Generative Learning of Densities on Manifolds

Dimitris G. Giovanis, Ellis Crabtree, Roger G. Ghanem, Ioannis G. Kevrekidis · 2025

Read Generative Learning of Densities on Manifolds by Dimitris G. Giovanis, Ellis Crabtree, Roger G. Ghanem, Ioannis G. Kevrekidis (2025) 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-2503-03963.txt.

Description

Dimitris Giovanis and co-authors present a way to generate new samples from data that live on curved surfaces, in a 2025 paper. Much real data, though stored in many dimensions, truly varies along a hidden low-dimensional shape.

The approach first maps the data to that hidden space with a tool called Diffusion Maps. It then learns the density there, either with a score-based diffusion model or by solving a stochastic equation. A second mapping lifts fresh samples back to the full space. Tests on a benchmark task and on a material with structure at two scales showed the method works. It opens sampling for complex data whose shape is not known ahead.

Catalog record

CREATORDimitris G. Giovanis, Ellis Crabtree, Roger G. Ghanem, Ioannis G. Kevrekidis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-03963

Browse the full library · About the Tower of Babel