New Riemannian Priors on the Univariate Normal Model - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Entropy Année : 2014

New Riemannian Priors on the Univariate Normal Model

Résumé

The current paper introduces new prior distributions on the univariate normal model, with the aim of applying them to the classification of univariate normal populations. These new prior distributions are entirely based on the Riemannian geometry of the univariate normal model, so that they can be thought of as "Riemannian priors". Precisely, if $\lbrace p_\theta ; \theta \in \Theta \rbrace$ is any parametrization of the univariate normal model, the paper considers prior distributions $G(\bar{\theta},\gamma)$ with hyperparameters $\bar{\theta}\in \Theta$ and $\gamma > 0$, whose density with respect to Riemannian volume is proportional to $\exp(-d^2(\theta,\bar{\theta})/2\gamma^2)$, where $d^2(\theta,\bar{\theta})$ is the square of Rao's Riemannian distance. The distributions $G(\bar{\theta},\gamma)$ are termed Gaussian distributions on the univariate normal model. The motivation for considering a distribution $G(\bar{\theta},\gamma)$ is that this distribution gives a geometric representation of a class or cluster of univariate normal populations. Indeed, $G(\bar{\theta},\gamma)$ has a unique mode $\bar{\theta}\in \Theta$ (precisely, $\bar{\theta}\in \Theta$ is the unique Riemannian center of mass of $G(\bar{\theta},\gamma)$, as shown in the paper), and its dispersion away from $\bar{\theta}\in \Theta$ is given by $\gamma$. Therefore, one thinks of members of the class represented by $G(\bar{\theta},\gamma)$ as being centered around $\bar{\theta}\in \Theta$ and lying within a typical distance determined by γ. The paper defines rigorously the Gaussian distributions $G(\bar{\theta},\gamma)$ and describes an algorithm for computing maximum likelihood estimates of their hyperparameters. Based on this algorithm and on the Laplace approximation, it describes how the distributions $G(\bar{\theta},\gamma)$ can be used as prior distributions for Bayesian classification of large univariate normal populations. In a concrete application to texture image classification, it is shown that this leads to an improvement in performance over the use of conjugate priors.
Fichier principal
Vignette du fichier
Said14_Entropy.pdf (2.36 Mo) Télécharger le fichier
Origine : Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-01084330 , version 1 (19-11-2014)

Identifiants

Citer

Salem Said, Lionel Bombrun, Yannick Berthoumieu. New Riemannian Priors on the Univariate Normal Model. Entropy, 2014, 16, pp.4015 - 4031. ⟨10.3390/e16074015⟩. ⟨hal-01084330⟩
187 Consultations
384 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More