Ángel Díaz Carral has joined the SimTech Cluster of Excellence as a Postdoctoral Researcher in the Statistical Model-Data Integration (SMDI) group led by Anneli Guthke. Ángel brings his expertise in machine learning and computational physics to SimTech’s interdisciplinary mission of advancing simulation technology.
The SMDI group focuses on integrating Bayesian uncertainty assessment with information-theoretic measures, aiming to merge the strengths of data-driven and physics-based modeling at the level of statistical evaluation. This innovative approach establishes new benchmarks for data-integrated simulation model assessment, driving advancements in uncertainty quantification, model diagnostics, error detection, and model improvement. These methodologies are applied in fields such as hydrology, hydrogeology, and environmental systems, with potential for cross-disciplinary impact.
Supporting the GeoMod4Future Project
At SimTech, Ángel is contributing to the GeoMod4Future project, an initiative that reimagines traditional geomodeling approaches to tackle pressing environmental challenges such as climate change and flood hazards. His work centers on developing sustainable AI models that enhance system understanding while minimizing resource consumption, aligning with SimTech's mission to create advanced simulation methods for better environmental understanding and protection.
The GeoMod4Future project aims to transform the geosciences by introducing network-like “neural stars”, a revolutionary concept inspired by recent machine learning advances. Unlike traditional models that are designed for specific purposes, neural stars establish a physically consistent and interpretable core through multidirectional training, enabling them to adapt to new questions without retraining. This approach optimizes the use of existing data, deepens understanding of natural systems, and creates models fit for diverse applications.
Ángel is collaborating closely with PD Dr. Uwe Ehret from the Karlsruhe Institute of Technology (KIT). With a PhD in machine learning and computational physics, his expertise in knowledge-driven machine learning for scientific applications, such as materials discovery, high-temperature superconductivity and bioinformatics, is instrumental in achieving the ambitious goals of GeoMed4Future.
Pioneering Open-Purpose Modeling
GeoMod4Future is setting a new standard by moving away from the traditional “fit-for-purpose” modeling paradigms. Instead, the project envisions open-purpose models capable of addressing diverse scenarios and revealing their uncertainties about previously unseen questions. By leveraging innovations such as probabilistic neural stars, multidirectional training, and smart selection layers, this initiative aims to deliver a proof-of-concept using real-world hydrological datasets within its 1.5-year timeframe.