Simulation Studies of the Speed of Recurrent Processing

Abstract

The speed of processing in the cortex can be fast. For exam- ple, the latency of neuronal responses in the visual system increases by only approximately 10-20 ms per area in the ventral pathway sequence V1 to V2 to V4 to Inferior Temporal visual cortex. Since individual neu- rons can be regarded as relatively slow computing elements, this may imply that such rapid processing can only be based on the feedforward connections across cortical areas. In this paper, we study this problem by using computer simulations of networks of spiking neurons. We eval- uate the speed with which different architectures, namely feed-forward and recurrent architectures, retrieve information stored in the synaptic efficacy. Through the implementation of continuous dynamics, we found that recurrent processing can take as little as 10-15 ms per layer. This is much faster than obtained with simpler models of cortical processing that are based on simultaneous updating of the firing rate of the indi- vidual units. These findings suggest that cortical information processing can be very fast even when local recurrent circuits are critically involved.