Self-sustained activity in networks of gain-modulated neurons

Abstract Simulation studies have shown that recurrently connected neurons are capable of sustaining non-uniform profiles of activity in the absence of tuned input. These attractor networks are the basis for models of working memory and other processes where information about transient stimuli is stored temporarily. In addition, there is strong evidence that neurons often interact by affecting each other's gain. Here I study a minimal recurrent network that takes gain interactions into account. I show analytically that, in agreement with results of computer simulations, a center-surround organization gives rise to two types of stable solutions: a uniform state in which all neurons fire at the same rate, and a self-sustained profile of activity that may be centered at any point in the network. This theoretical framework based on nonlinear neuronal interactions is, in general, a powerful tool for investigating recurrent network dynamics.