Intrinsic Wasserstein Rates for Score-Based Generative Models on Smooth Manifolds

Guoji Fu, Taiji Suzuki, Wee Sun Lee, Atsushi Nitanda · 2026

Read Intrinsic Wasserstein Rates for Score-Based Generative Models on Smooth Manifolds by Guoji Fu, Taiji Suzuki, Wee Sun Lee, Atsushi Nitanda (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15822.txt.

Introduction

Fu and three colleagues prove intrinsic Wasserstein rates for score-based generative models on smooth manifolds. Such models train in high-dimensional ambient space, yet much data lives on low-dimensional curved structures. For compact manifolds with Hölder densities, their estimator attains that Wasserstein-1 sample rate, up to log factors. The proof splits score fitting into large-noise and small-noise regimes and builds projection coordinates with a ReLU network.

Catalog record

CREATORGuoji Fu, Taiji Suzuki, Wee Sun Lee, Atsushi Nitanda
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE432547
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.15822

Browse the full library · About the Tower of Babel