Score-of-Mixture Training: Training One-Step Generative Models Made Simple via Score Estimation of Mixture Distributions
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911055316254720 |
|---|---|
| author | Jayashankar, Tejas Ryu, J. Jon Wornell, Gregory |
| author_facet | Jayashankar, Tejas Ryu, J. Jon Wornell, Gregory |
| contents | We propose Score-of-Mixture Training (SMT), a novel framework for training one-step generative models by minimizing a class of divergences called the $α$-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 64x64 show that SMT/SMD are competitive with and can even outperform existing methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_09609 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Score-of-Mixture Training: Training One-Step Generative Models Made Simple via Score Estimation of Mixture Distributions Jayashankar, Tejas Ryu, J. Jon Wornell, Gregory Machine Learning Artificial Intelligence We propose Score-of-Mixture Training (SMT), a novel framework for training one-step generative models by minimizing a class of divergences called the $α$-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 64x64 show that SMT/SMD are competitive with and can even outperform existing methods. |
| title | Score-of-Mixture Training: Training One-Step Generative Models Made Simple via Score Estimation of Mixture Distributions |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2502.09609 |