aiqbench analysis
This paper is about diffusion models, so it passes the topic gate. Its core finding is that the noise injected during sampling strongly affects whether models trained on different datasets converge on similar outputs, and that sample similarity is most sensitive in the middle of the sampling process, after coarse structure forms but before the specific sample is fixed. For readers, this is a research-level result about why generative models generalize rather than memorize, and it bears on how sampling settings should be chosen and evaluated.
The role of injected noise in diffusion model generalization
arXiv:2611.03764v1 Announce Type: new Abstract: Diffusion models trained on distinct datasets can produce highly similar samples when prompted with the same noise input, suggesting that they learn a generalized data distribution. We investigate the role of noise injected during the sampling process, and show that it plays an important role in making or breaking this strong generalization effect. Based on this finding we investigate the robustness of sample similarity to changes in the injected noise, and find that its impact is largest in the sampling regime where broad structure already exists but the precise identity of the sample is not yet determined.
Abstract:Diffusion models trained on distinct datasets can produce highly similar samples when prompted with the same noise input, suggesting that they learn a generalized data distribution. We investigate the role of noise injected during the sampling process, and show that it plays an important role in making or breaking this strong generalization effect. Based on this finding we investigate the robustness of sample similarity to changes in the injected noise, and find that its impact is largest in the sampling regime where broad structure already exists but the precise identity of the sample is not yet determined.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2611.03764 [cs.LG]
(or arXiv:2611.03764v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2611.03764 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 194 words
Related coverage
Beyond Sample Copying: Structural Memorization in Diffusion Models
arXiv · Oct 9, 2026
A Mechanistic Study of Algorithmic Generalization in Diffusion Models on Modular Arithmetic
arXiv · Oct 22, 2026
Tuning the Implicit Regularizer of Masked Diffusion Language Models: Enhancing Generalization via Insights from $k$-Parity
arXiv · Jun 5, 2026
From Score Learning to Discretized Sampling: An End-to-End Generalization Analysis of Diffusion Models
arXiv · Jul 28, 2026