Tag
This paper analyzes zero-shot conditional sampling with pretrained diffusion models for linear inverse problems, providing information-theoretic guarantees and proposing a projected-Langevin initialization method.