On 29 September 2026, the symposium Recent Advances in Industrial and Applied
Mathematics will take place at the Jheronimus Academy of Data Science in ’s-Hertogenbosch. During this event, we bring together leading voices from academia, research institutes, and industry to explore advances, applications, and open challenges in industrial and applied mathematics. This inspiring one‑day event offers a unique opportunity to connect with the broader applied mathematics community, discover emerging research directions, and build collaborations across sectors.

The program features keynote lectures by Peter Grünwald (CWI & UL) and Svetlana Dubinkina (VU), who will provide an overview of state-of-the-art developments and current research trends in mathematics. In addition, four duo presentations by academic and industrial partners will highlight recent advances in industrial applications of mathematics. Ample time will be dedicated to networking and discussion, creating an interactive environment for exchanging ideas and building new connections.

Abstracts

From Measurements to Forecasts: Inverse Modelling and Novel Sensing for Cable Temperatures
Laura Scarabosio (RU) and Sander Rieken (Alliander)

Alliander operates a large part of the Dutch electricity distribution grid. The energy transition is driving increasing electrification and decentralized renewable generation leading to widespread grid congestion. Since the permissible load of grid components such as cables and transformers is determined by thermal constraints, accurate forecasts of asset temperatures are needed to support operational decisions.

Temperature forecasts for underground cables can be obtained by combining load forecasts with thermal cable models. However, the
accuracy of these forecasts is strongly influenced by uncertainty in the thermal properties of the surrounding soil. We demonstrate how these parameters can be inferred from temperature observations using a particle-based Bayesian inverse modelling approach, thereby improving forecast accuracy. We then discuss how the impact of the remaining uncertainties on temperature forecasts can be quantified efficiently using multilevel sampling.

In addition to conventional measurements using thermal probes or distributed fiber-optic sensing, we investigate a novel approach in
which cable temperatures are estimated from time-domain reflectometry measurements. Together, the two inverse problems—estimating thermal model parameters from temperature data and estimating temperatures from reflectometric measurements—provide more accurate cable temperature forecasts for congested electricity grids.

GPU Simulations for Urban Planning
Maxim Knepflé (Tygron) and Kees Vuik (TU Delft)

Tygron is a Dutch spin‑off company from Delft University of Technology that develops and maintains the Tygron Platform. This cloud platform lets engineers and planners integrate, enhance, simulate, and visualize data to produce better substantiated analyses, designs and decisions for urban planning. It does this by running future design scenarios in a digital twin of the City that includes multiple deterministic simulation models to calculate the impact of design choices on for example: traffic noise, heat stress or rainfall flooding. These simulations must run both at high resolution and quickly to allow testing of many possible scenarios. To achieve this, Tygron uses GPU hardware and optimized software algorithms for parallelization and acceleration developed iteratively over the past 20 years. Today the platform runs one million simulations per year across 4000 projects with often many billions of grid cells.

To meet growing demand and the added load from generative AI agents that produce many more scenarios requiring verification our R&D is continuously evolving and further optimizing these algorithms. In our presentation we will explain key architectural design choices from three perspectives: 1) technological: how to maximize GPU hardware performance; 2) mathematical: algorithms used to obtain stable solutions; and 3) user: meeting expert’s needs for practical results. We will demonstrate this with an example showing how algorithms developed by William Kahan can maximize our GPU performance and accelerate urban planning in areas affected by extreme rainfall flooding.

E is the New P
Peter Grunwald (CWI & Leiden University)

How much evidence do the data give us about one hypothesis versus another? The standard way to measure evidence is still the p-value, despite a myriad of problems surrounding it. In this talk I will provide a gentle introduction to the e-value, a recently popularized notion of evidence which overcomes some of these issues. E-values, which have a very concrete interpretation in terms of betting strategies, were only given a name as recently as 2019. Since then, interest in them has exploded with five international workshops, 100s of papers, both theoretical and applied, many in top journals such as the Annals of Statistics.
Crucially, e-values allow for effortless testing under optional continuation of data collection and combination of data from different sources – something that practitioners yearn for, yet is simply impossible to do with the classical approaches. Relatedly, they serve as the basis for anytime-valid confidence intervals, which allow for continuous monitoring of uncertainty. Optional continuation and anytime-validity is just one way in which e-values provide more flexibility than p-values – they also allow to set a type of significance/confidence level alpha after seeing the data, which is a mortal sin in classical testing. In this talk I will introduce e-values, e-processes and AV confidence intervals, and discuss in detail the relation to existing approaches.

Main literature:
G., De Heide, Koolen. Safe Testing. Journal of the Royal Statistical Society Series B, 2024 (first version appeared on arXiv 2019).
G. Beyond Neyman-Pearson: e-values enable hypothesis testing with a data-driven alpha. Proceedings National Academy of Sciences of the USA (PNAS), 2024.

Data assimilation: from climate science to neuroscience
Svetlana Dubinkina (VU Amsterdam)

Data assimilation is concerned with estimating the initial state of a chaotic dynamical system from available measurement data. Over the years, this classical definition has been broadened to encompass parameter estimation and model error estimation. Furthermore, while data assimilation was initially applied primarily in fluid dynamics — for example in weather forecasting — it is now used across many other domains, such as neuroscience.

In this talk, I will introduce the Bayesian data assimilation framework, in which the goal is not a single estimate but an ensemble of estimates that reconstruct the posterior distribution given the prior and the data. The challenge is twofold: one must have theoretical guarantees that the inverse problem is well-posed, and one must sample the posterior in a computationally efficient manner. I will present theoretical results for the inverse problem of a model describing neuronal activity in the brain, as well as computational results for a high-dimensional PDE-constrained inverse problem.

Stacking Evidence across tests in R&D
Martijn Klein Horsman (Philips), Jan Willem Bikker (CQM)

CQM is a consultancy company with over four decades of experience in industrial R&D projects. One of its long-standing customers has developed consumer products for many years and seeks to reduce the test effort and improve decision making in development projects for a certain class of products. In these development projects, different types of tests are performed on prototype designs, from A (cheap) to D (expensive). The relatively cheap tests A, B assess the prototype early on in the projects, and the more expensive tests C, D involve a trained panel of assessors for confirmation. CQM co-develops with the R&D department a method for stacking evidence, which allows predictions of outcomes of expensive tests (C or D) conditional on observed test results, typically the cheaper ones. These predictions help in deciding to stop the project or form a prior in a Bayesian analysis of future test D. As a consequence, the approach is expected to reduce overall costs for the expensive tests, and has motivated substantial investment in its development.

 The model is based on historical development projects, where typically the 4-vectors of test results (A,B,C,D) have missing entries in historical dataset. The approach uses Bayesian inference for a multivariate model of test results (A,B,C,D), capturing conditional dependencies as in Bayesian networks, and incorporating measurement models akin to structural equation models. Strong priors based on expert opinion are needed to complement scarce data, but the multivariate nature poses challenges. In addition to the technical modelling aspects, I will discuss practical learnings from employing Bayesian analysis in an R&D organisation, including communication, prior specification, and the gradual development of statistical intuition.

Mathematics Keywords: Bayesian statistics, Bayesian network, SEM

Surrogate modeling for data-efficient engineering design under uncertainty
Dimitris Loukrezis (CWI) & Kujtim Gashi (QUGATE)

Uncertainty quantification (UQ) studies, such as uncertainty propagation and sensitivity analysis, have become an essential part of engineering design workflows, helping to assess quantitatively the impact of design or operation uncertainties upon crucial quantities of interest. Monte Carlo sampling remains the baseline UQ method, however, its slow convergence rate necessitates large sample sizes to reach
acceptable accuracies. This becomes particularly problematic when the underlying high-fidelity models are computationally expensive to evaluate. In such cases, UQ studies are typically enabled through the use of surrogate models, with data-driven, machine-learning surrogates enjoying increased popularity. However, data-driven surrogates face a fundamental challenge: high-quality data from experiments or high-fidelity simulations are expensive to generate, often limiting training datasets to a few hundred samples at most. This difficulty is further exacerbated by increased problem dimensionality.

This tandem talk focuses on polynomial chaos expansion (PCE), a surrogate modeling technique that is particularly well suited to UQ, as it provides an explicit polynomial representation of the stochastic response that enables the direct computation of statistical quantities and sensitivity measures. The first part of the talk (academic perspective) will present recent methodological advances in constructing sparse PCEs via adaptive algorithms that remain accurate in high-dimensional settings and under limited data availability. The second part of the talk (industrial perspective) will demonstrate the use of adaptive sparse PCEs in real-world engineering design under uncertainty, where they enable computationally demanding UQ tasksin a fraction of the usual turnaround time. The talk will also touch on related issues such as algorithmic robustness, implementation trade-offs, and integration into existing industrial workflows