Exact Posterior Score Estimation for Solving Linear Inverse Problems

arXiv·low signal

Diffusion and flow-based models learn strong data priors by training a denoiser to reverse Gaussian corruption; this work derives exact posterior score estimation to apply those priors to linear inverse problems. Of interest to practitioners using diffusion priors for reconstruction tasks (e.g., deblurring, inpainting, imaging) who want a more principled posterior sampling step.