Provable Posterior Sampling with Denoising Oracles via Tilted Transport

Neural Information Processing Systems 

From a Bayesian perspective, they offer a realistic modeling of data priors and facilitate solving inverse problems through posterior sampling.