Title: Bayesian score calibration for approximate models
Short description: This paper contains a method for correcting approximate models (e.g., posteriors distributions using approximate likelihoods, sampling approximations, approximate forward models, or several layers of approximation) in a simulation-based inference framework. It presents a generalised simulation-based inference framework using any strictly proper scoring rule. The paper also develops an asymptotic justification for replacing importance weights with unit weights under certain assumptions, and shows the usefulness of a coverage-based diagnostic.
Link to paper: https://jmlr.org/papers/v26/24-1179.html
Title: Bayesian score calibration for approximate models
Short description: This paper contains a method for correcting approximate models (e.g., posteriors distributions using approximate likelihoods, sampling approximations, approximate forward models, or several layers of approximation) in a simulation-based inference framework. It presents a generalised simulation-based inference framework using any strictly proper scoring rule. The paper also develops an asymptotic justification for replacing importance weights with unit weights under certain assumptions, and shows the usefulness of a coverage-based diagnostic.
Link to paper: https://jmlr.org/papers/v26/24-1179.html