NONLINEAR PRINCIPAL COMPONENTS-ANALYSIS OF NEURONAL SPIKE TRAIN DATA

Many recent approaches to decoding neural spike trains depend critical ly on the assumption that for low-pass filtered spike trains, the temp oral structure is optimally represented by a small number of linear pr ojections onto the data, We therefore tested this assumption of linear ity by comparing a linear factor analysis technique (principal compone nts analysis) with a nonlinear neural network based method. It is firs t shown that the nonlinear technique can reliably identify a neuronall y plausible nonlinearity in synthetic spike trains. However, when appl ied to the outputs from primary visual cortical neurons, this method s hows no evidence for significant temporal nonlinearities. The implicat ions of this are discussed.