USI - Email

Archive / INF Seminars / INF_2023_05_02_Esmail_Abdul_Fattah

USI - Email

	Approximate Bayesian inference based on dense matrices using INLA

Host: Prof. Olaf Schenk

Tuesday

02.05

USI Campus Est, room D5.01, Sector D

13:30 - 14:30


	Esmail Abdul Fattah King Abdullah University of Science and Technology (KAUST)

Abstract:
INLA method has become a commonly used tool for researchers and practitioners to do approximate Bayesian inference for various fields of applications. As its usage has grown, it becomes essential to incorporate more complex models and expand the method’s capabilities with more features. In this talk, we propose a prototype implementation for the method based on dense matrices. The sparsity assumption of INLA makes the inference computationally challenging due to the high number of required constraints necessary to fit improper models for aerial data. We take the advantage of the multi-core architectures and the abundance of memory in today’s computational resources to design a new approach that scales well and utilizes the presence of multiprocessors on shared and distributed memory.

Biography:
Esmail Abdul Fattah is a Ph.D. candidate in Statistics, studying under the supervision of Professor Håvard Rue in his research group at KAUST. His research interests span across disciplines, incorporating his background in computer and computational science, and mathematics with his current research in computational Bayesian statistics and modeling. These interests have developed over time to include developing Bayesian statistical tools within a high-performance computing (HPC) framework.