Three levels of lateral inhibition: A space-time study of the retina of the tiger salamander.

The space-time patterns of activity generated across arrays of retinal neurons can provide a sensitive measurement of the effects of neural interactions underlying retinal activity. We measured the excitatory and inhibitory components associated with these patterns at each cellular level in the retina and further dissected inhibitory components pharmacologically. Using perforated and loose patch recording, we measured the voltages, currents, or spiking at 91 lateral positions covering approximately 2 mm in response to a flashed 300-microm-wide bar. First, we showed how the effect of well known lateral inhibition at the outer retina, mediated by horizontal cells, evolved in time to compress the spatial representation of the stimulus bar at ON and OFF bipolar cell bodies as well as horizontal cells. Second, we showed, for the first time, how GABA(C) receptor mediated amacrine cell feedback to bipolar terminals compresses the spatial representation of the stimulus bar at ON bipolar terminals over time. Third, we showed that a third spatiotemporal compression exists at the ganglion cell layer that is mediated by feedforward amacrine cells via GABA(A) receptors. These three inhibitory mechanisms, via three different receptor types, appear to compensate for the effects of lateral diffusion of activity attributable to dendritic spread and electrical coupling between retinal neurons. As a consequence, the width of the final representation at the ganglion cell level approximates the dimensions of the original stimulus bar.

Effects of the synaptic transmission's dynamics on possible neural codes.

To examine the effects of paired pulse facilitation, long-term synaptic modifications as well as spike frequency adaptation on neural signal transmission, a simple model was applied. This way various input-output properties of the model units were described. Particularly, the transmission of the mean and S.D. of the simulated synaptic currents were studied. The results indicate that the transfer of the mean value of the membrane currents cannot be described in terms of synaptic weights. So firing rate can hardly be an efficient neural code, especially for adaptive channels of the central nervous system (CNS). On the contrary, the transfer of S.D. of synaptic currents behaves in accordance with the synaptic weights. So it is supported that S.D. of the synaptic currents is a biological relevant subclass of the variation codes [see Perkel, H., Bullock, T.H., 1968. Neurol coding. Neurosci. Res. Program Bull. 6, 221-344]. It is discussed how this code can be established and how it works.

Deviation code is a prospective candidate of the communication between adapting neurons.

The low Firing Rate (FR) of pyramidal cells (Thorpe et al., 1989, Pettigrew et al., 1968) prevents the use of rate coding. The Deviation Code (DC), in which the deviation from the mean is the signal, is a candidate code that may be used in neurobiological and formal neural networks. The neurons which use this code could have similar properties as the model neurons frequently used by the neural network (NN) theories. The biological verification of this type of coding is required. This coding might be relevant in the physiological investigation of learning rules and can help to understand the organization of different nuclei of the nervous system.

Increase of storage capacity of neural networks by preprocessing using convergence and divergence.

The greatest practical limitation of the associative memory models, especially the Hopfield model is the low storage capacity. It has been shown by Gardner, that the Hopfield type models storage limit is 2*N, where N is the number of the processing elements or neurons. For biased patterns, on the other hand, it is much greater. But in general the input patterns are not biased. To approach to this problem and to increase the storage capacity of the model, the input patterns have to be diluted by some conversion method particularly which uses convergence and divergence in neuroanatomical sense. Based on this model these parameters can be estimated. As a consequence of this bias and the divergence, the storage capacity is increased. This preprocessing method doesn't lead to the loss of information and keeps the error correcting ability of the model.