Design choices — noise schedule, parameterization, sampler, and model family — measurably change posterior accuracy in diffusion-based SBI; variance-preserving EDM diffusion with adaptive solvers leads on low-dimensional problems, flow matching is competitive on high-dimensional ones.
Montreal, Canada: MIT Press, pp
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Diffusion Models in Simulation-Based Inference: A Tutorial Review
Design choices — noise schedule, parameterization, sampler, and model family — measurably change posterior accuracy in diffusion-based SBI; variance-preserving EDM diffusion with adaptive solvers leads on low-dimensional problems, flow matching is competitive on high-dimensional ones.