One-step Diffusion Models with Bregman Density Ratio Matching
This work addresses the slow sampling problem in generative AI for applications requiring fast generation, though it is incremental as it builds on existing distillation methods.
The paper tackled the computational expense of multi-step sampling in diffusion models by proposing Di-Bregman, a framework for distillation using Bregman divergence-based density-ratio matching, achieving improved one-step FID on CIFAR-10 and text-to-image generation.
Diffusion and flow models achieve high generative quality but remain computationally expensive due to slow multi-step sampling. Distillation methods accelerate them by training fast student generators, yet most existing objectives lack a unified theoretical foundation. In this work, we propose Di-Bregman, a compact framework that formulates diffusion distillation as Bregman divergence-based density-ratio matching. This convex-analytic view connects several existing objectives through a common lens. Experiments on CIFAR-10 and text-to-image generation demonstrate that Di-Bregman achieves improved one-step FID over reverse-KL distillation and maintains high visual fidelity compared to the teacher model. Our results highlight Bregman density-ratio matching as a practical and theoretically-grounded route toward efficient one-step diffusion generation.