Jochen Braun: Clickety-clack: a non-equilibrium model of cortical activity performs optimal inference

Universität Magdeburg und Kingshuk Ghosh, Denver U, USA

Current theories of perceptual inference in the brain take an equilibrium viewpoint, in that neural activity categorizes a given sensory scene by reaching an energy minimum in a static multi-dimensional landscape. Here we propose an alternative, non-equilibrium framework where cortical activity reaches decisions by forming unstable dynamic states, which become immediately subject to revision. We see updating of decisions as an inherent activity of the brain that occurs even without new sensory information. Spontaneous stochastic fluctuations of activity (“shared variability” or “noise correlations”) play a central role in this. Building on perceptual evidence for a far-from-equilibrium birth-death process [1], we present a high-level model of cortical activity implemented with known features of primate neurophysiology and neuroanatomy. Linking all three levels of Marr, our model suggests that i) cortical decision-making relies on non-equilibrium dynamics and non-static energy landscapes, ii) the intrinsic stochasticity of neural activity is beneficial – not detrimental – for physiological function, iii) the decision-making performance of neural activity is maximized by a non-equilibrium variational principle (“maximum caliber”) [2].

[1] Cao et al. (2021) Binocular rivalry reveals an out-of-equilibrium neural dynamics suited for decision-making. eLife 10; e61581, 1–42.

[2] Ghosh et al. (2022) The maximum caliber variational principle for non-equilibria. Annu. Rev. Phys. Chem.71:213-78.

 

Guests are welcome!

 

Organized by

Benjamin Lindner / Margret Franke