Non-perturbative renormalization group analysis of nonlinear spiking networks

23 Jan 2023  ·  Braden A. W. Brinkman ·

The critical brain hypothesis posits that neural circuits may operate close to critical points of a phase transition, which has been argued to have functional benefits for neural computation. Theoretical and computational studies arguing for or against criticality in neural dynamics have largely relied on establishing power laws in neural data, while a proper understanding of critical phenomena requires a renormalization group (RG) analysis. However, neural activity is typically non-Gaussian, nonlinear, and non-local, rendering models that capture all of these features difficult to study using standard statistical physics techniques. We overcome these issues by adapting the non-perturbative renormalization group (NPRG) to work on network models of stochastic spiking neurons. Within a ``local potential approximation,'' we are able to calculate non-universal quantities such as the effective firing rate nonlinearity of the network, allowing improved quantitative estimates of network statistics. We also derive the dimensionless flow equation that admits universal critical points in the renormalization group flow of the model, and identify two important types of critical points: in networks with an absorbing state there is a fixed point corresponding to a non-equilibrium phase transition between sustained activity and extinction of activity, and in spontaneously active networks there is a physically meaningful \emph{complex valued} critical point, corresponding to a discontinuous transition between high and low firing rate states. Our analysis suggests these fixed points are related to two well-known universality classes, the non-equilibrium directed percolation class, and the kinetic Ising model with explicitly broken symmetry, respectively.

PDF Abstract
No code implementations yet. Submit your code now

Tasks


Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods


No methods listed for this paper. Add relevant methods here