Beyond bumps: Spiking networks that store sets of functions
Abstract: There are currently a number of models that use spiking neurons in recurrent networks to encode a stable Gaussian ‘bump’ of activation. These models successfully capture some behaviors of various neural systems (e.g., storing a single spatial location in working memory). However, they are limited to encoding single bumps of uniform height. We extend this previous work by showing how to construct and analyze realistic spiking networks that encode multiple ‘bumps’ of different heights. Our networks capture additional experimentally observed behavior (e.g., storing multiple spatial locations at the same time and the sensitivity of working memory to non-spatial parameters).
Loading