To estimate the missing data values directly.

To update the parameter estimates by maximizing the expected log-likelihood found in the E-step.

To calculate the posterior probability of the observed data.

To determine the number of iterations required for convergence.