Jonas Arruda, Niels Bracher, Ullrich Kothe, Jan Hasenauer, and Stefan T. Radev present a tutorial review on diffusion models for simulation-based inference. The authors synthesize developments covering training, inference, and evaluation design choices. They discuss concepts including guidance, score composition, flow matching, consistency models, and joint modeling, examining how noise schedules, parameterizations, and samplers affect efficiency and statistical accuracy. Case studies across parameter dimensionalities and simulation budgets illustrate these concepts alongside open research questions.
No score is assigned. Sources and their independence are shown in the citation chain below.