Acceleration of the EM algorithm using the vector epsilon algorithm (Q626226)

A technique is described for the acceleration of any EM-algorithm by the use of the vector \(\varepsilon\) algorithm scheme. It is shown that the accelerated iterations converge to the fixed point more quickly (in the Avram sense) than the initial EM iterations. Performances of the vector \(\varepsilon\) acceleration, Louis acceleration and initial EM algorithms are compared on simulation data for contingency tables with partially classified observations and incomplete bivariate normal data.