Diffusion models can be trained to learn the mapping between parameter space and observable space for solving inverse problems

A DM is trained to learn the mapping between the parameter space and the observable space.

http://arxiv.org/abs/2608.27080v1