The Expectation Maximization algorithm is an iterative method for computing approximate solutions to Maximum Likelihood (ML) parameter estimation problems (dempster). The EM algorithm is commonly used when the observations can be viewed as incomplete data, and the true ML estimates of the unknown parameters cannot be obtained in closed form. Under certain mild conditions (chengupta), the estimates of the unknown parameters obtained from successive iterations of the EM algorithm can be proved to converge to the parameter estimates corresponding to a local maximum of the likelihood function. Though EM algorithm is guaranteed to converge only to a local maximum of the likelihood function, in most of the practical scenarios,  the parameter estimates computed by using EM algorithm often serve as good approximaions to the true ML estimates.