A mixture model has a discrete and unobservable variable (i.e., latent) variable
associated with each data point. It can be interpreted as a pointer to the component
of a mixture generated the sample. This component computes weights the components
in the mixture, that is, the probability for each component that the next sample will
be drawn from it. In case of non-probabilistic models (k-mm and SOM) this is irrelevant.