The EM and the SEM algorithms for mixtures: Statistical and numerical aspects

This paper is devoted to the study of the statistical properties of two efficient algorithms for the mixture problem under the maximum likelihood approach: the EM algorithm and his probabilistic teacher version, the SEM algorithm. The authors show that in general the SEM algorithm performs better. In particular, they show that the SEM algorithm provides a rough estimation of the parameters standard deviations in a very competitive time compared with the expensive time with the Bootstrap estimates of standard deviations via the EM algorithm.