On multisample multinomial mixture model

This paper deals with the parameter estimation of mixed component distributions from a series of histograms, where mixing weights are variable. Exploring a series of histograms (rather than the total histogram) is studied in this paper since the fitting a series of histograms is more informative than that for a total histogram. To avoid the inconsistency in estimating the weights of classes, a mixed model with random weights is proposed. The modification of the iterative reweighted Gauss–Newton algorithm is suggested for the parameter estimation of ageneralized multivariate regression model applicable for singular covariance matrics of multivariate observations. The consistency and asymptotic normality for the parameter estimates of the model follows under our assumptions from Malyutov.