跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2506.03043 · v2

On accuracy of density estimation in generative models based on stochastic optimal control

首次提交 2025-06-04 00:26
版本更新 2026-11-17 01:19
雷达观测 2026-11-17 16:10
arXiv ↗PDF ↗

摘要

We study the statistical accuracy of Schrödinger bridge models based on an Ornstein-Uhlenbeck reference process. Given a known, compactly supported initial distribution and independent samples from an unknown target distribution, we estimate a right Schrödinger log-potential by minimizing empirical Kullback-Leibler (KL) risk over a prescribed function class. For bounded sub-Gaussian target densities, potentially with unbounded support, we establish a non-asymptotic high-probability oracle inequality for the KL-divergence between the target and estimated terminal marginals. Under additional two-sided Gaussian bounds and smoothness assumptions on the target log-density, we derive approximation guarantees for sparse ReLU networks and establish a KL convergence rate of O~(n−2s/(2s+d))\widetilde{O}(n^{-2s / (2s + d)}) for exact empirical risk minimization. This matches, up to logarithmic factors, the classical nonparametric KL-risk convergence exponent, while allowing target densities on Rd\mathbb R^d whose tails are controlled by Gaussian envelopes.

作者备注:62 pages, 2 figures

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 · 2026-11-17 01:19 · 最新观测版本

On accuracy of density estimation in generative models based on stochastic optimal control

We study the statistical accuracy of Schrödinger bridge models based on an Ornstein-Uhlenbeck reference process. Given a known, compactly supported initial distribution and independent samples from an unknown target distribution, we estimate a right Schrödinger log-potential by minimizing empirical Kullback-Leibler (KL) risk over a prescribed function class. For bounded sub-Gaussian target densities, potentially with unbounded support, we establish a non-asymptotic high-probability oracle inequality for the KL-divergence between the target and estimated terminal marginals. Under additional two-sided Gaussian bounds and smoothness assumptions on the target log-density, we derive approximation guarantees for sparse ReLU networks and establish a KL convergence rate of O~(n−2s/(2s+d))\widetilde{O}(n^{-2s / (2s + d)}) for exact empirical risk minimization. This matches, up to logarithmic factors, the classical nonparametric KL-risk convergence exponent, while allowing target densities on Rd\mathbb R^d whose tails are controlled by Gaussian envelopes.

查看此版本解读 →查看此版本原文 ↗