Estimation of a mean template from spike-train data.

Computing the template, or the mean, of a set of spike trains is a novel and important task in neural coding. Due to the random nature of spike trains taken from experimental recordings, probabilistic and statistical methods have gained prominence in examining underlying firing patterns. However, these methods focus on modeling neural activity at each given time and therefore their results depend heavily on model assumptions. Taking a model-free and metric-based approach, we analyze the space of spike trains directly and reach algorithms for estimating statistical summaries, such as the mean spike train, of a given set. In our data-driven approach the mean is defined directly in a function space in which the spike trains are viewed as individual points. Here we develop an efficient and convergence-proven algorithm to compute the mean spike train in a general scenario. Experimental result from a neural recoding in primate motor cortex indicates that the estimated means successfully capture the typical patterns in spike trains. In addition, these mean spike trains provide an accurate and efficient performance in decoding motor behaviors.