Masked Diffusion Models as Energy Minimization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chen, Sitong, Nie, Shen, Sun, Jiacheng, Feng, Zijin, Li, Zhenguo, Wen, Ji-Rong, Li, Chongxuan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917357332463616
author Chen, Sitong
Nie, Shen
Sun, Jiacheng
Feng, Zijin
Li, Zhenguo
Wen, Ji-Rong
Li, Chongxuan
author_facet Chen, Sitong
Nie, Shen
Sun, Jiacheng
Feng, Zijin
Li, Zhenguo
Wen, Ji-Rong
Li, Chongxuan
contents We present a systematic theoretical framework that interprets masked diffusion models (MDMs) as solutions to energy minimization problems in discrete optimal transport. Specifically, we prove that three distinct energy formulations--kinetic, conditional kinetic, and geodesic energy--are mathematically equivalent under the structure of MDMs, and that MDMs minimize all three when the mask schedule satisfies a closed-form optimality condition. This unification not only clarifies the theoretical foundations of MDMs, but also motivates practical improvements in sampling. By parameterizing interpolation schedules via Beta distributions, we reduce the schedule design space to a tractable 2D search, enabling efficient post-training tuning without model modification. Experiments on synthetic and real-world benchmarks demonstrate that our energy-inspired schedules outperform hand-crafted baselines, particularly in low-step sampling settings.
format Preprint
id arxiv_https___arxiv_org_abs_2509_13866
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Masked Diffusion Models as Energy Minimization
Chen, Sitong
Nie, Shen
Sun, Jiacheng
Feng, Zijin
Li, Zhenguo
Wen, Ji-Rong
Li, Chongxuan
Machine Learning
Artificial Intelligence
Computation and Language
We present a systematic theoretical framework that interprets masked diffusion models (MDMs) as solutions to energy minimization problems in discrete optimal transport. Specifically, we prove that three distinct energy formulations--kinetic, conditional kinetic, and geodesic energy--are mathematically equivalent under the structure of MDMs, and that MDMs minimize all three when the mask schedule satisfies a closed-form optimality condition. This unification not only clarifies the theoretical foundations of MDMs, but also motivates practical improvements in sampling. By parameterizing interpolation schedules via Beta distributions, we reduce the schedule design space to a tractable 2D search, enabling efficient post-training tuning without model modification. Experiments on synthetic and real-world benchmarks demonstrate that our energy-inspired schedules outperform hand-crafted baselines, particularly in low-step sampling settings.
title Masked Diffusion Models as Energy Minimization
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2509.13866