Saved in:
Bibliographic Details
Main Authors: Jiang, Lei, Ge, Wen, Cariou-Kotlarek, Niels, Yi, Mingxuan, Chen, Po-Yu, Yang, Lingyi, Buet-Golfouse, Francois, Mittal, Gaurav, Ni, Hao
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.16939
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916913636966400
author Jiang, Lei
Ge, Wen
Cariou-Kotlarek, Niels
Yi, Mingxuan
Chen, Po-Yu
Yang, Lingyi
Buet-Golfouse, Francois
Mittal, Gaurav
Ni, Hao
author_facet Jiang, Lei
Ge, Wen
Cariou-Kotlarek, Niels
Yi, Mingxuan
Chen, Po-Yu
Yang, Lingyi
Buet-Golfouse, Francois
Mittal, Gaurav
Ni, Hao
contents Diffusion models have achieved state-of-the-art results in generative modelling but remain computationally intensive at inference time, often requiring thousands of discretization steps. To this end, we propose Sig-DEG (Signature-based Differential Equation Generator), a novel generator for distilling pre-trained diffusion models, which can universally approximate the backward diffusion process at a coarse temporal resolution. Inspired by high-order approximations of stochastic differential equations (SDEs), Sig-DEG leverages partial signatures to efficiently summarize Brownian motion over sub-intervals and adopts a recurrent structure to enable accurate global approximation of the SDE solution. Distillation is formulated as a supervised learning task, where Sig-DEG is trained to match the outputs of a fine-resolution diffusion model on a coarse time grid. During inference, Sig-DEG enables fast generation, as the partial signature terms can be simulated exactly without requiring fine-grained Brownian paths. Experiments demonstrate that Sig-DEG achieves competitive generation quality while reducing the number of inference steps by an order of magnitude. Our results highlight the effectiveness of signature-based approximations for efficient generative modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2508_16939
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Sig-DEG for Distillation: Making Diffusion Models Faster and Lighter
Jiang, Lei
Ge, Wen
Cariou-Kotlarek, Niels
Yi, Mingxuan
Chen, Po-Yu
Yang, Lingyi
Buet-Golfouse, Francois
Mittal, Gaurav
Ni, Hao
Machine Learning
Probability
Diffusion models have achieved state-of-the-art results in generative modelling but remain computationally intensive at inference time, often requiring thousands of discretization steps. To this end, we propose Sig-DEG (Signature-based Differential Equation Generator), a novel generator for distilling pre-trained diffusion models, which can universally approximate the backward diffusion process at a coarse temporal resolution. Inspired by high-order approximations of stochastic differential equations (SDEs), Sig-DEG leverages partial signatures to efficiently summarize Brownian motion over sub-intervals and adopts a recurrent structure to enable accurate global approximation of the SDE solution. Distillation is formulated as a supervised learning task, where Sig-DEG is trained to match the outputs of a fine-resolution diffusion model on a coarse time grid. During inference, Sig-DEG enables fast generation, as the partial signature terms can be simulated exactly without requiring fine-grained Brownian paths. Experiments demonstrate that Sig-DEG achieves competitive generation quality while reducing the number of inference steps by an order of magnitude. Our results highlight the effectiveness of signature-based approximations for efficient generative modeling.
title Sig-DEG for Distillation: Making Diffusion Models Faster and Lighter
topic Machine Learning
Probability
url https://arxiv.org/abs/2508.16939