Score-of-Mixture Training: One-Step Generative Model Training Made Simple via Score Estimation of Mixture Distributions
Abstract
We propose Score-of-Mixture Training (SMT), a novel framework for training one-step generative models by minimizing a class of divergences called the $\alpha$-skew Jensen–Shannon divergence. At its core, SMT estimates the score of mixture distributions between real and fake samples across multiple noise levels. Similar to consistency models, our approach supports both training from scratch (SMT) and distillation using a pretrained diffusion model, which we call Score-of-Mixture Distillation (SMD). It is simple to implement, requires minimal hyperparameter tuning, and ensures stable training. Experiments on CIFAR-10 and ImageNet 64$\times$64 show that SMT/SMD are competitive with and can even outperform existing methods.
Cite
Text
Jayashankar et al. "Score-of-Mixture Training: One-Step Generative Model Training Made Simple via Score Estimation of Mixture Distributions." Proceedings of the 42nd International Conference on Machine Learning, 2025.Markdown
[Jayashankar et al. "Score-of-Mixture Training: One-Step Generative Model Training Made Simple via Score Estimation of Mixture Distributions." Proceedings of the 42nd International Conference on Machine Learning, 2025.](https://mlanthology.org/icml/2025/jayashankar2025icml-scoreofmixture/)BibTeX
@inproceedings{jayashankar2025icml-scoreofmixture,
title = {{Score-of-Mixture Training: One-Step Generative Model Training Made Simple via Score Estimation of Mixture Distributions}},
author = {Jayashankar, Tejas and Ryu, Jongha Jon and Wornell, Gregory W.},
booktitle = {Proceedings of the 42nd International Conference on Machine Learning},
year = {2025},
pages = {27021-27049},
volume = {267},
url = {https://mlanthology.org/icml/2025/jayashankar2025icml-scoreofmixture/}
}