INLA-SPDE
A combination of Bayesian approximate inference methods in which INLA replaces MCMC sampling with Laplace approximations and SPDE approximates a continuous-space Matérn Gaussian process as a sparse Gaussian Markov random field (GMRF), enabling fast and scalable posterior inference.
Timeline (last 90 days)
ISBO (INLA-SPDE Spatio-Temporal Bayesian Optimization) is a scalable Bayesian optimization framework for spatio-temporal point process data that combines the Log-Gaussian Cox Process (LGCP) with the INLA-SPDE inference method.
ISBO adopts the INLA-SPDE method, using a Matérn field on a mesh to approximate a dense Gaussian process as a sparse Gaussian Markov random field (GMRF), thereby supporting fast posterior inference.
The INLA-SPDE method relies on mesh partitioning and Matérn field prior settings, and the impact of the choice of mesh resolution and prior parameters on the results still deserves further investigation.
All Facts (3)
ISBO adopts the INLA-SPDE method, using a Matérn field on a mesh to approximate a dense Gaussian process as a sparse Gaussian Markov random field (GMRF), thereby supporting fast posterior inference.
50%UnverifiedThe INLA-SPDE method relies on mesh partitioning and Matérn field prior settings, and the impact of the choice of mesh resolution and prior parameters on the results still deserves further investigation.
50%UnverifiedISBO (INLA-SPDE Spatio-Temporal Bayesian Optimization) is a scalable Bayesian optimization framework for spatio-temporal point process data that combines the Log-Gaussian Cox Process (LGCP) with the INLA-SPDE inference method.
50%