New hallmarks of criticality in recurrent neural networks

TL;DR

This paper demonstrates through a computational model that cortical neural networks operating near criticality exhibit key features like irregular spiking and variability decline, linking criticality to neural variability and network behavior.

Contribution

It reveals that criticality in recurrent neural networks explains irregular spiking and variability patterns, providing a unifying principle for cortical neural dynamics.

Findings

Irregular spiking emerges near criticality.

Response variability decline at stimulus onset is linked to criticality.

Spiking irregularity is maximized at the network's critical point.

Abstract

A rigorous understanding of brain dynamics and function requires a conceptual bridge between multiple levels of organization, including neural spiking and network-level population activity. Mounting evidence suggests that neural networks of cerebral cortex operate at criticality. How operating near this network state impacts the variability of neuronal spiking is largely unknown. Here we show in a computational model that two prevalent features of cortical single-neuron activity, irregular spiking and the decline of response variability at stimulus onset, are both emergent properties of a recurrent network operating near criticality. Importantly, our work reveals that the relation between the irregularity of spiking and the number of input connections to a neuron, i.e., the in-degree, is maximized at criticality. Our findings establish criticality as a unifying principle for the variability of single-neuron spiking and the collective behavior of recurrent circuits in cerebral cortex.