BAYESCOMP Group Secures CRG Funding to Revolutionize Latent Variable Modeling
The KAUST Bayesian Computational Statistics and Modeling (BAYESCOMP) Research Group is pleased to announce the award of a 2025 Competitive Research Grant (CRG) for the project: "Advancing scalable Bayesian methods for latent variable modelling." The project is funded at USD $900,000 for 36 months, with the research program officially commencing in April 2026.
About
Team
The project is led at KAUST by Prof. Håvard Rue (Lead PI) alongside researcher Dr. Haziq Jamil (Co-I). Together, the team will combine cutting-edge statistical theory with scalable computational algorithms to push the boundaries of modern psychometrics and structural equation modeling.
Research Focus
Latent variable models (LVMs)—including factor analysis, structural equation modeling (SEM), and item response theory (IRT)—lie at the heart of modern scientific inquiry. Across disciplines ranging from education and psychology to the biomedical and environmental sciences, researchers increasingly rely on these models to infer underlying abilities, attitudes, risks, or mechanisms from noisy or indirect measurements.
However, as datasets grow in complexity, conventional LVM estimation methods face severe limitations. Current Bayesian approaches that rely on Markov Chain Monte Carlo (MCMC) are computationally demanding and frequently struggle to scale when handling high-dimensional data, non-Gaussian outcomes, or spatiotemporal dependencies.
This newly funded project tackles these critical bottlenecks by leveraging the Integrated Nested Laplace Approximation (INLA), a state-of-the-art deterministic framework for fast and accurate Bayesian inference in latent Gaussian models. By adapting and reformulating complex LVMs within the INLA ecosystem, the project will deliver orders-of-magnitude gains in speed and stability, making computationally prohibitive models highly practical for everyday applied research.
Application Drivers and Outputs
A central pillar of this initiative is the development of robust, user-friendly software that bridges advanced statistical machinery with accessible workflows. The team has already introduced the INLAvaan R package as a preliminary proof of concept. This package mirrors familiar modeling syntax while seamlessly connecting to the powerful INLA computational engine.
The methodologies developed will be rigorously validated through cross-disciplinary case studies. Crucially, this research is deeply aligned with Saudi Vision 2030 and the Kingdom's national priorities in Health and Wellness, Human Capital Development, and Economies of the Future. Example applications will tackle real-world challenges such as
- Detecting geographic disparities in large-scale educational assessments;
- Analyzing intensive longitudinal data from mobile sensors; and
- Modeling complex health and survival risks.
By providing scalable, dependency-aware tools, this research will position KAUST at the international forefront of data science and next-generation statistical modeling.
