David Bolin

From autonomous vehicles to environmental monitoring, #KAUST Statistics Professor David Bolin’s research focuses on developing statistical methods that make sense of complex, structured data. Leading the Stochastic Processes and Mathematical Statistics group, Bolin and his team work at the interface of probability theory and real-world systems through stochastic partial differential equations (#PDEs). His research addresses challenges where traditional statistical assumptions break down, such as traffic data constrained by road networks and environmental data shaped by ocean dynamics. These methods support applications ranging from traffic safety analysis, with the potential to inform evidence-based policy decisions, to early-warning models for ecological change in the Red Sea. By pairing methodological advances with accessible software, Bolin’s work emphasizes adoption, ensuring that theory translates into tools that can guide research, planning and policy. Read more about how Bolin is bridging mathematical statistics and real-world impact: cemse.kaust.edu.sa/articles/2026/… #CEMSE #KAUSTStatistics #ReadSea #SmartCities #RedSeaResearch #StochasticPDEs #DataScience

Interesting workshop which also will have a tutorial on the MetricGraph package.

Great work about finding hotspots for traffic accidents using the MetricGraph and excursions packages.

Check out our latest paper on model checking for latent Gaussian models using  R-INLA and STAN by @Rafael_M_Cabral , @jdavidbolin , and @HavardRue1 : arxiv.org/abs/2307.12365 We propose a workflow to check model assumptions and the robustness of the statistical answers. #RINLA

The paper on the approach that facilitates using SPDE models with arbitrary smoothness in @bayescomp_inla and @inlabru is now out in JCGS. The method recently won one of the 2023 KAUST competitions on Spatial Stats for Large Data, so it works really well. tandfonline.com/doi/full/10.10…

Huge congratulations 👏👏 to @Rafael_M_Cabral for successfully passing his PhD defense. He was advised by @jdavidbolin and @HavardRue1 and has made great contributions to latent non gaussian models.

After about one year of development, the MetricGraph #RStats package is now public! It facilitates working with data and random fields on linear networks and metric graphs, and has interfaces to @inlabru and @bayescomp_inla. For detailed tutorials, see davidbolin.github.io/MetricGraph/

Last year, @Jonas_Wallin, @AlexandreBSimas and I wrote a paper on Gaussian fields on metric graphs. It was a bit dense, so it is now split it in three parts: Regularity: arxiv.org/abs/2205.06163 Markov properties: arxiv.org/abs/2304.03190 Inference: arxiv.org/abs/2304.10372

📝Read our new post by R. Cabral, D. Bolin & H. Rue (@KAUST_News) on fitting robust non-Gaussian models in Stan and R-INLA. youngstats.github.io/post/2023/01/1… #statistics #blogpost @StatFen @BernoulliSoc @InstMathStat

Great work by Rafael Cabral (student of Håvard Rue and me at @cemseKAUST) on fitting latent non-Gaussian models.

Version 1.2.0 of rSPDE is now on CRAN. One new feature is the support for using fractional SPDE models in @inlabru. See davidbolin.github.io/rSPDE/articles… for details. #RINLA #rstats #rspatial

In December last year, @AlexandreBSimas and I updated the rSPDE package davidbolin.github.io/rSPDE/ with functions that facilitate using fractional SPDE models in R-INLA. The paper that describes the approach that was used to do this is now finally out: arxiv.org/abs/2209.04670

We're not slowing down with Expressing & Exploiting Structure in Modeling, Theory, & Computation w/ #Gaussian Processes, hearing talks by @jdavidbolin, @Princeton's Ruiyi Yang, @NorthwesternU's Matt Plumlee, @HoumanOwhadi, & @SandiaLabs's Mamikon Gulian imsi.institute/activities/gau…

@ASeatonSpatial @millerdl @dan_p_simpson @FinnLindgren The problematic parameter is the range, not the smoothness. Even if the smoothness is fixed one cannot estimate the range consistently.

@millerdl abstract looks interesting. Although I thought you cannot identify marg var, range and differentiability param all at once + I don't have an intuition for estimating differentiability, don't I need data at "infinitesimal" resolution? @dan_p_simpson @FinnLindgren

Looking forward to attending NORDSTAT again next year, where I will be organising a session about Gaussian processes on networks and graphs.

Our paper on an ICA model for fMRI using surface-based spatial priors & empirical population priors was accepted at JCGS! Using spatial priors dramatically improves power to identify areas of the brain engaged in different functional networks. @jdavidbolin arxiv.org/abs/2005.13388

Thanks, it was a nice surprise!

@dan_p_simpson Thanks!! But some might argue that it is a great day when you need to prove your own Sobolev embedding theorem. :)

New paper about Gaussian processes on metric graphs, such as street networks. With @Jonas_Wallin and Alexandre Simas. This took a long time to develop, but I am really happy about the outcome. R implementations for INLA etc are coming soon. arxiv.org/abs/2205.06163