Prediction of linear and non-linear responses of MGB neurons by system identification methods.

In sensory physiology, various System Identification methods are implemented to formalize stimulus-response relationships. We applied the Volterra approach for characterizing input-output relationships of cells in the medial geniculate body (MGB) of an awake squirrel monkey. Intraspecific communication calls comprised the inputs and the corresponding cellular evoked responses--the outputs. A set of vocalization was used to calculate the kernels of the transformation, and these kernels subserved to predict the responses of the cell to a different set of vocalizations. It was found that it is possible to predict the response (PSTH) of MGB cells to natural vocalizations, based on envelopes of the spectral components of the vocalization. Some of the responses could be predicted by assuming a linear transformation function, whereas other responses could be predicted by non-linear (second order) kernels. These two modes of transformation, which are also reflected by a distinct spatial distribution of the linear vis-à-vis non-linear responding cells, apparently represent a new revelation of parallel processing of auditory information.

Identification of MGB cells by volterra kernels. III. A glance into the black box.

Neuronal systems can be described by their transfer functions, which can be represented by a Volterra series expansion. While the high level of abstraction which characterizes this representation enables a global description, it is problematic, to some extent, in the context of linking the formal representation of the system to its actual structure. The formal representation is unique, yet there are multiple physical realizations of this representation. Separating the system's output into its logical components (linear, cross-linear, and self nonlinear, in this study), and inspecting the relative contribution of these components, might provide a key towards a linkage between the formal and actual representations. Based on results drawn from identification of MGB cells of the squirrel monkey, it is shown that the relative contributions can be described in neurobiological terms such as excitation and inhibition and thus be attributed to actual subsystems.

Identification of MGB cells by Volterra kernels. I. Prediction of responses to species specific vocalizations.

The function of CNS sites is frequently explored by an analysis of its input-output relationships. However, such research are often confined to a qualitative and subjective inspection of raw data. System Identification methods can be used to formalize the stimulus--response relations, and one of them, the Volterra approach, is employed here in order to define these relations in the MGB of the squirrel monkey, natural vocalizations being the stimuli. In order to validate the formal representation of the system under study, the predictability power of the model is tested. Having the distances between responses (PSTH) and predicted response quantified, it is found that the predictions made by the model are, in general, "closer" to the actual responses then some arbitrarily chosen responses. It is concluded that there are cells in the MGB that can be characterized by their Volterra kernels, and further research on the cell's functional role can be based on these kernels.

Analysis of recall and recognition in a certain class of adaptive networks.

A general model of adaptive networks which perform recall is analyzed in view of qualitative psychological findings. The assumptions underlying the basic model are few and general in the sense that no specifications of structure or mechanisms of adaptation are imposed. The analysis of the model is towards the addition of various features drawn from the global input-output relations expected from the networks. The memorizing process of recognition is found to be intrinsic to the model, and four of the most prominent relations between the performance of recall and recognition are shown to either exist in model or to be realizable by means of few additional plausible features.