Multi-packet regions in stabilized continuous attractor networks

Continuous attractor neural networks are recurrent networks with center-surround interaction profiles that are common ingredients in many neuroscientific models. We study realizations of multiple non-equidistant activity packets in this model. These states are not stable without further stabilizing mechanisms, but we show they can exist for long periods. While these states must be avoided in winner-take-all applications, they demonstrate that multiple working memories can be sustained in a model with global inhibition.