Study examines how injected noise shapes diffusion model generalization
A new arXiv paper reports that diffusion models trained on different datasets can generate highly similar samples from the same noise input, suggesting they learn shared generalizable structure rather than memorizing training data.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-04 05:00 · arXiv cs.LG
The role of injected noise in diffusion model generalization