Can a competitive neural network explain selective attention in insect target tracking neurons?

Abstract

Small target motion detecting (STMD) neurons in the dragonfly brain are neural correlates of a highly-specialized and ethologically-significant feature detection function, and the recent discovery of selective attention in STMDs has clear implications for the ability of dragonflies to track and pursue one target from among several. We used a biophysically-plausible neural network model, based on competitive units fed by NMDA-type synaptic inputs and including lateral feedback inhibition, to model these attentional effects in numerical simulations. With appropriate forward gain, the model displays a winner-takes-all behavior that partially captures the selective attention documented in electrophysiological recordings from STMDs. It cannot, however, explain the full range of results that have now been observed in wide-field STMDs, in particular a bias toward attention to targets dependent on their traversal of continuous trajectories.