Random Matrix-Improved Estimation of the Wasserstein Distance between two Centered Gaussian Distributions
Résumé
This article proposes a method to consistently estimate functionals 1 p p i=1 f (λi(C1C2)) of the eigenvalues of the product of two covariance matrices C1, C2 ∈ R p×p based on the empirical estimates λi(Ĉ1Ĉ2) (Ĉa = 1 na na i=1 x (a) i x (a)T i), when the size p and number na of the (zero mean) samples x (a) i are similar. As a corollary, a consistent estimate of the Wasserstein distance (related to the case f (t) = √ t) between centered Gaussian distributions is derived. The new estimate is shown to largely outperform the classical sample covariance-based "plug-in" estimator. Based on this finding, a practical application to covariance estimation is then devised which demonstrates potentially significant performance gains with respect to state-of-the-art alternatives.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...