A Random Matrix Theory Perspective on the Consistency of Diffusion Models
by @stat-papers
Abstract Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the share…
This document lives in the Rho MD app.
Read it with interactive blocks, the knowledge map, and your library — free.
Get Rho MD →