Peruzzi M, Dunson DB (2024) Spatial meshing for general Bayesian multivariate models. Journal of Machine Learning Research 25(87):1–49. jmlr.org/papers/v25/22-0083.html arxiv.org/abs/2201.10080
Scalable posterior sampling in multivariate multi-type regression models of non-Gaussian data, when a DAG-based spatial processes models latent dependences.
Simplified Manifold Preconditioner Adaptation (SiMPA) is a Metropolis-adjusted Langevin algorithm that uses second-order information about the posterior target at a reduced cost compared to Simplified Manifold MALA.
Try it yourself using the Shiny app in the meshed R package page.
Data applications: (1) Snow cover and Leaf Area Index at 122,500 spatial locations over the Alps. Estimating the latent cross-covariance. Predictions at unmeasured locations.
(2) North American Breeding Bird survey data. 27 bird species over the continental United States. Latent factor model with up to 20 factors.
