{"id":"arxiv/2608.06560","url":"https://arxiv.org/abs/2608.06560","tags":["math.OC","math.ST","stat.ML"],"score":0,"title":"Wasserstein Mahalanobis Distances for Recovering Latent Geometry","domain":"math.OC","source":"arxiv","authors":["Chuxiangbo Wang","Shiying Li","Caroline Moosmüller"],"reviews":0,"subject":"math.OC","summary":"The Mahalanobis distance is a fundamental covariance-adapted metric for multivariate data and plays a central role in recovering latent geometry from nonlinear observations. We extend this principle from vector-valued data to probability measures by introducing a Wasserstein Mahalanobis distance. Our construction replaces Euclidean displacement vectors with optimal transport displacement fields and local covariance matrices with covariance operators defined on Wasserstein tangent spaces. We show that this construction inherits the geometry-recovery property underlying nonlinear independent…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-06","reactions":{}}