Dynamical properties of self-sustained and driven neural networks

Abstract

In the awake brain, cerebral cortex displays asynchronous-irregular (AI) states, where neurons fire irregularly and with low correlation. Neural networks can display AI states that are self-sustained through recurrent connections, or in some cases, need an external input to sustain activity. In this paper, we aim at comparing these two dynamics and their consequences on responsiveness. We first show that the first Lyapunov exponent (FLE) can differ between self-sustained and driven networks, the former displaying a higher FLE than the late. Next, we show that this impact the dynamics of the system, leading to a tendency for self-sustained networks to be more responsive, both properties that can also be captured by mean-field models. We conclude that there is a dynamical and excitability difference between the two types of networks besides their apparent similar collective firing. The model predicts that calculating FLE from population activities in experimental data could provide a way to identify if real neural networks are self-sustained or driven.