Collective Activity of Many Bistable Assemblies Reproduces Characteristic Dynamics of Multistable Perception.

Area of Science:

• Neuroscience
• Perception
• Computational Modeling

Background:

• Perceptual decisions involve evidence accumulation influenced by noise.
• Multistable perception, characterized by spontaneous reversals, exhibits consistent statistical properties.
• Understanding these statistics constrains models of neural representations.

Purpose of the Study:

• To identify a stochastic process that explains the scaling properties of reversal times in multistable perception.
• To test whether a generalized Ehrenfest process can quantitatively reproduce observed perceptual dynamics.
• To link these dynamics to plausible neural substrates.

Main Methods:

• Treated reversal timing as a first-passage time problem.
• Modeled collective activity of a finite population of bistable units (generalized Ehrenfest process).
• Compared the generalized Ehrenfest process with Poisson, Wiener, and Ornstein-Uhlenbeck processes.

Main Results:

• The generalized Ehrenfest process quantitatively reproduced the characteristic scaling property of reversal times.
• This model explained the relationship between mean reversal times and their variance/skewness.
• Other considered processes (Poisson, Wiener, Ornstein-Uhlenbeck) failed to reproduce these scaling properties.

Conclusions:

• Perceptual representations are likely granular, probabilistic, and operate far from equilibrium.
• A finite population of bistable neural assemblies, like cortical columns, can underlie multistable perception.
• These findings support a model of perception as a statistical inference mechanism.