Forward-only Diffusion Probabilistic Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Luo, Ziwei, Gustafsson, Fredrik K., Sjölund, Jens, Schön, Thomas B.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914057797238784
author Luo, Ziwei
Gustafsson, Fredrik K.
Sjölund, Jens
Schön, Thomas B.
author_facet Luo, Ziwei
Gustafsson, Fredrik K.
Sjölund, Jens
Schön, Thomas B.
contents This work presents a forward-only diffusion (FoD) approach for generative modelling. In contrast to traditional diffusion models that rely on a coupled forward-backward diffusion scheme, FoD directly learns data generation through a single forward diffusion process, yielding a simple yet efficient generative framework. The core of FoD is a state-dependent stochastic differential equation that involves a mean-reverting term in both the drift and diffusion functions. This mean-reversion property guarantees the convergence to clean data, naturally simulating a stochastic interpolation between source and target distributions. More importantly, FoD is analytically tractable and is trained using a simple stochastic flow matching objective, enabling a few-step non-Markov chain sampling during inference. The proposed FoD model, despite its simplicity, achieves state-of-the-art performance on various image restoration tasks. Its general applicability on image-conditioned generation is also demonstrated via qualitative results on image-to-image translation. Our code is available at https://github.com/Algolzw/FoD.
format Preprint
id arxiv_https___arxiv_org_abs_2505_16733
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Forward-only Diffusion Probabilistic Models
Luo, Ziwei
Gustafsson, Fredrik K.
Sjölund, Jens
Schön, Thomas B.
Machine Learning
This work presents a forward-only diffusion (FoD) approach for generative modelling. In contrast to traditional diffusion models that rely on a coupled forward-backward diffusion scheme, FoD directly learns data generation through a single forward diffusion process, yielding a simple yet efficient generative framework. The core of FoD is a state-dependent stochastic differential equation that involves a mean-reverting term in both the drift and diffusion functions. This mean-reversion property guarantees the convergence to clean data, naturally simulating a stochastic interpolation between source and target distributions. More importantly, FoD is analytically tractable and is trained using a simple stochastic flow matching objective, enabling a few-step non-Markov chain sampling during inference. The proposed FoD model, despite its simplicity, achieves state-of-the-art performance on various image restoration tasks. Its general applicability on image-conditioned generation is also demonstrated via qualitative results on image-to-image translation. Our code is available at https://github.com/Algolzw/FoD.
title Forward-only Diffusion Probabilistic Models
topic Machine Learning
url https://arxiv.org/abs/2505.16733