Parallelized integrated nested Laplace approximations for fast Bayesian inference

There is a growing demand for performing larger-scale Bayesian inference tasks, arising from greater data availability and higher-dimensional model parameter spaces. In this work we present parallelization strategies the methodology of integrated nested Laplace approximations (INLA), popular framework approximate on class Latent Gaussian models. Our approach makes use thread-level parallelism, parallel line search procedure using robust regression in INLA’s optimization phase state-of-the-art sparse linear solver PARDISO. We leverage mutually independent function evaluations algorithm as well advanced algebra techniques. This way can flexibly utilize power today’s multi-core architectures. demonstrate performance our new scheme number different real-world applications. The introduction parallelism leads to speedups factor 10 more all larger already current version open-source R-INLA package, making its improved conveniently available users.