Abstract. 
When observers view for an extended time an ambiguous visual scene with two
or more different interpretations they report switching between different
perceptions. Binocular rivalry is a well-known example of perceptual
bistability (two perceptions), for which different mechanistic models have
been proposed. However, attempts to generalize these models to stimuli with
more than two competing percepts are scarce in the literature.

In this talk, we will focus on experiments with two superimposed drifting
gratings (plaids), identical except for their movement directions (normal to
their respective orientations).  There are two sources of ambiguity: their
relative motion and the depth order.  If the angle between the normal
vectors for motion is small, coherence is favored (the gratings move
together as a single pattern).  If the angle is large, transparency is
favored (the gratings slide across one another).  For some range of angles,
tristable perception is experienced: a coherent percept  and two
transparent percepts with alternating depth ordering.

We will present a firing rate-based neural network model for the dynamics
of alternations during perceptual tristability for plaids. The model
involves stochastic dynamics of multiple-attractor systems. We have tested
two different architectures: (a) three populations compete at the same level
(b) two pairs of populations compete hierarchically.  We will discuss the
different results the two architectures yield to.

This is joint work with John Rinzel (NYU)