Compressed Sampling for Spike Sorting

Our goal is to demonstrate the usefulness and efficiency of a compressed sensing approach when applied to the problem of spike sorting. This approach involves four steps towards transforming the raw spikes into a sparse basis with the fewest number of measurements require to achieve accurate clustering. We first train a dictionary which converts a spike into a sparse signal. We then sample the signal and reconstruct it into a sparse domain using a standard compressive sensing approach. Finally, we cluster the sparse signal and compare it to our truth data. We demonstrate that we are able to achieve very high cluster purity with our approach, and can do so with a minimal number of samples required from the original signal.