In theoretical neuroscience, because of the limited available single neuron data, neural activity is usually modeled considering one local network representing a single brain area. While much has been learnt in this way, many tasks activate a network of areas and, by neglecting inter-areal interactions, whose effects are increasingly acknowledged and partially taken into account is some models, the limits of a local modeling approach are more and more apparent. As an example, recordings in several brain areas -at least seven- of a monkey engaged in a perceptual discrimination task have revealed task-induced neural activation in all of them, and that neurons coding for the stimulus, working memory and the decision are found in most of the recorded areas . Despite a number of theoretical attempts, mostly at the local level, an understanding of the organization of such neural activity is still lacking. Because of the dense number of inter-areal (white-matter) connections, a distributed task-induced neural activation suggests that a model at the brain scale level is necessary. In this context, the first question to address is how spontaneous neural activity organizes at the brain level in the absence of a task, and how the stimulus-induced activity in sensory areas propagates on the network during a task.
In the last decades, studies of spontaneous brain activity, as measured by neuroimaging techniques (fMRI, MEG) during rest, have revealed a specific organization of the slow temporal fluctuations of these signals in a set of so-called resting state networks. Recently, models of the brain spontaneous activity considering local networks of excitatory and inhibitory neurons connected through the network defined by inter-areal connections have been proposed. Assuming that local networks are in an asynchronous low activity state during spontaneous activity, for the emerging global state to best fit the experimental data, it was found that local networks must be relatively strongly coupled and in a balanced state , this state being structurally imposed here by an adequate choice of the connection weights. However, and like in alternative models, stimulus-induced activation is found to propagate only weakly on the global network.
To explain a number of experimental observations in V1 in a unified way, a recent study has proposed that the local excitatory network is intrinsically unstable but stabilized by feedback inhibition . Using such local networks in a large-scale brain model, we show here that, beyond fitting the organization of spontaneous activity, stimulus-induced activation propagates on the global network. Furthermore, as a local network goes towards a dynamically balanced state for sufficiently strong input, this state is also observed at the global network level.
 R. Romo, V. de Lafuente Progr. Neurobiol. 103:41-75 (2013).
 G. Deco, A. Ponce-Alvarez, P. Hagmann, G.L. Romani, D. Mantini, M. Corbetta J. Neurosci. 34:7886-7898 (2014).
 A. Ahmadian, D.B. Rubin, K.D. Miller Neural Comput. 25:1994-2037 (2013).