Wasserstein Riemannian geometry of Gaussian densities

DOI10.1007/S41884-018-0014-4MaRDI QIDQ1713654zbMATH OpenOpenAlexWikidataFDO

Authors Luigi Malagò, Luigi Montrucchio, Giovanni Pistone

Publication date 28 January 2019

Published in Information Geometry (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1801.09269

zbMATH Keywords

Gaussian distributioninformation geometryLevi-Civita covariant derivativenatural gradientnormal coordinatesoptimization on positive-definite symmetric matricesRiemannian exponentialRiemannian metricsWasserstein distance

Mathematics Subject Classification ID

Positive matrices and their generalizations; cones of matrices (15B48) Global geometric and topological methods (à la Gromov); differential geometric analysis on metric spaces (53C23) Special Riemannian manifolds (Einstein, Sasakian, etc.) (53C25) Geometric probability and stochastic geometry (60D05)

Abstract: The Wasserstein distance on multivariate non-degenerate Gaussian densities is a Riemannian distance. After reviewing the properties of the distance and the metric geodesic, we present an explicit form of the Riemannian metrics on positive-definite matrices and compute its tensor form with respect to the trace inner product. The tensor is a matrix which is the solution to a Lyapunov equation. We compute the explicit formula for the Riemannian exponential, the normal coordinates charts and the Riemannian gradient. Finally, the Levi-Civita covariant derivative is computed in matrix form together with the differential equation for the parallel transport. While all computations are given in matrix form, nonetheless we discuss also the use of a special moving frame.

Recommendations

• Wasserstein Riemannian geometry of Gamma densities
• On Wasserstein geometry of the space of Gaussian measures
• Wasserstein geometry of Gaussian measures
• Wasserstein barycenters in the manifold of all positive definite matrices
• Geometry of matrix decompositions seen through optimal transport and information geometry

Cites work

• A class of Wasserstein metrics for probability distributions
• A convexity principle for interacting gases
• A infinite-dimensional geometric structure on the space of all the probability measures equivalent to a given one
• Combinatorial optimization with information geometry: the Newton method
• Computational Methods for Linear Matrix Equations
• Estimation of non-normalized statistical models by score matching
• scientific article; zbMATH DE number 3174491 (Why is no real title available?)
• scientific article; zbMATH DE number 5131267 (Why is no real title available?)
• scientific article; zbMATH DE number 3928102 (Why is no real title available?)
• scientific article; zbMATH DE number 52737 (Why is no real title available?)
• scientific article; zbMATH DE number 1261669 (Why is no real title available?)
• scientific article; zbMATH DE number 713322 (Why is no real title available?)
• scientific article; zbMATH DE number 1964693 (Why is no real title available?)
• scientific article; zbMATH DE number 1560711 (Why is no real title available?)
• scientific article; zbMATH DE number 5223994 (Why is no real title available?)
• Infinite dimensional analysis. A hitchhiker's guide.
• Information geometry and its applications
• Information Geometry of the Gaussian Distribution in View of Stochastic Optimization
• Kernel density estimation on spaces of Gaussian distributions and symmetric positive definite matrices
• Metric spaces, convexity and nonpositive curvature
• Nonparametric information geometry
• On a Formula for the L2 Wasserstein Metric between Measures on Euclidean and Hilbert Spaces
• On the Bures-Wasserstein distance between positive definite matrices
• On the optimal mapping of distributions
• Polar factorization and monotone rearrangement of vector‐valued functions
• Polar factorization of maps on Riemannian manifolds
• Proper local scoring rules
• Riemannian geometry.
• Some geometric calculations on Wasserstein space
• The Density Manifold and Configuration Space Quantization
• The distance between two random vectors wigh given dispersion matrices
• The Frechet distance between multivariate normal distributions
• The Fritz John necessary optimality conditions in the presence of equality and inequality constraints
• THE GEOMETRY OF DISSIPATIVE EVOLUTION EQUATIONS: THE POROUS MEDIUM EQUATION
• Trail to a Lyapunov equation solver
• Wasserstein geometry of Gaussian measures

Cited in

(42)

• Optimal transport natural gradient for statistical manifolds with continuous sample space
• Accelerated information gradient flow
• Alpha Procrustes metrics between positive definite operators: a unifying formulation for the Bures-Wasserstein and Log-Euclidean/Log-Hilbert-Schmidt metrics
• Bures-Wasserstein geometry for positive-definite Hermitian matrices and their trace-one subset
• When optimal transport meets information geometry
• Entropy-regularized 2-Wasserstein distance between Gaussian measures
• The Schrödinger problem on the non-commutative Fisher-Rao space
• Geometry and fixed-rate quantization in Riemannian metric spaces induced by separable Bregman divergences
• Curvature of the manifold of fixed-rank positive-semidefinite matrices endowed with the Bures-Wasserstein metric
• Hessian transport gradient flows
• Geometry of matrix decompositions seen through optimal transport and information geometry
• \(O(n)\)-invariant Riemannian metrics on SPD matrices
• Nonlinear Diffusion: Geodesic Convexity is Equivalent to Wasserstein Contraction
• Averaging symmetric positive-definite matrices
• Riemannian distances between covariance operators and Gaussian processes
• A fanning scheme for the parallel transport along geodesics on Riemannian manifolds
• Geodesic PCA versus Log-PCA of Histograms in the Wasserstein Space
• Wasserstein Riemannian geometry of Gamma densities
• Riemannian conjugate gradient methods: general framework and specific algorithms with convergence analyses
• Wasserstein Riemannian geometry on statistical manifold
• Geodesic PCA in the Wasserstein space by convex PCA
• On Wasserstein geometry of the space of Gaussian measures
• Bures–Wasserstein Minimizing Geodesics between Covariance Matrices of Different Ranks
• Invariance properties of the natural gradient in overparametrised systems
• Characterization of invariant inner products
• Learning with symmetric positive definite matrices via generalized Bures-Wasserstein geometry
• Fisher-Rao Riemannian geometry of equivalent Gaussian measures on Hilbert space
• Affine statistical bundle modeled on a Gaussian Orlicz-Sobolev space
• Geometry and convergence of natural policy gradient methods
• Spatiotemporal analysis using Riemannian composition of diffusion operators
• Designing universal causal deep learning models: The geometric (Hyper)transformer
• Proxying credit curves via Wasserstein distances
• A Wasserstein-type distance for Gaussian mixtures on vector bundles with applications to shape analysis
• On Wasserstein distances for affine transformations of random vectors
• Approximation of splines in Wasserstein spaces
• Wasserstein geometry of Gaussian measures
• Covariance-modulated optimal transport and gradient flows
• Riemannian-geometric generalizations of quantum fidelities and Bures-Wasserstein distance
• On the Wasserstein Median of Probability Measures
• Multifidelity covariance estimation via regression on the manifold of symmetric positive definite matrices
• Adapted optimal transport between Gaussian processes in discrete time
• Information geometry of the Otto metric

This page was built for publication: Wasserstein Riemannian geometry of Gaussian densities

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1713654)