Abstract

In what dynamical regime do cortical circuit operate? What are the main building blocks of collective neuronal dynamics that enable the generation of behaviour? Classical models of cortical dynamics account for the structure of spontaneous activity rather well, but fail to explain transient, behaviour-related activity. In particular, the high spatial and temporal complexity of activity recorded in the motor cortex during limb movements has long presented a significant challenge to modelers. Here, I will describe a new class of models in which strong and complex recurrent excitation is stabilized by detailed feedback inhibition. In contrast with previous models, ours are able to transiently and selectively amplify certain network states, which can be reached through appropriate external stimulation during movement planning. Following a go cue, the input is withdrawn and the network’s intrinsic dynamics then produces single-neuron and collective activity in good agreement with empirical observations. These transients can further be decoded into complex movements. I will show that the complex dynamics of our networks can be understood from a simple building block, already known as “balanced amplification” in the context of primary visual cortex dynamics. In this dynamical regime, excitation and inhibition are as tightly balanced as recently reported in experiments across several brain areas, suggesting inhibitory control of complex excitatory interactions as a generic principle of cortical dynamics.