Geodesic regression

by   Frank Hansen, et al.

The theory of geodesic regression aims to find a geodesic curve which is an optimal fit to a given set of data. In this article we restrict ourselves to the Riemannian manifold of positive definite operators (matrices) on a Hilbert space of finite dimension. There is a unique geodesic curve connecting two positive definite operators, and it is given by the weighted geometric mean. The function that measures the squared Riemannian metric distance between an operator and a geodesic curve is not convex nor geodesically convex in the operators generating the curve. This is a marked difference to the situation in linear regression. The literature mainly tries to find numerical solutions that approximate the optimal curve in a single point. We suggest to apply a distance measure slightly coarser than the Riemannian metric. The ensuing control function faithfully identifies geodesic curves, and it coincides with the standard control function based on the Riemannian metric for commuting operators. The control function constructed in this way is geodesically convex. We are therefore able to find a global and uniquely defined optimal fit to any given set of data. The generators of the geodesic curve may also be determined as the unique solution to two operator equations.


page 1

page 2

page 3

page 4


Positive definite matrices and the S-divergence

Positive definite matrices abound in a dazzling variety of applications....

Multifidelity Covariance Estimation via Regression on the Manifold of Symmetric Positive Definite Matrices

We introduce a multifidelity estimator of covariance matrices formulated...

Curve Reconstruction in Riemannian Manifolds: Ordering Motion Frames

In this article we extend the computational geometric curve reconstructi...

Learning Riemannian Metrics

We propose a solution to the problem of estimating a Riemannian metric a...

Riemannian Langevin Monte Carlo schemes for sampling PSD matrices with fixed rank

This paper introduces two explicit schemes to sample matrices from Gibbs...

Multivariate manifold-valued curve regression in time

Fréchet global regression is extended to the context of bivariate curve ...

Representation and Characterization of Non-Stationary Processes by Dilation Operators and Induced Shape Space Manifolds

We have introduce a new vision of stochastic processes through the geome...

Please sign up or login with your details

Forgot password? Click here to reset