We propose a united approach to maximum likelihood estimation, classification, and statistical learning in the context of finite mixture models, based on observations that can be considered a ...