Adaptation and other dynamic effects on neural signal transfer

Abstract

In the light of the latest results concerning the dynamics of synaptic transmission (Markram et al., 1996) and spike frequency adaptation, the question how the signals are transferred between neurons and what are the meaningful signals in the neural information processing, has to be reconsidered. We constructed simple models of these phenomena and computed qualitatively the neural transfer properties. In these models we examined only the transmission of two parameters of the neurons' membrane current: the mean and the standard deviation. In the cerebral cortex there is usually significant convergence between neurons and so, if the inputs follow the Central Limit Theorem, these two parameters can fully describe the summed input of the cell. From these transfer properties we can conclude, that the big observed coefficient of variation (Softky and Koch, 1993) can be produced, at least partly, by these mechanisms. The low firing rate and spike frequency adaptation do not allow us to use simple firing rate code for longer period simulation of real neurons, as it is assumed in the majority of regular artificial neural network models (ANN) (Gerstner et al., 1992). The measured effects of long term potentiation on the dynamics of synaptic transmission show, that the transfer of membrane currents mean does not change appropriately. Namely, its modification can not be considered as synaptic weight change. Our result shows, that there could be an additional parameter, the standard deviation of the synaptic currents, which can provide the same sort of transfer properties as the average firing rate in typical ANN models. The formation and operation of this type of “code” is discussed.