Disentanglement in T-space for Faster and Distributed Training of Diffusion Models with Fewer Latent-states

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gupta, Samarth, Gadde, Raghudeep, Chen, Rui, Martinez, Aleix M.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913998250704896
author Gupta, Samarth
Gadde, Raghudeep
Chen, Rui
Martinez, Aleix M.
author_facet Gupta, Samarth
Gadde, Raghudeep
Chen, Rui
Martinez, Aleix M.
contents We challenge a fundamental assumption of diffusion models, namely, that a large number of latent-states or time-steps is required for training so that the reverse generative process is close to a Gaussian. We first show that with careful selection of a noise schedule, diffusion models trained over a small number of latent states (i.e. $T \sim 32$) match the performance of models trained over a much large number of latent states ($T \sim 1,000$). Second, we push this limit (on the minimum number of latent states required) to a single latent-state, which we refer to as complete disentanglement in T-space. We show that high quality samples can be easily generated by the disentangled model obtained by combining several independently trained single latent-state models. We provide extensive experiments to show that the proposed disentangled model provides 4-6$\times$ faster convergence measured across a variety of metrics on two different datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2508_14413
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Disentanglement in T-space for Faster and Distributed Training of Diffusion Models with Fewer Latent-states
Gupta, Samarth
Gadde, Raghudeep
Chen, Rui
Martinez, Aleix M.
Machine Learning
Computer Vision and Pattern Recognition
We challenge a fundamental assumption of diffusion models, namely, that a large number of latent-states or time-steps is required for training so that the reverse generative process is close to a Gaussian. We first show that with careful selection of a noise schedule, diffusion models trained over a small number of latent states (i.e. $T \sim 32$) match the performance of models trained over a much large number of latent states ($T \sim 1,000$). Second, we push this limit (on the minimum number of latent states required) to a single latent-state, which we refer to as complete disentanglement in T-space. We show that high quality samples can be easily generated by the disentangled model obtained by combining several independently trained single latent-state models. We provide extensive experiments to show that the proposed disentangled model provides 4-6$\times$ faster convergence measured across a variety of metrics on two different datasets.
title Disentanglement in T-space for Faster and Distributed Training of Diffusion Models with Fewer Latent-states
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.14413