Deconvolution for the Wasserstein Metric and Geometric Inference
نویسندگان
چکیده
Recently, [4] have dened a distance function to measures to answer geometric inference problems in a probabilistic setting. According to their result, the topological properties of a shape can be recovered by using the distance to a known measure ν, if ν is close enough to a measure µ concentrated on this shape. Here, close enough means that the Wasserstein distance W 2 between µ and ν is suciently small. Given a point cloud, a natural candidate for ν is the empirical measure µ n. Nevertheless, in many situations the data points are not located on the geometric shape but in the neighborhood of it, and µ n can be too far from µ. In a deconvolution framework, we consider a slight modication of the classical kernel deconvolution estimator, and we give a consistency result and rates of convergence for this estimator. Some simulated experiments illustrate the deconvolution method and its application to geometric inference on various shapes and with various noise distributions. Résumé : La notion de fonction distance à une mesure récemment introduite dans [4] permet de répondre à des problèmes d'inférence géométrique dans un cadre probabiliste : les propriétés topologiques d'un compact K ⊂ R d peuvent être estimées à l'aide de la fonction distance à une mesure de probabilité connue ν si celle-ci se trouve susamment proche (au sens de la distance de Wasserstein W 2) d'une mesure µ dont K est le support. En pratique lorsque les observations sont corrompues par du bruit, la mesure empirique associée aux observations n'est généralement pas assez proche de µ pour pouvoir être utilisée directement. Dans cet article, on propose une solution à ce problème en considérant un modèle de convolution pour lequel la loi du bruit est supposée connue. On considère une variante de l'estimateur par noyau de déconvolution classique dont on établit la consistence et des vitesses de convergence. On illustre la méthode proposée et ses applications en inférence géométrique sur diérentes formes géométriques et diérentes distributions de bruit sur les observations.
منابع مشابه
Minimax rates of convergence for Wasserstein deconvolution with supersmooth errors in any dimension
The subject of this paper is the estimation of a probability measure on R from data observed with an additive noise, under the Wasserstein metric of order p (with p ≥ 1). We assume that the distribution of the errors is known and belongs to a class of supersmooth distributions, and we give optimal rates of convergence for the Wasserstein metric of order p. In particular, we show how to use the ...
متن کاملMetric Currents and Geometry of Wasserstein Spaces
We investigate some geometric aspects of Wasserstein spaces through the continuity equation as worked out in mass transportation theory. By defining a suitable homology on the flat torus T, we prove that the space Pp(T) has non-trivial homology in a metric sense. As a byproduct of the developed tools, we show that every parametrization of a Mather’s minimal measure on T corresponds to a mass mi...
متن کاملImage Labeling Based on Graphical Models Using Wasserstein Messages and Geometric Assignment
We introduce a novel approach to Maximum A Posteriori inference based on discrete graphical models. By utilizing local Wasserstein distances for coupling assignment measures across edges of the underlying graph, a given discrete objective function is smoothly approximated and restricted to the assignment manifold. A corresponding multiplicative update scheme combines in a single process (i) geo...
متن کاملROBUSTNESS OF THE TRIPLE IMPLICATION INFERENCE METHOD BASED ON THE WEIGHTED LOGIC METRIC
This paper focuses on the robustness problem of full implication triple implication inference method for fuzzy reasoning. First of all, based on strong regular implication, the weighted logic metric for measuring distance between two fuzzy sets is proposed. Besides, under this metric, some robustness results of the triple implication method are obtained, which demonstrates that the triple impli...
متن کاملOn Wasserstein Geometry of the Space of Gaussian Measures
Abstract. The space which consists of measures having finite second moment is an infinite dimensional metric space endowed with Wasserstein distance, while the space of Gaussian measures on Euclidean space is parameterized by mean and covariance matrices, hence a finite dimensional manifold. By restricting to the space of Gaussian measures inside the space of probability measures, we manage to ...
متن کامل