• Corpus ID: 2607135

Geometric methods for estimation of structured covariances

  title={Geometric methods for estimation of structured covariances},
  author={Lipeng Ning and Xianhua Jiang and Tryphon T. Georgiou},
Author(s): Ning, Lipeng; Jiang, Xianhua; Georgiou, Tryphon | Abstract: We consider problems of estimation of structured covariance matrices, and in particular of matrices with a Toeplitz structure. We follow a geometric viewpoint that is based on some suitable notion of distance. To this end, we overview and compare several alternatives metrics and divergence measures. We advocate a specific one which represents the Wasserstein distance between the corresponding Gaussians distributions and show… 

Figures from this paper

Geometric methods for structured covariance estimation

The metric induced by Monge-Kantorovich transportation of the respective probability measures leads to an efficient linear matrix inequality (LMI) formulation of the approximation problem and relates to approximation in the Hellinger metric.

The Wasserstein metric in Factor Analysis

  • L. NingT. Georgiou
  • Mathematics, Computer Science
    SIAM Conf. on Control and its Applications
  • 2013
This work considers the problem of approximating a (nonnegative definite) covariance matrix by the sum of two structured covariances – one which is diagonal and one which has low-rank and proposes alternating between optimization with respect to each of the two summands.

Riemannian Optimization and Approximate Joint Diagonalization for Blind Source Separation

This paper builds the first complete Riemannian optimization framework suited for BSS and AJD handling three classical constraints, and allowing to use a large panel of general optimization algorithms on manifolds.


In many practical applications second order moments are used for estimation of power spectrum. This can be cast as an inverse problem where we seek a spectrum consistent with a given state-covariance

Linear Models Based on Noisy Data and the Scheme.

. We address the problem of identifying linear relations among variables based on noisy measurements. This is, of course, a central question in problems involving “Big Data.” Often a key assumption

Linear models based on noisy data and the Frisch scheme

The problem of identifying linear relations among variables based on noisy measurements, a central question in the search for structure in large data sets, is addressed by describing certain fundamental contributions by the founders of the field and providing alternative modern proofs to certain key results.

On the Geometry of Maximum Entropy Problems

This work shows that a simple geometric result suffices to derive the form of the optimal solution in a large class of finite- and infinite-dimensional maximum entropy problems concerning probability distributions, spectral densities, and covariance matrices, and applies this orthogonality principle to the new problem of completing a block-circulant covariance matrix when an a priori estimate is available.

Optimal Mass Transport for Statistical Estimation, Image Analysis, Information Geometry, and Control

A wide range of generalizations and insights are considered for the purpose of tackling problems of AFOSR interest, including matrix-valued statistics and fusion of information, optical flow, controlled active vision, tracking and dynamic textures.

Metrics for multivariate power spectra

Divergence measures are constructed based on the degradation of prediction-error and smoothing-error variances which induce Riemannian metrics which generalize earlier results for scalar PSD's.



Estimation of structured covariance matrices

An iterative algorithm that solves the necessary gradient equations for moderate-sized problems with reasonable computational ease is presented and theoretical convergence properties of the basic algorithm are investigated and robust modifications discussed.

Hellinger Versus Kullback–Leibler Multivariable Spectrum Approximation

This paper introduces a new metric on multivariable spectral densities induced by the family of their spectral factors, which, in the scalar case, reduces to the Hellinger distance, and solves the corresponding constrained optimization problem via duality theory.

Geometry of canonical correlation on the state space of a quantum system

A Riemannian metric is defined on the state space of a finite quantum system by the canonical correlation (or Kubo–Mori/Bogoliubov scalar product). This metric is infinitesimally induced by the

Differential-geometrical methods in statistics

1. Introduction.- I. Geometrical Structures of a Family of Probability Distributions.- 2. Differential Geometry of Statistical Models.- 2.1. Manifold of statistical model.- 2.2. Tangent space.- 2.3.

The Metric of Bures and the Geometric Phase.

After the appearance of the papers of Berry [1], Simon [2], and of Wilczek and Zee [3], I tried to understand [4], whether there is a reasonable extension of the geometric phase - or, more

On the optimal mapping of distributions

We consider the problem of mappingX→Y, whereX andY have given distributions, so as to minimize the expected value of ∣X–Y∣2. This is equivalent to finding the joint distribution of the random

Elements of Information Theory

The author examines the role of entropy, inequality, and randomness in the design of codes and the construction of codes in the rapidly changing environment.

A new method for moving-average parameter estimation

An apparently original method for moving-average parameter estimation, based on covariance fitting and convex optimization, is introduced, shown by means of numerical simulation to provide much more accurate parameter estimates, in difficult scenarios, than a related existing method does.

Toeplitz Forms And Their Applications

Part I: Toeplitz Forms: Preliminaries Orthogonal polynomials. Algebraic properties Orthogonal polynomials. Limit properties The trigonometric moment problem Eigenvalues of Toeplitz forms