Spike classification with multivariate t-distribution mixture model via improved Expectation-Maximization algorithm

Recent research has developed various methods in automatic spike classification, including Expectation-Maximization (EM) clustering based on multivariate t-distribution mixture models. In our study, we improved the EM iterative algorithm with a significantly better ascent gradient in the high-dimensional feature space of spikes. Our simulations showed that this improvement of the EM algorithm could reduce the computation time with no significant change in classification error. Applications of this new algorithm yielded better computation cost and a more robust performance in real experimental spike data analysis.