Riemannian Gaussian Distributions on the Space of Symmetric Positive Definite Matrices

DOI10.1109/TIT.2017.2653803MaRDI QIDQ5280889zbMATH OpenOpenAlexWikidataFDO

Authors Salem Said, Lionel Bombrun, Yannick Berthoumieu, Jonathan H. Manton

Publication date 27 July 2017

Published in IEEE Transactions on Information Theory (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1507.01760

Mathematics Subject Classification ID

Estimation in multivariate analysis (62H12) Random matrices (probabilistic aspects) (60B20) Measures of information, entropy (94A17) Methods of local Riemannian geometry (53B21) Positive matrices and their generalizations; cones of matrices (15B48) Matrices over special rings (quaternions, finite fields, etc.) (15B33)

Abstract: Data which lie in the space $mathcal{P}_{m,}$, of $mimesm$ symmetric positive definite matrices, (sometimes called tensor data), play a fundamental role in applications including medical imaging, computer vision, and radar signal processing. An open challenge, for these applications, is to find a class of probability distributions, which is able to capture the statistical properties of data in $mathcal{P}_{m,}$, as they arise in real-world situations. The present paper meets this challenge by introducing Riemannian Gaussian distributions on $mathcal{P}_{m,}$. Distributions of this kind were first considered by Pennec in $2006$. However, the present paper gives an exact expression of their probability density function for the first time in existing literature. This leads to two original contributions. First, a detailed study of statistical inference for Riemannian Gaussian distributions, uncovering the connection between maximum likelihood estimation and the concept of Riemannian centre of mass, widely used in applications. Second, the derivation and implementation of an expectation-maximisation algorithm, for the estimation of mixtures of Riemannian Gaussian distributions. The paper applies this new algorithm, to the classification of data in $mathcal{P}_{m,}$, (concretely, to the problem of texture classification, in computer vision), showing that it yields significantly better performance, in comparison to recent approaches.

Cited in

(26)

• Kernel density estimation on spaces of Gaussian distributions and symmetric positive definite matrices
• Warped Riemannian metrics for location-scale models
• A hyperbolic approach for learning communities on graphs
• Differential geometry with extreme eigenvalues in the positive semidefinite cone
• Probabilistic learning vector quantization on manifold of symmetric positive definite matrices
• A Riemannian structure for correlation matrices
• Statistics on the Stiefel manifold: theory and applications
• From Bayesian inference to MCMC and convex optimisation in Hadamard manifolds
• Gaussian distributions on Riemannian symmetric spaces in the large \(N\) limit
• Online learning of Riemannian hidden Markov models in homogeneous Hadamard spaces
• Intrinsic wavelet regression for curves of Hermitian positive definite matrices
• Nonparametric matrix regression function estimation over symmetric positive definite matrices
• Riemannian Gaussian distributions, random matrix ensembles and diffusion kernels
• Wasserstein barycenters in the manifold of all positive definite matrices
• Averaging symmetric positive-definite matrices
• Geometric mean of partial positive definite matrices with missing entries
• Time-varying spectral matrix estimation via intrinsic wavelet regression for surfaces of Hermitian positive definite matrices
• Inference for partially observed Riemannian Ornstein-Uhlenbeck diffusions of covariance matrices
• Quantization and clustering on Riemannian manifolds with an application to air traffic analysis
• Intrinsic Data Depth for Hermitian Positive Definite Matrices
• Optimal transport and variational Bayesian inference
• Gaussian Distributions on Riemannian Symmetric Spaces: Statistical Learning With Structured Covariance Matrices
• Exponential-Wrapped Distributions on Symmetric Spaces
• A nonlocal denoising algorithm for manifold-valued images using second order statistics
• Tracking and Regret Bounds for Online Zeroth-Order Euclidean and Riemannian Optimization
• First order methods for optimization on Riemannian manifolds

This page was built for publication: Riemannian Gaussian Distributions on the Space of Symmetric Positive Definite Matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5280889)