跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2404.00551 · v2

Gaussian-Channel Regularity and Distribution Learning with Flow Matching

首次提交 2024-03-31 11:39
版本更新 2026-11-16 23:41
雷达观测 2026-11-17 16:10
arXiv ↗PDF ↗

摘要

We study distribution learning by flow matching with linear interpolation from a Gaussian source under Gaussian-channel regularity, a condition on the target distribution. For an integer r≥2r\geq2, we establish spatial regularity of order rr for the population velocity, together with temporal Lipschitz continuity on the full interval and local time analyticity before the terminal time. We use these properties to construct a finite multiscale ReLU sieve with a spatial Lipschitz bound independent of the truncation radius. For an exact minimizer of the empirical flow matching risk over this sieve, we obtain an upper bound on the expected Wasserstein-22 error of order n−r/(d+2r)n^{-r/(d+2r)}, up to a logarithmic factor, where nn is the sample size and dd is the dimension. The bound includes Euler discretization up to time one, without early stopping. The conditions hold at r=2r=2 for targets with strongly convex C2C^2 potentials and two-sided Hessian bounds. They also hold at every fixed order r≥2r\geq2 for distributions obtained by convolving a law of bounded support with a nondegenerate isotropic Gaussian. For sufficiently large fixed constants defining the target class, we prove a minimax lower bound of order max⁡{n−1/2,n−(r+1)/(d+2r)}\max\{n^{-1/2},n^{-(r+1)/(d+2r)}\}.

作者备注:56 pages, 2 tables

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v2,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v2 · 2026-11-16 23:41 · 最新观测版本

Gaussian-Channel Regularity and Distribution Learning with Flow Matching

We study distribution learning by flow matching with linear interpolation from a Gaussian source under Gaussian-channel regularity, a condition on the target distribution. For an integer r≥2r\geq2, we establish spatial regularity of order rr for the population velocity, together with temporal Lipschitz continuity on the full interval and local time analyticity before the terminal time. We use these properties to construct a finite multiscale ReLU sieve with a spatial Lipschitz bound independent of the truncation radius. For an exact minimizer of the empirical flow matching risk over this sieve, we obtain an upper bound on the expected Wasserstein-22 error of order n−r/(d+2r)n^{-r/(d+2r)}, up to a logarithmic factor, where nn is the sample size and dd is the dimension. The bound includes Euler discretization up to time one, without early stopping. The conditions hold at r=2r=2 for targets with strongly convex C2C^2 potentials and two-sided Hessian bounds. They also hold at every fixed order r≥2r\geq2 for distributions obtained by convolving a law of bounded support with a nondegenerate isotropic Gaussian. For sufficiently large fixed constants defining the target class, we prove a minimax lower bound of order max⁡{n−1/2,n−(r+1)/(d+2r)}\max\{n^{-1/2},n^{-(r+1)/(d+2r)}\}.

查看此版本解读 →查看此版本原文 ↗