On the Statistical Capacity of Deep Generative Models

Edric Tam, David B. Dunson · 2025

Read On the Statistical Capacity of Deep Generative Models by Edric Tam, David B. Dunson (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-2501-07763.txt.

Description

Edric Tam and David Dunson examine the statistical capacity of deep generative models. Their 2025 arXiv paper questions a common hopeful assumption. Many assume big networks with enough data can sample any target distribution well. The authors build a shared framework to test that belief. It covers variational autoencoders and generative adversarial networks.

They show broad classes of these models are not universal generators. With Gaussian latent variables, outputs stay concentrated with light tails. Similar limits hold for log-concave latent distributions too. Diffusion models fall under the same verdict by a reduction step. Manifold latent spaces get parallel guarantees from geometry. The results set honest bounds on what generative models can mimic.

Catalog record

CREATOREdric Tam, David B. Dunson
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-07763

Browse the full library · About the Tower of Babel