Two-sample Test with Kernel Projected Wasserstein Distance
arXiv:2102.06449
Abstract
We develop a kernel projected Wasserstein distance for the two-sample test, an essential building block in statistics and machine learning: given two sets of samples, to determine whether they are from the same distribution. This method operates by finding the nonlinear mapping in the data space which maximizes the distance between projected distributions. In contrast to existing works about projected Wasserstein distance, the proposed method circumvents the curse of dimensionality more efficiently. We present practical algorithms for computing this distance function together with the non-asymptotic uncertainty quantification of empirical estimates. Numerical examples validate our theoretical results and demonstrate good performance of the proposed method.
arXiv comment50 pages, 11 figures, 3 tables, accepted as oral presentation in AISTATS-22
References in corpus (13)
- Global rates of convergence for nonconvex optimization on manifolds
- Generative Models and Model Criticism via Optimized Maximum Mean Discrepancy
- Sample Complexity of Sinkhorn divergences
- Convergence and Concentration of Empirical Measures under Wasserstein Distance in Unbounded Functional Spaces
- A Kernel Test of Goodness of Fit
- Learning Deep Kernels for Non-Parametric Two-Sample Tests
- Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distances
- Statistical bounds for entropic optimal transport: sample complexity and the central limit theorem
- Principal Differences Analysis: Interpretable Characterization of Differences between Distributions
- Two-sample Test using Projected Wasserstein Distance
- Vector Transport-Free SVRG with General Retraction for Riemannian Optimization: Complexity Analysis and Practical Implementation
- Asymptotic Guarantees for Generative Modeling Based on the Smooth Wasserstein Distance
- Neural Tangent Kernel Maximum Mean Discrepancy