Transformations in Semi-Parametric Bayesian Synthetic Likelihood

Abstract

Bayesian synthetic likelihood (BSL) is a popular method for performing approximate Bayesian inference when the likelihood function is intractable. In synthetic likelihood methods, the likelihood function is approximated parametrically via model simulations, and then standard likelihood-based techniques are used to perform inference. The Gaussian synthetic likelihood estimator has become ubiquitous in BSL literature, primarily for its simplicity and ease of implementation. However, it is often too restrictive and may lead to poor posterior approximations. Recently, a more flexible semi-parametric Bayesian synthetic likelihood (semiBSL) estimator has been introduced, which is significantly more robust to irregularly distributed summary statistics. In this work, we propose a number of extensions to semiBSL. First, we consider even more flexible estimators of the marginal distributions using transformation kernel density estimation. Second, we propose whitening semiBSL (wsemiBSL) -- a method to significantly improve the computational efficiency of semiBSL. wsemiBSL uses an approximate whitening transformation to decorrelate summary statistics at each algorithm iteration. The methods developed herein significantly improve the versatility and efficiency of BSL algorithms.