Robust encoding by collective bursting in biologically plausible neural networks

Abstract

We describe a novel type of bursting that we observe in simulations of large recurrent networks of biophysically plausible, intrinsically non-bursting neurons. The mechanism responsible for the bursting is a combination of excitatory feedback received from neighbouring neurons, together with an activity-dependent adaptation mechanism that slows down spiking. This collective bursting is shown to encode external inputs in the intervals between bursts. The interspike intervals during each burst are irregular and have a high output rate that is insensitive to the input strength. The encoding is reliable and precise, even when individual neurons have imperfect, varying properties and is robust to failure of large numbers of neurons.