Stability of neural network activity regimes achieved by the balance between excitatory and inhibitory synaptic scaling

Neural circuits, such as the cerebral cortex, display a resting basal activity that never ceases. In order to better understand the functioning of these systems, and how they may process information, it takes to understand what is needed for the maintenance of an uncorrelated basal activity. This is the so-called background state, in which an input representation may be embedded while avoiding pathological regimes such as network-wide synchronization. One of the processes that is considered to be critical for this is the experimentally observed plasticity mechanism of homeostatic synaptic scaling. This phenomenon consists on the reaction by the network upon a sudden activity disturbance, in which its synaptic weights are adjusted so that a set spiking frequency can be regained. In neurons of the visual cortex, it is seen that the amplitude of excitatory signals is enhanced if the activity is damped, and decreased if the activity is augmented. The inhibitory connections follow the opposite direction of change. While the effects of the plasticity on excitatory connections have been recently studied, the concurrent inhibitory plasticity and their results over the network balance, as well as the following interferences with other long-term plasticity mechanisms, are not clear. Our work is based on the analysis of a mean-field model adapted to account for the synaptic scaling and the comparison of its results with those of detailed numerical simulations of integrate-and-fire neural networks to assert the stability of the basal activity state. We find necessary relationships between macroscopic parameters, such as the ratio between the recurrent and feedforward connections according to the frequency to which the network is expected to converge in order to keep the activity from diverging.