A family of anisotropic distributions on the hyperbolic plane

07/11/2017
Publication GSI2017
OAI : oai:www.see.asso.fr:17410:22556
contenu protégé  Document accessible sous conditions - vous devez vous connecter ou vous enregistrer pour accéder à ou acquérir ce document.
- Accès libre pour les ayants-droit
 

Résumé

Most of the parametric families of distributions on manifold are constituted of radial distributions. The main reason is that quantifying the anisotropy of a distribution on a manifold is not as straightforward as in vector spaces and usually leads to numerical computations.
Based on a simple de nition of the covariance on manifolds, this paper presents a way of constructing anisotropic distributions on the hyperbolic space whose covariance matrices are explicitly known. The approach remains valid on every manifold homeomorphic to vector spaces.

A family of anisotropic distributions on the hyperbolic plane

Collection

application/pdf A family of anisotropic distributions on the hyperbolic plane Emmanuel Chevallier
Détails de l'article
contenu protégé  Document accessible sous conditions - vous devez vous connecter ou vous enregistrer pour accéder à ou acquérir ce document.
- Accès libre pour les ayants-droit

Most of the parametric families of distributions on manifold are constituted of radial distributions. The main reason is that quantifying the anisotropy of a distribution on a manifold is not as straightforward as in vector spaces and usually leads to numerical computations.
Based on a simple de nition of the covariance on manifolds, this paper presents a way of constructing anisotropic distributions on the hyperbolic space whose covariance matrices are explicitly known. The approach remains valid on every manifold homeomorphic to vector spaces.
A family of anisotropic distributions on the hyperbolic plane

Média

Voir la vidéo

Métriques

0
0
289.97 Ko
 application/pdf
bitcache://0da3be3ed689aece37ea9894716bcb0f54f2adfc

Licence

Creative Commons Aucune (Tous droits réservés)

Sponsors

Sponsors Platine

alanturinginstitutelogo.png
logothales.jpg

Sponsors Bronze

logo_enac-bleuok.jpg
imag150x185_couleur_rvb.jpg

Sponsors scientifique

logo_smf_cmjn.gif

Sponsors

smai.png
gdrmia_logo.png
gdr_geosto_logo.png
gdr-isis.png
logo-minesparistech.jpg
logo_x.jpeg
springer-logo.png
logo-psl.png

Organisateurs

logo_see.gif
<resource  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xmlns="http://datacite.org/schema/kernel-4"
                xsi:schemaLocation="http://datacite.org/schema/kernel-4 http://schema.datacite.org/meta/kernel-4/metadata.xsd">
        <identifier identifierType="DOI">10.23723/17410/22556</identifier><creators><creator><creatorName>Emmanuel Chevallier</creatorName></creator></creators><titles>
            <title>A family of anisotropic distributions on the hyperbolic plane</title></titles>
        <publisher>SEE</publisher>
        <publicationYear>2018</publicationYear>
        <resourceType resourceTypeGeneral="Text">Text</resourceType><dates>
	    <date dateType="Created">Wed 7 Mar 2018</date>
	    <date dateType="Updated">Wed 7 Mar 2018</date>
            <date dateType="Submitted">Mon 19 Nov 2018</date>
	</dates>
        <alternateIdentifiers>
	    <alternateIdentifier alternateIdentifierType="bitstream">0da3be3ed689aece37ea9894716bcb0f54f2adfc</alternateIdentifier>
	</alternateIdentifiers>
        <formats>
	    <format>application/pdf</format>
	</formats>
	<version>37270</version>
        <descriptions>
            <description descriptionType="Abstract">Most of the parametric families of distributions on manifold are constituted of radial distributions. The main reason is that quantifying the anisotropy of a distribution on a manifold is not as straightforward as in vector spaces and usually leads to numerical computations.<br />
Based on a simple de nition of the covariance on manifolds, this paper presents a way of constructing anisotropic distributions on the hyperbolic space whose covariance matrices are explicitly known. The approach remains valid on every manifold homeomorphic to vector spaces.
</description>
        </descriptions>
    </resource>
.

A family of anisotropic distributions on the hyperbolic plane Emmanuel Chevallier1 Weizmann Institute of Science Abstract. Most of the parametric families of distributions on manifold are constituted of radial distributions. The main reason is that quanti- fying the anisotropy of a distribution on a manifold is not as straightfor- ward as in vector spaces and usually leads to numerical computations. Based on a simple definition of the covariance on manifolds, this paper presents a way of constructing anisotropic distributions on the hyperbolic space whose covariance matrices are explicitly known. The approach re- mains valid on every manifold homeomorphic to vector spaces. 1 Introduction Probability density estimation on Riemannian manifolds is the subject of several recent studies. The different approaches can be separated into two categories, the parametric and non-parametric ones. The context of Riemannian manifolds brings difficulties of two kinds. Firstly, the theoretical results about distributions and the convergence of estimators known for random variables valued in Rn have to be adapted to the case of random variables valued in Riemannian manifolds, see [2, 9, 3, 1, 8, 12–15]. Secondly, the construction of probability distribution and of density estimators should require a reasonable amount of computational com- plexity, see [8, 12, 13, 16–18]. A generalization of the Gaussian distribution on manifolds was proposed in [8]. Although the expression of the proposed law is hard to compute on general manifolds, expressions of radial Gaussians on sym- metric spaces can be found in [12–14]. On isotropic spaces, an isotropic density is simply a radial density. The anisotropy of a density can be evaluated with the notion of covariance proposed in [8]. In this paper, we are interested in the construction of anisotropic distribu- tions on the hyperbolic space. The problem of anisotropic normal distributions on manifold have been addressed in [19] through anisotropic diffusion. The con- struction is valid on arbitrary manifolds but requires important computations. The hyperbolic space is a very particular Riemannian manifold: it is at the same time isotropic and diffeomorphic to a vector space. These two specificities signif- icantly ease the construction of probability distributions and probability density estimators. Generally, it is difficult to control the covariance of a distribution on a Riemannian manifold, e.g. the covariance of the Gaussian law proposed in [8]. We propose a simple way of constructing distributions whose covariance is fully controlled. The method is derived from the density kernel proposed by [1]. These distributions can be used in the non parametric kernel density estimator but also to design mixture models for parametric density estimation. The paper is organised as follows. Section 2 is a very brief introduction to the hyperbolic plane. Section 3 reviews some general facts about probabilities on Riemannian manifolds. Section 4 describes how to built anisotropic density functions on the hyperbolic space. 2 The hyperbolic space The hyperbolic geometry results of a modification of the fifth Euclid’s postulate on parallel lines. In two dimensions, given an line D and a point p / ∈ D, the hyperbolic geometry is an example where there are at least two lines going through p, which do not intersect D. Let us consider the open unit disk of the Euclidean plane endowed with the Riemannian metric: ds2 D = 4 dx2 + dy2 (1 − x2 − y2)2 (1) where x and y are the Cartesian coordinates. The unit disk D endowed with dsD is called the Poincaré disk and is a model of the two-dimensional hyperbolic geometry. The construction is generalized to higher dimensions. Let ISO be the isometry group of D. It can be shown that: – D is homogeneous: ∀p, q ∈ D, ∃φ ∈ ISO, φ(p) = q, points are indistinguish- able. – D is isotropic: for any couple of geodesics γ1 and γ2 going through a point p ∈ D, there exists φ ∈ ISO such that φ(p) = p and φ(γ1) = γ2. In other words, directions are indistinguishable. – the Riemannian exponential applications are bijective. – D has a constant negative curvature. Let x denote the coordinates of elements of TpD in an orthogonal basis. x is mapped to a point on D by the Reimannian exponential application noted expp and form thus a chart of D. This chart is called an exponential chart at the point p. Given a reference point p the point of polar coordinates (r, α) of the hyper- bolic space is defined as the point at distance r of p on the geodesic with initial direction α ∈ S1 . Since the hyperbolic space is isotropic, the expression of the metric in polar coordinates only depends on r, ds2 = dr2 + sinh(r)2 dα2 , (2) see [10, 11]. 3 Distributions on D 3.1 Densities The metric of a Riemmanian manifold provides a measure of volumes vol. In a chart, if G is the matrix of the metric, the density of vol with respect to the Lebesgue measure of the chart is dvol dLeb = |det( √ G)| where √ G is the matrix square root of G. Let µ be a measure on M. If µ has a density f with respect to the Lebesgue measure of a chart, then the density with respect to the Riemannian volume measure is given by dµ dvol = dµ dLeb dLeb dvol = 1 |det( √ G)| f. (3) 3.2 Intrinsic means Given a distribution µ, the variance at p be defined by σ2 (p) = Z D d(p, .)2 dµ. When the variance is finite everywhere, its minima are called mean points. The hyperbolic space is a Cartan-Hadamar manifold, that is to say it is complete, simply connected and of negative curvature. On Cartan-Hadamar manifolds, when the variance is finite everywhere, the mean exists and is unique, see [8] corollary 2. It is achieved at p such that Z TpD xdµ̃ = 0, where µ̃ is the image of the measure µ by the inverse of the exponential appli- cation at p. 3.3 Covariance on manifold The covariance of a random vector is the matrix formed by the covariance of its coordinates. In a vector space the coordinates of a vector are given in terms of projection on the corresponding axis. On a Riemannian manifold the notions of projection on a geodesic usually do not lead to explicit expressions. Even if it does not conserve all the properties of the covariance of vectors, when possible, the simplest generalisation to manifolds is to take the Euclidean covariance after lifting the distribution on a tangent space by the inverse of the exponential map, see [8]. Since on the hyperbolic space the exponential application a bijection, it is always possible to lift distributions on tangent spaces. Given a distribution µ and a orthogonal basis of TpD, the covariance at p ∈ D is thus defined as Σp(µ) = Z TpD xxt dµ̃ This definition of covariance was used to define a notion of principal geodesic analysis on manifolds in [20]. It can be noted that the covariance at the point p is a point in TD ⊗ TD. 4 Constructing anisotropic distributions The author of [8] proposes a generalization of Gaussian distributions on man- ifolds as the distribution that maximizes the entropy given its barycenter and covariance. This generalization leads to a density of the form, N(p,Γ )(expp(x)) = k. exp  − xt Γx 2  Given p and the covariance matrix Σp, the main difficulties are to obtain ex- pressions of the normalizing factor k and of the concentration matrix Γ. Since hyperbolic space is homogenous, k and Γ only depend on the matrix Σp. The expression of k and Γ when Σp is a (positive) mutliple of the identity matrix can be found in [12]. However, it is difficult to obtain these relations when Σp is not diagonal. It might be interesting to define parametric families of distributions whose means and covariances can easily be controlled, even if they do not verify the same statistical properties as the Gaussian distributions. Let K : R+ → R+ be a function such that, i. R R2 K(kyk) dy = 1 ii. R R2 kyk2 K(kyk) dy = 2 Given Γ a symmetric positive definite matrix, we have then Z R2 1 p det(Γ) K( √ xtΓ−1x)dx = 1. Let p be a point in D. Set an orthonormal basis of the tangent space TpD and consider the distribution νp,Γ on TpD whose density with respect to the Lebesgue measure of TpD is given by 1 √ det(Σ) K( √ xtΓ−1x), where x and Γ are expressed in the reference basis. Let µp,Γ = expp∗(νp,Γ ) be the pushforward measure of νp,Γ by the Riemannian exponential at p. Theorem 1. p is the unique mean of µp,Γ . Proof. It can be checked that µp,Γ has a finite variance everywhere. Moreover, Z TpD √ Γ−1x 1 √ det Γ K( √ xtΓ−1x) dx = 0. The integrability of the function can be deduced from i and ii and the nullity from its symmetry. Therefore according to paragraph 3.2 p is the unique mean of µp,Γ . Theorem 2. The covariance Σp of µp,Γ at p and the concentration matrix Γ are equal. Proof. In the reference basis, making use of ii with the change of variables y = √ Γ−1x Σp = Z R2 xxt 1 p det(Γ) K( √ xtΓ−1x)dx = Γ1/2 Z R2 yyt K( p yty)dyΓ1/2 = Γ1/2 Z R Z 2π 0 r2  cos(θ) sin(θ)   cos(θ) sin(θ) t K(r)rdrdθ ! Γ1/2 = Γ1/2  1 2 Z R r2 IK(r)2πrdr  Γ1/2 = Γ1/2 I  1 2 Z R2 kyk2 K(kyk) dy  Γ1/2 = Γ. The tangent space TpD endowed with the reference basis provides a parametriza- tion of the hyperbolic space. By definition, the density of µp,Γ in this parametriza- tion is given by 1 √ det(Σ) K( √ xtΣ−1x). In order to obtain the density with respect to the Riemannian measure this term should be multiplied by the density of the Lebesgue measure of the parametrization with respect to the Riemannian mea- sure, see Eq. 3. In an adapted orthonormal basis of TpD, Eq. 2 leads to the following expression of the matrix of the metric, G = 1 0 0 sinh(r)2 r2 ! . Thus, det( √ G) = sinh(r) r . Eq.3 leads to the density ratio, dx dvol (x) = ||x|| sinh(||x||) , where dx is the Lebesgue measure induced by the reference basis. Recall that in this parametrization, the Euclidean norm of x is the distance between expp(x) and p. The density of µp,Γ with respect to the Riemannian measure is given by f(expp(x)) = ||x|| sinh(||x||) p det(Σ) K √ xtΣ−1x  . Fig. 1 shows the level lines when K is Gaussian. Fig. 1. In this example K(x) = 1 √ 2π e−x2 and Σ has 1 and 1 4 as eigenvalues. The level lines of the corresponding density f are flattened circle but are not ellipses. 5 Estimating the mean and the covariance Let the function K and the distribution µp,Γ be as defined in section 4. Given a set of draws drawn from this distribution it is important to have estimators of the two parameters: the mean and the covariance. In order to estimate the unknown parameters (p, Σp) given a set of independent samples (p1, .., pn), it is usual to try to maximize the likelihood function. The log-likelihood of a set of samples is defined as L(p1, .., pn; (p̂, Σ̂)) = X i log   ||xi|| sinh(||xi||) q det(Σ̂) K q xt iΣ̂−1xi    = X i log   ||xi|| sinh(||xi||) q det(Σ̂)   + log  K q xt iΣ̂−1xi  . The major difficulty is that it is not possible to optimize the mean and the covari- ance separately. Thus there might not be explicit expressions of the maximum likelihood. However, the mean and the covariance have natural estimators. It is already known that the empirical barycenter is a strongly consistent estimator of the barycenter, see [21] theorem 2.3. Given an estimate of the barycenter, it is possible to compute the empirical covariance in the corresponding tangent plane, Σ̂p̂ = 1 N X xixt i (4) Using a similar construction as the Sasakian metric, see [22], the vector bundle TD ⊗ TD can be endowed with a Riemannian metric. Although we do not prove it in this paper, we are convinced that almost surely d((p̂, Σ̂p̂), (p, Σ)) −→ n→+∞ 0, where d is the Riemannian distance on TD ⊗ TD. 6 Conclusion In this paper we proposed a set of parametric families of anisotropic distribu- tions on the hyperbolic plane. The main interest of these distributions is that the covariance matrix and concentration matrix are equal. The empirical mean and covariance provide thus simple estimators of the parameters of the distribu- tion. Working with anisotropic distributions is expected to reduce the number of distributions used in mixture models and thus to reduce the computational complexity of the parameter estimation of the mixture models. On the one hand, our future work will focus on deriving convergence rates of the estimation of the covariance. On the other hand, we will study the use of these distributions in problems of radar signal classification. References 1. Pelletier, B. Kernel density estimation on Riemannian manifolds. Stat. Probab. Lett. 2005, 73, 297–304. 2. Hendriks, H. Nonparametric estimation of a probability density on a Riemannian manifold using Fourier expansions. Ann. Stat. 1990, 18, 832–849. 3. Huckemann, S.; Kim, P.; Koo, J.; Munk, A. Mobius deconvolution on the hyper- bolic plan with application to impedance density estimation. Ann. Stat. 2010, 38, 2465–2498. 4. Barbaresco, F. Robust statistical radar processing in Fréchet metric space: OS- HDR-CFAR and OS-STAP processing in siegel homogeneous bounded domains. In Proceedings of the 2011 12th International Radar Symposium (IRS), Leipzig, Germany, 7–9 Septerber 2011. 5. Barbaresco, F. Information Geometry of Covariance Matrix: Cartan-Siegel Homogeneous Bounded Domains, Mostow/Berger Fibration and Fréchet Me- dian. In Matrix Information Geometry; Bhatia, R., Nielsen, F., Eds.; Springer: Berlin/Heidelberg, Germany, 2012; pp. 199–256. 6. Barbaresco, F. Information geometry manifold of Toeplitz Hermitian positive def- inite covariance matrices: Mostow/Berger fibration and Berezin quantization of Cartan-Siegel domains. In Int. J. Emerg. Trends Signal Process. (IJETSP) 2013, 1, 1–87. 7. Cannon, J. W.; Floyd, W. J.; Kenyon, R.; and Parry, W. R. Hyperbolic Geome- try.MSRI Publications, 31, 1997. 8. Pennec, X. Intrinsic Statistics on Riemannian Manifolds: Basic Tools for Geomet- ric Measurements. In J Math Imaging Vis,25: 127, 2006 9. Kim,P. T. and Richards, D. St. P. Deconvolution Density Es- timation on the Space of Positive Definite Symmetric Matrices. In N onparametric Statistics and Mixture Models , 147–168. 2011. 10. Grigoryan, A. Heat kernel and analysis on manifolds. American mathematical so c., 47, 2012. 11. Anker, J-P.; Ostellari, P. The heat kernel on noncompact symmetric spaces. In Lie groups and symmetric spaces, AMS Transl.Ser.2, 210,27-46,2003. 12. Said, S., Bombrun, L. and Berthoumieu, Y. New Riemannian priors on the uni- variate normal model. In Entropy, 16(7), 4015-4031, 2014 13. Said, S.; Bombrun, L.; Berthoumieu, Y. and Jonathan Manton,J. Riemannian Gaussian Distributions on the Space of Symmetric Positive Definite Matrices. arXiv:1507.01760 [math.ST], 2015 14. Said, S.; Hajri, H.; Bombrun, L. and Vemuri, B.C. Gaussian distributions on Rie- mannian symmetric spaces: statistical learning with structured covariance matri- ces. arXiv:1607.06929 [math.ST], 2016 15. Le Bihan, N.; Flamant, J.; Manton, J.H. Density estimation on the rotation group using diffusive wavelets. arXiv:1512.06023, 2015 16. Chevallier, E., Barbaresco, F., and Angulo, J. Probability density estimation on the hyperbolic space applied to radar processing. In I nternational Conference on Networked Geometric Science of Information (pp. 753-761), 2015. 17. Chevallier, E.; Forget, T.; Barbaresco, F. and Angulo, J. Kernel Density Estima- tion on the Siegel Space with an Application to Radar Processing. In Entropy, 18(11), 396, 2016. 18. Chevallier,E.; Kalunga, E. and Angulo, J.; Kernel Density Estimation on Spaces of Gaussian Distributions and Symmetric Positive Definite Matrices. In SIAM J. Imaging Sci., 10(1), 191–215. 19. Sommer, S. Anisotropic distributions on manifolds: Template estimation and most probable paths. In I nternational Conference on Information Processing in Medical Imaging, Springer International Publishing, 193-204, june 2015. 20. Fletcher, P. T.; Lu, C.; Pizer, S. M. and Joshi, S. Principal geodesic analysis for the study of nonlinear statistics of shape. In IEEE transactions on medical imaging, 23(8), 995-1005, 2004. 21. Bhattacharya, R. and Patrangenaru, V. Large sample theory of intrinsic and extrinsic sample means on manifolds. I. In Annals of statistics, 1-29, 2003 22. Kappos, E. Natural Metric On Tangent Bundles. Master’s thesis, 2001