Diffusion samples can drift toward an empty center, a post says
When a sample is mostly noise, a diffusion model’s best guess is close to the mean, according to the post. That can pull samples toward the dataset’s center—even when no data exists there.
TLDR
A post describes an early-stage behavior in diffusion models: when a sample is mostly noise, the model’s best guess is close to the mean. The author says samples can therefore move toward the dataset’s center, even if it contains no data, before rebounding toward the data manifold—the region where the data lies.
Combined views
42
1 Source, first seen ago