Diffusion models pull early samples toward dataset center even where no data exists

alec_helbling · x · 2026-09-08

The author points out a curious property of diffusion models: when a sample is mostly noise, the model's best guess is close to the mean, so early in sampling, samples can be pulled toward the center of the dataset—even where no data exists—before rebounding toward the data manifold.

Original post →

More from Research

Research channel →