stat.MLMay 19, 2026

Optimizing Computational-Statistical Runtime for Wasserstein Distance Estimation

Authors: Peter Matthew JacobsJeff M. Phillips

Abstract

Squared Wasserstein distance is a frequently used tool to measure discrepancy between probability distributions. This distance is typically computed between empirical measures of size nn from two underlying random samples. Unfortunately, even in lower dimensional Euclidean space problems (d{2,3})\left( d \in \{2,3\} \right), algorithms for Wasserstein distance computation with approximate or exact precision guarantees scale poorly in the runtime as a function of nn and the desired precision. In response, we consider the computational-statistical runtime, where the goal is to estimate from samples the Wasserstein distance between potentially smooth measures up to εε-additive error in expectation with respect to the sampling; we allow O(1)O(1) computational cost for collecting a sample. Towards this, we develop a Sample-Sketch-Solve paradigm where we introduce a regular cartesian grid sketch of the samples. We show that (especially under αα-Hölder smooth distributions) this can compress the data without increasing asymptotic error, and also regularizes the structure which enables faster exact algorithms. Ultimately, we approximate W22(P,Q)W_2^2(P,Q) within εε error in εmax(2,d+1+o(1)1+α)ε^{-\max(2,\frac{d+1+o(1)}{1+α})} time for 0<α<10 < α< 1 Hölder smooth distributions P,QP,Q on (0,1)d(0,1)^{d}; an optimal Θ(ε2)Θ(ε^{-2}) for α>1/2α> 1/2 when d=2d=2 and nearly optimal as α1α\to 1 when d=3d = 3.

Explore similar work

CardsList
  1. The Observable Wasserstein Distance

    May 11, 2026Edivaldo Lopes dos Santos, Leandro Vicente Mauri, Washington Mio +1Wasserstein DistanceMetric Spaces