Fig. 6: Randomly shuffling recurrent connections eliminates nTWs and the ability to forecast.
From: Waves traveling over a map of visual space can ignite short-term predictions of sensory input
a Left: the topographic network model used throughout this study, featuring feedforward projections of the image input (red lines) and local distance-dependent horizontal connectivity (blue lines). There are also synaptic time delays proportional to a node pair’s separation distance within the horizontal recurrent circuitry. Right: by randomizing the horizontal connection weights and time delays, the topography in the network is removed. b Closed-loop forecasts generated by the topographic network. c The network activity of the topographic network in response to frames of a natural movie input. d Network activity of the shuffled network. e Closed-loop forecasts generated by the shuffled network.
