Avalanche scaling in large neural populations with distributed coupling to multiple dynamical latent variables

• Type of Material: Article
• Authors: Mia Morrell, New York University; Ilya Nemenman, Emory University; Audrey J Sederberg, University of Minnesota
• Language: English
• Date: 2023-01-02
• Publisher: NIH
• Publication Version: Preprint (Prior to Peer Review)
• Copyright Statement: © 2023 Cornell University
• License: Creative Commons Attribution 4.0 International
• Title of Journal or Parent Work: arXiv
• Volume: 2023
• Grant/Funding Information: AS was supported in part by NIH grant 1RF1MH130413-01 and by startup funds from the University of Minnesota Medical School.; IN was supported in part by the Simons Foundation Investigator program, the Simons-Emory Consortium on Motor Control, NSF grant BCS/1822677 and NIH grant 2R01NS084844.
• Keywords: neural activity data

Abstract

Observations of power laws in neural activity data have raised the intriguing notion that brains may operate in a critical state. One example of this critical state is “avalanche criticality,” which has been observed in a range of systems, including cultured neurons, zebrafish, and human EEG. More recently, power laws have also been observed in neural populations in the mouse under a coarse-graining procedure, and they were explained as a consequence of the neural activity being coupled to multiple latent dynamical variables. An intriguing possibility is that avalanche criticality emerges due to a similar mechanism. Here, we determine the conditions under which dynamical latent variables give rise to avalanche criticality. We find that a single, quasi-static latent variable can generate critical avalanches, but that multiple latent variables lead to critical behavior in a broader parameter range. We identify two regimes of avalanches, both of which are critical, but differ in the amount of information carried about the latent variable. Our results suggest that avalanche criticality arises in neural systems in which there is an emergent dynamical variable or shared inputs creating an effective latent dynamical variable, and when this variable can be inferred from the population activity.