$d$-Linear Generation Error Bound for Distributed Diffusion Models
Keywords: Distributed diffusion models, Generation error bound
TL;DR: This is the first work to provide the generation error bound for distributed diffusion models.
Abstract: The recent rise of distributed diffusion models has been driven by the explosive growth of data and the increasing demand for data generation. However, distributed diffusion models face unique challenges in resource-constrained environments. Existing approaches lack theoretical support, particularly with respect to generation error in such settings. In this paper, we are the first to derive the generation error bound for distributed diffusion models with arbitrary pruning, not assuming perfect score approximation. By analyzing the convergence of the score estimation model trained with arbitrary pruning in a distributed manner, we highlight the impact of complex factors such as model evolution dynamics and arbitrary pruning on the generation performance. This theoretical generation error bound is linear in the data dimension $d$, aligning with state-of-the-art results in the single-worker paradigm.
Primary Area: optimization
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9547
Loading