| |
| Recently, several methods have been developed for automatic spike classification, including the Expectation Maximization (EM) clustering based on multivariate t-distribution mixture models. However, studies showed that EM algorithm has linearity convergence, and in terms of spike classification, this method could not be used practically for the large time expending. In this study, we introduced an optimized EM algorithm for spike classification with multivariate t-distribution finite mixture models. Our algorithm optimizes the EM iterative algorithm with classic ascent gradient in high dimension characteristic space of spikes, thus the convergence becomes better than its original linearity. The application of our optimized EM algorithm in real spike data showed a faster and more robust performance in the convergence, and better capability of the spike classification. |
|
| Keywords:Spike Classification, EM Algorithm, Ascent Gradient |
|
