A second-order approximation to the log-likelihood surface for mixture models, with application to the EM algorithm.

This article considers a of recent origin approximation to the log-likelihood surface in mixture archetypes This approximation is based upon both the mean and variance of the full-data log likelihood above imputations of assignments of observations to composings This approximation is accurate to next to the first order, and holds for general missing data question s The approximation provides a fresh method for calculating the observ information using the EM algorithm, and motivates a Gauss-Newton manner for finding the MLE. This Gauss-Newton means is implemented together with the ideas behind the SAGE algorithm. The resulting algorithm outperforms the EM CEMM and a further Gauss-Newton algorithm when