Approximate Bayesian inference based on Laplace approximation and quadrature has become increasingly popular for its efficiency in fitting latent Gaussian models (LGM). However, many useful models can only be fitted as LGMs if some conditioning parameters are fixed. Such models are termed conditional LGMs, with examples including change-point detection, non-linear regression, and many others. Existing methods for fitting conditional LGMs rely on grid search or sampling-based approaches to explore the posterior density of the conditioning parameters; both require a large number of evaluations of the unnormalized posterior density of the conditioning parameters. Since each evaluation requires fitting a separate LGM, these methods become computationally prohibitive beyond simple scenarios. In this work, the Bayesian Optimization Sequential Surrogate (BOSS) algorithm is introduced, which combines Bayesian optimization with approximate Bayesian inference methods to significantly reduce the computational resources required for fitting conditional LGMs. With orders of magnitude fewer evaluations than those required by the existing methods, BOSS efficiently generates sequential design points that capture the majority of the posterior mass of the conditioning parameters and subsequently yields an accurate surrogate posterior distribution that can be easily normalized. The efficiency, accuracy, and practical utility of BOSS are demonstrated through extensive simulation studies and real-world applications in epidemiology, environmental sciences, and astrophysics.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号