Alpha Procrustes metrics between positive definite operators: a unifying formulation for the Bures-Wasserstein and Log-Euclidean/Log-Hilbert-Schmidt metrics

DOI10.1016/J.LAA.2021.11.011zbMATH Open1491.15037arXiv1908.09275OpenAlexW3214736133WikidataQ114151961 ScholiaQ114151961MaRDI QIDQ2069860FDOQ2069860

Authors: Hà Quang Minh 

Publication date: 21 January 2022

Published in: Linear Algebra and its Applications (Search for Journal in Brave)

Abstract: This work presents a parametrized family of distances, namely the Alpha Procrustes distances, on the set of symmetric, positive definite (SPD) matrices. The Alpha Procrustes distances provide a unified formulation encompassing both the Bures-Wasserstein and Log-Euclidean distances between SPD matrices. We show that the Alpha Procrustes distances are the Riemannian distances corresponding to a family of Riemannian metrics on the manifold of SPD matrices, which encompass both the Log-Euclidean and Wasserstein Riemannian metrics. This formulation is then generalized to the set of positive definite Hilbert-Schmidt operators on a Hilbert space, unifying the infinite-dimensional Bures-Wasserstein and Log-Hilbert-Schmidt distances. In the setting of reproducing kernel Hilbert spaces (RKHS) covariance operators, we obtain closed form formulas for all the distances via the corresponding kernel Gram matrices. From a statistical viewpoint, the Alpha Procrustes distances give rise to a parametrized family of distances between Gaussian measures on Euclidean space, in the finite-dimensional case, and separable Hilbert spaces, in the infinite-dimensional case, encompassing the 2-Wasserstein distance, with closed form formulas via Gram matrices in the RKHS setting. The presented formulations are new both in the finite and infinite-dimensional settings.

Full work available at URL: https://arxiv.org/abs/1908.09275

Recommendations

• A unified formulation for the Bures-Wasserstein and log-Euclidean/log-Hilbert-Schmidt distances between positive definite operators
• Riemannian distances between covariance operators and Gaussian processes
• Alpha-beta log-determinant divergences between positive definite trace class operators
• Infinite-dimensional log-determinant divergences. III: Log-Euclidean and log-Hilbert-Schmidt divergences
• Affine-invariant Riemannian distance between infinite-dimensional covariance operators

zbMATH Keywords

reproducing kernel Hilbert spacesWasserstein distanceGaussian measurescovariance operatorsProcrustes distanceLog-Euclidean distanceLog-Hilbert-Schmidt distancepositive definite matrices and operators

Mathematics Subject Classification ID

Positive matrices and their generalizations; cones of matrices (15B48) Positive linear operators and order-bounded operators (47B65) Operator means involving linear operators, shorted linear operators, etc. (47A64) Linear operators in reproducing-kernel Hilbert spaces (including de Branges, de Branges-Rovnyak, and other structured spaces) (47B32) Probabilistic measure theory (60A10)

Cites Work

• Support Vector Machines
• A class of Wasserstein metrics for probability distributions
• Geometric Means in a Novel Vector Space Structure on Symmetric Positive‐Definite Matrices
• Wasserstein geometry of Gaussian measures
• Riemannian geometry
• The Frechet distance between multivariate normal distributions
• A Riemannian framework for tensor computing
• Title not available (Why is that?)
• On a Formula for the L2 Wasserstein Metric between Measures on Euclidean and Hilbert Spaces
• Title not available (Why is that?)
• Solution of the Equation $AX + XB = C$ by Inversion of an $M \times M$ or $N \times N$ Matrix
• The distance between two random vectors wigh given dispersion matrices
• On lower bounds for the \(L^ 2\)-Wasserstein metric in a Hilbert space
• On an inequality of Lieb and Thirring
• Means of Hermitian positive-definite matrices based on the log-determinant \(\alpha\)-divergence function
• Log-determinant divergences revisited: alpha-beta and gamma log-det divergences
• Non-Euclidean statistics for covariance matrices, with applications to diffusion tensor imaging
• Wasserstein Riemannian geometry of Gaussian densities
• Trace and Eigenvalue Inequalities for Ordinary and Hadamard Products of Positive Semidefinite Hermitian Matrices
• Nonpositive curvature: A geometrical approach to Hilbert-Schmidt operators
• Infinite-dimensional log-determinant divergences between positive definite trace class operators
• Infinite-dimensional log-determinant divergences between positive definite Hilbert-Schmidt operators
• Alpha-beta log-determinant divergences between positive definite trace class operators
• A unified formulation for the Bures-Wasserstein and log-Euclidean/log-Hilbert-Schmidt distances between positive definite operators
• Is affine-invariance well defined on SPD matrices? A principled continuum of metrics
• Infinite-dimensional log-determinant divergences. III: Log-Euclidean and log-Hilbert-Schmidt divergences
• Riemannian Geometry of Symmetric Positive Definite Matrices via Cholesky Decomposition
• Geometric Distance Between Positive Definite Matrices of Different Dimensions

Cited In (7)

• Relative operator entropy properties related to some weighted metrics
• Convergence and finite sample approximations of entropic regularized Wasserstein distances in Gaussian and RKHS settings
• Entropic regularization of Wasserstein distance between infinite-dimensional Gaussian measures and Gaussian processes
• Riemannian distances between covariance operators and Gaussian processes
• A unified formulation for the Bures-Wasserstein and log-Euclidean/log-Hilbert-Schmidt distances between positive definite operators
• Tsallis relative operator entropy properties with some weighted metrics
• Infinite-dimensional log-determinant divergences. III: Log-Euclidean and log-Hilbert-Schmidt divergences

This page was built for publication: Alpha Procrustes metrics between positive definite operators: a unifying formulation for the Bures-Wasserstein and Log-Euclidean/Log-Hilbert-Schmidt metrics

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2069860)