A tutorial review published on arXiv and accepted in TMLR with Survey Certification examines the growing role of diffusion models in simulation-based inference (SBI). The authors—Jonas Arruda, Niels Bracher, Ullrich Köthe, Jan Hasenauer, and Stefan T. Radev—synthesize recent developments in using score-based diffusion models to estimate latent parameters from simulated and real data.

The review covers design choices for training, inference, and evaluation, and highlights concepts such as guidance, score composition, flow matching, consistency models, and joint modeling. It also discusses how noise schedules, parameterizations, and samplers affect efficiency and statistical accuracy.

The authors illustrate the approaches with case studies across different parameter dimensionalities, simulation budgets, and model types, and outline open questions for future research. The tutorial is intended as a practical resource for researchers applying diffusion models to SBI problems.