A second type of criticality in the brain uncovers rich multiple-neuron dynamics

Experimental evidence for cortical networks operating in the balanced state is overwhelming [1,2,3]. In this state, strong recurrent inhibition yields almost vanishing correlations in the input to neurons [4,5]. The balanced state, however, only restricts average correlations in the network due to the large convergence of connections. Here we show that balanced networks can show a rich correlation structure between individual neurons that is explained by the effective connectivity of the network [6]. The latter is determined by the anatomical connections and the sensitivity of neurons to inputs. Large heterogeneity in effective connections causes nearly unstable linearized dynamics in various directions of the high-dimensional space of all neurons, leading to multiple-neuron responses with largely different time courses [7]. As a consequence distributions of correlations across neurons become broad, but approximately centered around zero. A large dispersion of correlations, as for example obtained from recordings in macaque motor cortex, can therefore be used as an indicator of a rich dynamical repertoire, which is hidden from macroscopic brain signals, but essential for high performance in such concepts as reservoir computing [8,9]