Distribution asymptotique de l'affinité L 2 de densités gaussiennes

Abstract

After recalling the L 2 affinity measure between two multidimensionnal Gaussian density functions, we prove its asymptotic normality when parameters of one density are replaced by their maximum likelihood estimators. We extend this result to the vectorial case and we use it to allocate an estimated density to a class of densities as does the discriminant analysis method. As an application, a dating method for archeological data is proposed.