SNaP: One-Step Posterior Sampling Over 30x Faster Than Iterative Methods

WashU's Kamilov group introduced SNaP, a MeanFlow-based one-step posterior sampler for noisy inverse problems that needs just one network evaluation (about 12ms) per sample and is over 30x faster than iterative baselines, validated on deblurring, super-resolution, inpainting, and MRI reconstruction.

2026-09-29 ~ 2026-09-29 · 3 related posts

• SNaP paper: one-step posterior sampling for noisy inverse problems, 30x faster — prof_kamilov · 2026-09-29
• SNaP one-step posterior sampling runs 30x faster than iterative samplers — prof_kamilov · 2026-09-29
• SNaP: one-step posterior sampler hits 12ms per sample, 30-2250x faster than iterative — prof_kamilov · 2026-09-29