Abstract: We study information matrices for statistical models by the -Wasserstein metric. We call them Wasserstein information matrices (WIMs), which are analogs of classical Fisher information matrices. We introduce Wasserstein score functions and study covariance operators in statistical models. Using them, we establish Wasserstein-Cramer-Rao bounds for estimations and explore their comparisons with classical results. We next consider the asymptotic behaviors and efficiency of estimators. We derive the on-line asymptotic efficiency for Wasserstein natural gradient. Besides, we study a Poincar'e efficiency for Wasserstein natural gradient of maximal likelihood estimation. Several analytical examples of WIMs are presented, including location-scale families, independent families, and rectified linear unit (ReLU) generative models.
Information matrices for statistical models, similar to the classical Fisher information matrices, are studied by means of the \(L^2\)-Wasserstein metric. New Wasserstein score functions are introduced to this end and covariance operators are investigated in statistical models, delivering, in particular, Wasserstein-Cramer-Rao bounds for estimations. Online asymptotic efficiency for the Wasserstein natural gradient and a Poincaré efficiency for the Wasserstein natural gradient of maximal likelihood estimation are then established. Some analytical examples of such matrices are provided, too.
Recommendations
- Information geometry of Wasserstein divergence
- On a prior based on the Wasserstein information matrix
- Information geometry connecting Wasserstein distance and Kullback-Leibler divergence via the entropy-relaxed transportation problem
- Information geometry of positive definite matrices
- scientific article; zbMATH DE number 3872359
- Information matrices for Laplace and Pareto mixtures
- Information geometry of the Gaussian space
- Information matrices for normal and Laplace mixtures
- Fisher information regularization schemes for Wasserstein gradient flows
Cites work
- A formalization of the natural gradient method for general similarity measures
- Affine natural proximal learning
- deBruijn identities: From Shannon, Kullback-Leibler and Fisher to generalized φ-entropies, φ-divergences and φ-Fisher informations
- Differential-geometrical methods in statistics.
- Elements of Information Theory
- Generalization of an inequality by Talagrand and links with the logarithmic Sobolev inequality
- scientific article; zbMATH DE number 1077733 (Why is no real title available?)
- scientific article; zbMATH DE number 1909499 (Why is no real title available?)
- Information geometry
- Logarithmic divergences from optimal transport and Rényi geometry
- Natural gradient via optimal transport
- On parameter estimation with the Wasserstein distance
- Online natural gradient as a Kalman filter
- Optimal Transport
- Optimal transport natural gradient for statistical manifolds with continuous sample space
- Parametric Fokker-Planck equation
- Ricci curvature for parametric statistics via optimal transport
- Some geometric calculations on Wasserstein space
- The Density Manifold and Configuration Space Quantization
- THE GEOMETRY OF DISSIPATIVE EVOLUTION EQUATIONS: THE POROUS MEDIUM EQUATION
- Transport information geometry: Riemannian calculus on probability simplex
- Wasserstein covariance for multiple random densities
- Wasserstein proximal of GANs
- Wasserstein statistics in one-dimensional location scale models
Cited in
(9)- On a prior based on the Wasserstein information matrix
- Techniques for continuous optimal transport problem
- Geometric thermodynamics for the Fokker-Planck equation: stochastic thermodynamic links between information geometry and optimal transport
- Information geometry of Wasserstein statistics on shapes and affine deformations
- Numerical analysis on neural network projected schemes for approximating one dimensional Wasserstein gradient flows
- Manifold learning in metric spaces
- Embedding distributional data
- Parameterized Wasserstein gradient flow
- Information geometric regularization of the barotropic Euler equation
This page was built for publication: Wasserstein information matrix
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6170229)
