Nonlinear stimulus representations in neural circuits with approximate excitatory-inhibitory balance

Abstract

Balanced excitation and inhibition is widely observed in cortical recordings. How does this balance shape neural computations and stimulus representations? This problem is often studied using computational models of neuronal networks in a dynamically balanced state. However, these balanced network models predict a linear relationship between stimuli and population responses, in contrast to the nonlinearity of cortical computations. We show that every balanced network architecture admits some stimuli that break the balanced state and these breaks in balance push the network into a "semi-balanced state" characterized by excess inhibition to some neurons, but an absence of excess excitation. The semi-balanced state is unavoidable in networks driven by multiple stimuli, consistent with experimental data, has a direct mathematical relationship to artificial neural networks, and permits nonlinear stimulus representations and nonlinear computations.