an open question. We develop a new, attractor-based framework in which alternations are induced by noise, and are absent without it. Our framework starts with an energy minimization model and then realized with rate-based and spiking network models, providing different levels of description of perceptual bi-stability. The model behavior is compared with experimental results from binocular rivalry, the most extensively studied bi-stable phenomenon. The model reproduces two long-standing experimental results, Levelt (1968) propositions II and IV. It also predicts a reduction of activity during rivalry compared to non rivaling vision, consistent with recent neuroimaging findings. The model can be implemented in an architecture that includes inhibitory populations locally connected to the competing excitatory populations and driven globally by an excitatory pool. This architecture readily generalizes to several competing populations, providing a natural extension for multi-stability phenomena.