Linear Solvers for Large-Scale Bayesian Modeling

In this talk we explore what it means to perform Bayesian inference and introduce the methodology of integrated nested Laplace approximations (INLA).

Overview

We take a dive behind the scenes and discuss the computational core operations that arise in the context of INLA and how these are addressed in the R-INLA software package and other implementations. This includes a discussions on sparse matrices, structured sparsity patterns, block dense operations and CPU vs. GPU-accelerated computations.

Presenters

Brief Biography

Lisa Gaedke-Merzhäuser is a Postdoctoral Fellow in the BayesComp group at KAUST, led by Prof. Håvard Rue. She holds a PhD in Computational Science from Università della Svizzera italiana (USI) in Lugano, Switzerland. Her research interests lie in fusing statistical learning techniques with methods from high-performance computing. She has been developing INLA_DIST, a distributed memory GPU-accelerated version of INLA for large-scale spatio-temporal models.