Two-Sample Testing via Path-based Inference
Abstract
Modern deep generative models are primarily studied for their ability to generate realistic samples, yet the generative dynamics they learn can also serve as objects of statistical inference. We develop this idea for two-sample testing, the problem of deciding whether the same distribution generated two finite datasets. Using stochastic interpolants, we connect both distributions to a shared Gaussian bottleneck, so that each half of the resulting path is a Gaussian channel acting on a single population. We prove that the null hypothesis holds if and only if the population denoiser, or equivalently, the velocity fields of the two halves, coincide at any single noise level, which amounts to a reflection symmetry of the path about the bottleneck. Deviations from this symmetry yield a continuum of two-sample witnesses, which we estimate via held-out regression risks on learned denoisers and velocities and aggregate along the path; under an information-theoretic weighting, the aggregated discrepancy equals the Jeffreys divergence between the noise-smoothed distributions. Calibrating the resulting statistics by permutation yields tests that are valid in finite samples for any trained networks and consistent when the fields are learned accurately. On a synthetic benchmark and three image benchmarks, the proposed tests improve power over the strongest baseline by up to 33 percentage points at an equal total sample budget, with the best choice of regression representation and path weighting depending on the data modality. These results show that generative paths provide a principled representation for statistical testing, extending stochastic-interpolant models beyond generation.
Related
- Strong Approximations for Multivariate Integrated Empirical Processes: Two-Sample Testing, Change-Point Detection, and Inference with Estimated Parameters
- Data-Matrix-Based Overall Goodness-of-Fit Testing for Confirmatory Factor Analysis: In-Sample Prediction Error for Observed Variables
- OLS Testing with Predictors Scaled to Have Unit Sample Variance, and Similar Results for a Robust Regression Estimator
- Sample-Efficient Multiple Testing with Adaptive Data Collection
- Inference for Causal Path-Specific Effects in Longitudinal Studies with Repeated Treatments and Multiple Mediators
BibTeX
@misc{english2026twosample,
title = {{Two-Sample Testing via Path-based Inference}},
author = {Eshant English and Wei-Cheng Lai and Yanfeng Yang and Kenji Fukumizu and Taiji Suzuki and Christoph Lippert},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.05684},
eprint = {2610.05684},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.05684},
}
RIS
TY - UNPB
TI - Two-Sample Testing via Path-based Inference
AU - Eshant English
AU - Wei-Cheng Lai
AU - Yanfeng Yang
AU - Kenji Fukumizu
AU - Taiji Suzuki
AU - Christoph Lippert
PY - 2026
DA - 2026/10/05
PB - arXiv
DO - 10.48550/arxiv.2610.05684
UR - https://arxiv.org/abs/2610.05684
AB - Modern deep generative models are primarily studied for their ability to generate realistic samples, yet the generative dynamics they learn can also serve as objects of statistical inference. We develop this idea for two-sample testing, the problem of deciding whether the same distribution generated two finite datasets. Using stochastic interpolants, we connect both distributions to a shared Gaussian bottleneck, so that each half of the resulting path is a Gaussian channel acting on a single population. We prove that the null hypothesis holds if and only if the population denoiser, or equivalently, the velocity fields of the two halves, coincide at any single noise level, which amounts to a reflection symmetry of the path about the bottleneck. Deviations from this symmetry yield a continuum of two-sample witnesses, which we estimate via held-out regression risks on learned denoisers and velocities and aggregate along the path; under an information-theoretic weighting, the aggregated discrepancy equals the Jeffreys divergence between the noise-smoothed distributions. Calibrating the resulting statistics by permutation yields tests that are valid in finite samples for any trained networks and consistent when the fields are learned accurately. On a synthetic benchmark and three image benchmarks, the proposed tests improve power over the strongest baseline by up to 33 percentage points at an equal total sample budget, with the best choice of regression representation and path weighting depending on the data modality. These results show that generative paths provide a principled representation for statistical testing, extending stochastic-interpolant models beyond generation.
ER -