New Bayesian inference methods improve simulation-based calibration for complex models

Researchers have developed new methods for simulation-based Bayesian inference (SBI) for models with intractable likelihoods but tractable forward simulation. The first contribution introduces a sequential procedure using progressively localized, data-informed conditional density approximations as proposal distributions, with a final importance sampling step to correct the posterior distribution. The second contribution is localized simulation-based calibration (SBC), which probes calibration over broader neighborhoods than the posterior at a low computational cost by fitting and reusing local surrogates and ratio estimators across pseudo-observations. These methods were evaluated on three case studies, including an epidemiological application. AI

RANK_REASON The cluster contains a single academic paper detailing new statistical methods. [lever_c_demoted from research: ic=1 ai=0.4]

Read on arXiv stat.ML →

• Bayesian inference
• Gaussian mixtures-of-experts
• importance sampling
• Mixtures-of-experts of autoregressive time series: asymptotic normality and model specification
• Przegląd Epidemiologiczny
• simulation-based calibration
• simulation-based inference

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →