Item

Abstract

Criticality is considered an attractive candidate state for the dynamics and function of the brain, because in models criticality maximizes a number of properties related to information transmission and computations. These include the dynamic range, the susceptibility, the sensitivity to input, the correlation length, and the pattern diversity. And indeed, numerous studies accumulated evidence that supports the criticality hypothesis. However, some observations are also contradictory. The latter might in part be explained by the fact that criticality is a “full system” property, whereas experimental neural recordings can only assess a tiny part of the full network (“subsampling problem”). Here, we first recapitulate the basic properties of a dynamical system at and around the critical point in a pedagogic manner. We show how subsampling can bias inference about the underling network dynamics, and then present two recent analytical approaches to overcome the subsampling bias, based either on neural avalanches, or on the time-series of neural activity proper. The novel approach typically allows quantify the distance to criticality from less then ten recorded neurons and a few minutes of recording only. Thereby, it offers a novel and very powerful approach to assess criticality and task-related deviations thereof.