Innovating Signal Processing for Spike Train Data

It is well known that the conventional algorithms of optimal signal processing developed for random processes can not be easily applied to the analysis and quantification of spike trains. This talk will present our efforts to derive a reproducing kernel Hilbert space (RKHS) for spike train analysis. The advantage of a RKHS is that it has a linear structure and therefore the conventional techniques of principal component analysis, optimal filtering, classification and clustering can be readily applied. We will briefly present the methodology and show some preliminary examples with synthetic and real spike data.