Â鶹ӰԺ

Control of macroscopic neuronal network activity is optimal at criticality

There is growing evidence that the cortex operates at a critical state where the strength of excitatory and inhibitory neurons is precisely balanced. Recent research has shown that many properties related to information processing are optimized at criticality, including dynamic range, information transmission, and the variability of synchronization. By analyzing neuronal network models, we show here that, in addition, the macroscopic activity of the network is most easily controlled at criticality. We first analyze a simple binary neuron model, where nodes may be either active or quiescent at each time. The balance in the relative numbers and synapse strengths of excitatory and inhibitory nodes determines whether excitations are amplified, maintained, or decay on average (corresponding to supercritical, critical, and subcritical states, respectively). We consider the problem of controlling the total network activity to a given target value, using a feedback loop with either global or local information. We show numerically and theoretically that the control error, averaged across a range of targets, is minimized when the network is critical. We attempt to validate our findings for the binary neuron model by simulating a conductance-based neuron model. However we find that for the conductance-based model, the question of tuning activity becomes much more complicated.