Learning Mixtures of Gaussians Using the DDPM Objective