04:00
2026-09-25
machinebrief.com
artificial-intelligence
FB-GDM: Fully-Bayesian Guided Diffusion Models for High-Dimensional Linear Inverse Problems via Unsupervised Variational Inference
Researchers introduced FB-GDM, a fully-Bayesian guided diffusion method that eliminates the per-task scalar hyperparameter calibration required by Diffusion Posterior Sampling (DPS) and Pseudoinverse-…