This research will develop and evaluate nonlinear methods of analyzing information content in spike activity recorded from ensembles of hippocampus neurons in behaving animals. Existing liner algorithms (discriminant analysis) have been used to extract behaviorally-significant information encoded among populations of neurons when firing is time-locked to specific task relevant events of a short-term memory (DNMS) task (Deadwyler et al., 1966). Nonlinear methods have the advantage of allowing analyses of neuron activity independent of other events in order to detect patterns missed by linear analyses. Detection of neural codes embedded within ongoing ensemble activity not necessarily synchronized to external events will provide a means of """"""""predicting"""""""" behavioral responses prior to when those identified linear patterns emerge. The research will adapt Volterra representations of nonlinear systems and draw upon nonparametric optimizations currently in the systems theoretic literature. A computationally-efficient neural network equivalent of Volterra-class systems (three-layer perceptron) will be adopted for computation of high-order kernel representations of neural firing patterns. Analyses will utilize existing as well as new data recorded from microelectrode arrays implanted in highly trained rats and will initially be evaluated relative to linear discriminant analyses previously published by this laboratory (Deadwyler and Hampson, 1997). The applicant will develop and characterize specific algorithms, and neural network design, and a generalizable nonlinear approach for extracting behaviorally-significant information from populations of hippocampal neurons.
