Critical windows: non-asymptotic theory for feature emergence in diffusion models

arXiv:2403.01633v1 Announce Type: new
Abstract: We develop theory to understand an intriguing property of diffusion models for image generation that we term critical windows. Empirically, it has been observed that there are narrow time intervals in sampling during which particular features of the final image emerge, e.g. the image class or background color (Ho et al., 2020b; Georgiev et al., 2023; Raya & Ambrogioni, 2023; Sclocchi et al., 2024; Biroli et al., 2024). While this is advantageous for interpretability as …

abstract arxiv class cs.cv cs.lg diffusion diffusion models emergence feature features image image generation narrow property sampling stat.ml theory type windows