On Stochastic Versions of the EM Algorithm

We compare three different stochastic versions of the EM algorithm: The SEM algorithm, the SAEM algorithm and the MCEM algorithm. We suggest that the most relevant contribution of the MCEM methodology is what we call the simulated annealing MCEM algorithm, which turns out to be very close to SAEM. We focus particularly on the mixture of distributions problem. In this context, we review the available theoretical results on the convergence of these algorithms and on the behavior of SEM as the sample size tends to infinity. The second part is devoted to intensive Monte Carlo numerical simulations and a real data study. We show that, for some particular mixture situations, the SEM algorithm is almost always preferable to the EM and simulated annealing versions SAEM and MCEM. For some very intricate mixtures, however, none of these algorithms can be confidently used. Then, SEM can be used as an efficient data exploratory tool for locating significant maxima of the likelihood function. In the real data case, we show that the SEM stationary distribution provides a contrasted view of the loglikelihood by emphasizing sensible maxima. Abstract. We compare three diierent stochastic versions of the EM algorithm: The SEM algorithm, the SAEM algorithm and the MCEM algorithm. We suggest that the most relevant contribution of the MCEM methodology is what we call the simulated annealing MCEM algorithm, which turns out to be very close to SAEM. We focus particularly on the mixture of distributions problem. In this context, we review the available theoretical results on the convergence of these algorithms and on the behavior of SEM as the sample size tends to innnity. The second part is devoted to intensive Monte Carlo numerical simulations and a real data study. We show that, for some particular mixture situations, the SEM algorithm is almost always preferable to the EM and simulated annealing versions SAEM and MCEM. For some very intricate mixtures, however, none of these algorithms can be conndently used. Then, SEM can be used as an eecient data exploratory tool for locating signiicant maxima of the likelihood function. In the real data case, we show that the SEM stationary distribution provides a contrasted view of the loglikelihood by emphasizing sensible maxima.