Synchronizing Probability Measures on Rotations via Optimal Transport

Mayer control problem with probabilistic uncertainty on initial positions

Let μ∈ P2(Rd), where P2(Rd) denotes the space of square integrable probability measures, and consider a Borel-measurable function Φ:P2(Rd)→R. In this paper we develop an antithetic Monte Carlo estimator (A-MLMC) for Φ(μ), which achieves sharp error bound under mild regularity assumptions. The estimator takes as input the empirical laws μN=1N∑ i=1NδX i, where (a) (Xi)i=1N is a sequence of i.i.d. samples from μ or (b) (Xi)i=1N is a system of interacting particles (diffusions) corresponding to a McKean–Vlasov stochastic differential equation (McKV-SDE). Each case requires a separate analysis. For a mean-field particle system, we also consider the empirical law induced by its Euler discretisation which gives a fully implementable algorithm. As by-products of our analysis, we establish a dimension-independent rate of uniform strong propagation of chaos, as well as an L2 estimate of the antithetic difference for i.i.d. random variables corresponding to general functionals defined on the space of probability measures.

In this short communication we prove that the subspace Pn,n−1(X)of all probability measures P(X), whose supports consist of exactly n points is an (n−1)-dimensional topological manifold. A number of subspaces of the space of all probability measures having infinite dimension in the sense of dim, which are manifolds, are identified. We also consider individual subsets of the infinite compact set X, on which the space of probability measures is homotopy dense in the entire space. Three theorems on the topological properties of manifolds—subspaces of homotopy dense probability measures in the space of probability measures with finite supports on a compactum—are formulated and proven, and special cases of finite and infinite compactums are considered.

The existence of an inverse limit of an inverse system of (probability) measure spaces has been investigated since the very beginning of the modern probability theory. Results from Kolmogorov [11], Bochner [2], Choksi [6], Metivier [15], Bourbaki [4], Mallory and Sion [12] among others have paved the way of the deep understanding of this problem. All the above results, however, call for some topological concepts, or at least the ones which are closely related topological ones. In this paper we investigate purely measurable inverse systems of (probability) measure spaces, and give a sufficient condition for the existence of a unique inverse limit. An example for the considered purely measurable inverse systems of (probability) measure spaces is also given.

This paper presents primal heuristics for the computation of Wasserstein Barycenters of a given set of discrete probability measures. The computation of a Wasserstein Barycenter is formulated as an optimization problem over the space of discrete probability measures. In practice, the barycenter is a discrete probability measure which minimizes the sum of the pairwise Wasserstein distances between the barycenter itself and each input measure. While this problem can be formulated using Linear Programming techniques, it remains a challenging problem due to the size of real-life instances. In this paper, we propose simple but efficient primal heuristics, which exploit the properties of the optimal plan obtained while computing the Wasserstein Distance between a pair of probability measures. In order to evaluate the proposed primal heuristics, we have performed extensive computational tests using random Gaussian distributions, the MNIST handwritten digit dataset, and the Fashion MNIST dataset introduced by Zalando. We also used Translated MNIST, a modification of MNIST which contains original images, rescaled randomly and translated into a larger image. We compare the barycenters computed by our heuristics with the exact solutions obtained with a commercial Linear Programming solver, and with a state-of-the-art algorithm based on Gaussian convolutions. Our results show that the proposed heuristics yield in very short run time and an average optimality gap significantly smaller than 1%.

The aim of this paper is to address optimality of stochastic control strategies via dynamic programming subject to total variation distance ambiguity on the conditional distribution of the controlled process. We formulate the stochastic control problem using minimax theory, in which the control minimizes the payoff while the conditional distribution, from the total variation distance set, maximizes it. First, we investigate the maximization of a linear functional on the space of probability measures on abstract spaces, among those probability measures which are within a total variation distance from a nominal probability measure, and then we give the maximizing probability measure in closed form. Second, we utilize the solution of the maximization to solve minimax stochastic control with deterministic control strategies, under a Markovian and a non-Markovian assumption, on the conditional distributions of the controlled process. The results of this part include (1) minimax optimization subject to total va...

In his classical argument, Rao derives the Riemannian distance corresponding to the Fisher metric using a mapping between the space of positive measures and Euclidean space. He obtains the Hellinger distance on the full space of measures and the Fisher distance on the subset of probability measures. In order to highlight the interplay between Fisher theory and quantum information theory, we extend this construction to the space of positive-definite Hermitian matrices using Riemannian submersions and quotient manifolds. The analog of the Hellinger distance turns out to be the Bures–Wasserstein (BW) distance, a distance measure appearing in optimal transport, quantum information, and optimisation theory. First we present an existing derivation of the Riemannian metric and geodesics associated with this distance. Subsequently, we present a novel derivation of the Riemannian distance and geodesics for this metric on the subset of trace-one matrices, analogous to the Fisher distance for probability measures.

Clinical predictive models are vulnerable to degradation in performance due to changes in the distribution of the data (distribution divergence) at application time. Significant reductions in model performance can lead to suboptimal medical decisions and harm to patients. Distribution divergence in healthcare data can arise from changes in medical practice, patient demographics, equipment, and measurement standards. However, estimating model performance at application time is challenging when labels are not readily available, which is often the case in healthcare. One solution to this challenge is to develop unsupervised methods of measuring distribution divergence that are predictive of changes in performance of clinical models. In this article, we investigate the capability of divergence metrics that can be computed without labels in estimating model performance under conditions of distribution divergence. In particular, we examine two popular integral probability metrics, i.e., Wasserstein distance and maximum mean discrepancy, and measure their correlation with model performance in the context of predicting mortality and prolonged stay in the intensive care unit (ICU). When models were trained on data from one hospital’s ICU and assessed on data from ICUs in other hospitals, model performance was significantly correlated with the degree of divergence across hospitals as measured by the distribution divergence metrics. Moreover, regression models could predict model performance from divergence metrics with small errors.

A definition of Lévy processes with values in the space of probability measures was introduced by Shiga and Tanaka (Electronic J. Prob. 11 (2006)). It is shown that the Lévy process with values in the space of probability measures in law has a modification satisfying a certain condition. The modification is a Lévy process in the sense of Shiga and Tanaka. The Lévy-Itô decomposition of sample paths of the Lévy process satisfying the condition is derived.

On the optimal rate for the convergence problem in mean field control

We introduce the method of Geodesic Principal Component Analysis (GPCA) on the space of probability measures on the line, with finite second moment, endowed with the Wasserstein metric. We discuss the advantages of this approach, over a standard functional PCA of probability densities in the Hilbert space of square-integrable functions. We establish the consistency of the method by showing that the empirical GPCA converges to its population counterpart, as the sample size tends to infinity. A key property in the study of GPCA is the isometry between the Wasserstein space and a closed convex subset of the space of square-integrable functions, with respect to an appropriate measure. Therefore, we consider the general problem of PCA in a closed convex subset of a separable Hilbert space, which serves as basis for the analysis of GPCA and also has interest in its own right. We provide illustrative examples on simple statistical models, to show the benefits of this approach for data analysis. The method is also applied to a real dataset of population pyramids.

Nonparametric and nonlinear measures of statistical dependence between pairs of random variables are important tools in modern data analysis. In particular the emergence of large data sets can now support the relaxation of linearity assumptions implicit in traditional association scores such as correlation. Here we describe a Bayesian nonparametric procedure that leads to a tractable, explicit and analytic quantification of the relative evidence for dependence vs independence. Our approach uses Polya tree priors on the space of probability measures which can then be embedded within a decision theoretic test for dependence. Polya tree priors can accommodate known uncertainty in the form of the underlying sampling distribution and provides an explicit posterior probability measure of both dependence and independence. Well known advantages of having an explicit probability measure include: easy comparison of evidence across different studies; encoding prior information; quantifying changes in dependence across different experimental conditions, and the integration of results within formal decision analysis.

Global solutions to an equation for mutation process in bacteria and its preservation of positive supports

We present a self-contained and comprehensive study of the Fisher-Rao space of matrix-valued non-commutative probability measures, and of the related Hellinger space. Our non-commutative Fisher-Rao space is a natural generalization of the classical commutative Fisher-Rao space of probability measures and of the Bures-Wasserstein space of Hermitian positive-definite matrices. We introduce and justify a canonical entropy on the non-commutative Fisher-Rao space, which differs from the von Neumann entropy. We consequently derive the analogues of the heat flow, of the Fisher information, and of the dynamical Schrödinger problem. We show the \(\Gamma \)-convergence of the \(\varepsilon \)-Schrödinger problem towards the geodesic problem for the Fisher-Rao space, and, as a byproduct, the strict geodesic convexity of the entropy.

Domain adaptation is a major area of interest in intelligent equipment maintenance and fault diagnosis in recent years. Traditional machine/deep-learning-based fault diagnosis methods assume that the source and target domains share the same distribution, which may fail and lead to catastrophic damages. Many domain adaptation-based fault diagnosis methods have been proposed to address the domain shift problem. However, most of them only align global domain distributions and ignore class relationships between domains, which leads to a decline in diagnostic performance. To overcome this deficiency, a dual-view alignment-based domain adaptation network (DVADAN) for fault diagnosis is proposed in this paper. Specifically, the proposed dual-view alignment, consisting of a global (marginal) alignment constructed with maximum mean discrepancy and a local (conditional) alignment calculating the class-centers by Wasserstein distance, is developed to reduce domain distribution discrepancy. Extensive experiments on two test rigs validated the effectiveness of the proposed DVADAN and showed its superiority over state-of-art fault diagnosis methods.