Despite extensive ongoing research, there is no general agreement about how information is transmitted and processed in biological neural networks. Many neurobiological experiments suggest that there is a strong correlation between the rate of action potential production and the stimulus strength, at least in early sensory areas. In addition, one also finds changing temporal correlations between spike trains of neurons by varying stimuli [1,2,3]. Thus, there are indications that the average frequency of spike trains as well as their temporal correlations carry important information within the neural systems. In this paper, these two coding concepts are combined and attributed to two complementary modes of network operations: one mode utilizing the firing rates of neurons for continuous representation and calculation of stimulus properties, the other mode relying on the temporal timing of spike trains to construct and mark stable percepts by a dynamical process of coherence-detection.
As a worked-out example, a network composed of rate- and spike-coding neurons is presented. The network performs disparity calculations between two stereo images, solving in effect the task of depth perception. As shown here, both operational modes are necessary for the successful operation of the network.