Precision of pulse-coupled networks of integrate-and-fire neurons

Abstract

Some sensory tasks in the nervous system require highly precise spike trains to be generated in the presence of intrinsic neuronal noise. Collective enhancement of precision (CEP) can occur when spike trains of many neurons are pooled together into a more precise population discharge. We study CEP in a network of N model neurons connected by recurrent excitation. Each neuron is driven by a periodic inhibitory spike train with independent jitter in the spike arrival time. The network discharge is characterized by [iopmath latex="$sigma_W$"] W [/iopmath], the dispersion in the spike times within one cycle, and [iopmath latex="$sigma_B$"] B [/iopmath], the jitter in the network-averaged spike time between cycles. In an uncoupled network [iopmath latex="$sigma_Bsim1/sqrt{N}$"] B ~ 1/(N)1/2 [/iopmath] and [iopmath latex="$sigma_W$"] W [/iopmath] is independent of N. In a strongly coupled network [iopmath latex="$sigma_Bsim 1/sqrt{log N}$"] B ~ 1/(log N)1/2 [/iopmath] and [iopmath latex="$sigma_W$"] W [/iopmath] is close to zero. At intermediate coupling strengths, [iopmath latex="$sigma_W$"] W [/iopmath] is reduced, while [iopmath latex="$sigma_B$"] B [/iopmath] remains close to its uncoupled value. The population discharge then has optimal biophysical properties compared with the uncoupled network.