Bistable Perception Modeled as Competing

We propose stochastic integration at two neural levels as a model for bistable perception. In this model, two sets of meta-stable populations are driven by visual input, while two further sets are driven by the phenomenal percept. A perceptual reversal occurs whenever the activity associated with one percept exceeds a threshold. Perceptual alternations result from the continuous repetition of this race to threshold. Our model accounts for several hitherto puzzling aspects of bistable perception: the wide range of alternation rates observed under different conditions, the highly consistent statistics, the perceptual stabilization with interrupted displays, and the history-dependence of phenomenal appearance. It also predicts details of the dynamics of bistable perception that have so far not been examined. We conclude that bistable perception reflects the collective nature of neural decision making, rather than specific biophysical properties of individual neurons.