A Manifold Two-Sample Test Study: Integral Probability Metric with Neural Networks
Jie Wang, Minshuo Chen, Tuo Zhao, Wenjing Liao, Yao Xie

TL;DR
This paper introduces new two-sample tests based on integral probability metrics that leverage neural networks to efficiently detect distribution differences in high-dimensional data supported on low-dimensional manifolds, with theoretical guarantees.
Contribution
The paper proposes neural network-based IPM tests for high-dimensional manifold-supported data, achieving optimal risk bounds and adaptivity to intrinsic data structure.
Findings
Neural network IPM test matches the risk bounds of H"older IPM.
Tests are adaptive to intrinsic low-dimensional structure.
Proposed methods are computationally feasible for high-dimensional data.
Abstract
Two-sample tests are important areas aiming to determine whether two collections of observations follow the same distribution or not. We propose two-sample tests based on integral probability metric (IPM) for high-dimensional samples supported on a low-dimensional manifold. We characterize the properties of proposed tests with respect to the number of samples and the structure of the manifold with intrinsic dimension . When an atlas is given, we propose two-step test to identify the difference between general distributions, which achieves the type-II risk in the order of . When an atlas is not given, we propose H\"older IPM test that applies for data distributions with -H\"older densities, which achieves the type-II risk in the order of . To mitigate the heavy computation burden of evaluating the H\"older IPM, we approximate the…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsStatistical Methods and Inference · Advanced Statistical Methods and Models · Statistical Methods and Bayesian Inference
