On the maximization of likelihoods belonging to the exponential family using ideas related to the Levenberg-Marquardt approach