Brent Doiron, University of Pittsburgh

The space and time of cortical shared variability

A characteristic feature of cortical activity is its sizable dynamic and trial-to-trial variability. A popular framework to study the mechanics of variability is one where networks of neurons show a balance between large recurrent excitation and inhibition. Classic treatments of balanced networks establish an asynchronous state, where network-wide shared variability is vanishingly small. This is at odds with population recordings from cortex that clearly establish that shared variability is widespread and is often manipulated by cortical state. We explore balanced networks with a spatial structure in connectivity, with nearby neuron pairs wire together with a higher probability that more distant pairs. We derive key constraints for the asynchronous solution, and show that when the spatial spread of recurrent wiring is broader than that of feedforward projections a solution with correlated spiking activity is uncovered. Further, we explore how the timescales of inhibition versus excitation also control population-wide variability and show a low dimensional variability is possible with slow inhibition. In all case we compare of theoretical predictions with population recordings in the visual system of non-human primates.  

Organized by

Benjamin Lindner