A Neural Model for a Randomized Frequency-Spatial Transformation

We examine a random neural network model of the cortex, composed of neurons having short membrane time constants and stochastic dynamics. We show that such limited memory resources suffice for a frequency-spatial transformation (FST): Depending on the frequency of the input signal, different neural assemblies generate sustained cortical activity that persists after the input stimuli is removed. These assemblies may be only indirectly connected to the input region via the cortical mesh of connections. The FST scheme proposed demonstrates that random neural networks may respond specifically to different input stimuli.