Apply for the NHR Graduate School and become part of NHR4CES

Advances in technology and algorithms have enabled researchers to use large-scale computational simulations to study complex physical phenomena such as weather, mechanics, and biology. As simulation data becomes increasingly large and complex, efficient processing and analysis have become significant challenges. Visualization provides the means for efficient visual analysis, exploration, and communication of such data but needs to be adapted to support HPC-related applications. Novel efficient techniques and In-situ visualization help address this by focusing on computational resources where needed and enabling real-time data analysis during the simulation. As part of the CSG Visualization, you develop and implement such novel strategies to support the effective and efficient visual analysis of complex data on HPC clusters.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Basic Knowledge of Computer Graphics Concepts
• Good programming skills (e.g., C++, CUDA)

Working language/required language skills:

• English

Description of supervisor and scientific institution:

• Dr. Tim Gerrits, RWTH Aachen University, Lead of the Visualization Team

The PhD position on LLM-based Data Wrangling for HPC Applications aims to explore the integration of large language models (LLMs) into the data preparation pipeline for high- performance computing (HPC) workflows, with a particular focus on tabular data. In many scientific domains, HPC simulations and analyses rely on complex tabular datasets that require extensive wrangling—cleaning, transformation, normalization, and semantic alignment—before they can be effectively used. This process is often manual, errorprone, and domain-specific. Recent advances in LLMs offer a promising avenue for automating and generalizing these tasks by leveraging their ability to understand and generate structured data representations. The research will investigate how LLMs can be adapted, fine-tuned, or augmented to support scalable, context-aware data wrangling for large and heterogeneous scientific tables, enabling more efficient data preparation and accelerating scientific discovery in HPC environments.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Master’s degree in computer science or a related field, with a strong background in data management and AI
• Experience with large language models, tabular data processing, and deep learning frameworks (e.g., PyTorch, TensorFlow) is expected
• Solid knowledge of data management and database systems is required
• Familiarity with HPC environments or scientific workflows is a plus
• Strong programming skills and proficiency in English are essential

Working language/required language skills:

• English

Description of supervisor and scientific institution:

• Prof. Dr. Carsten Binnig, TU Darmstadt, Professor on Data and AI Systems

The PhD position on Integrating Vector Search into HPC Infrastructure focuses on rethinking vector search systems, crucial for applications such as similarity search in scientific data, machine learning, and AI, by leveraging the unique capabilities of high-performance computing (HPC) environments. Traditional vector databases typically do not fully exploit HPC hardware features such as high-throughput interconnects, parallel file systems, and large-scale GPU clusters. This research will investigate how to redesign vector search algorithms and systems to natively integrate with HPC infrastructure, optimizing for low-latency communication, efficient use of hierarchical storage, and scalable GPU acceleration. The goal is to enable fast, large-scale similarity search over highdimensional vectors within scientific workflows, thereby expanding the role of vector search in data-driven HPC applications.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Applicants should have a master’s degree in computer science or a related field, with strong skills in systems programming and parallel computing
• Experience with vector search, GPU programming (e.g., CUDA), and HPC technologies such as MPI, fast storage, and high-speed networks are highly desirable
• Familiarity with machine learning and data-intensive applications is a plus
• Proficiency in English and strong programming skills are required.

Working language/required language skills:

• English

Description of supervisor and scientific institution:

• Prof. Dr. Carsten Binnig, TU Darmstadt, Professor on Data and AI Systems

The project Building Foundation Models for Operational Processes aims to pioneer a new class of foundation models tailored to the unique structure and semantics of operational process data. While current foundation models and large language models (LLMs) are predominantly trained on textual corpora, enterprise systems like SAP and Salesforce generate rich, structured, event-centric data that remains underutilized by modern AI.
Leveraging the high-performance computing (HPC) resources of NHR4CES, this project will develop and train large-scale models capable of understanding, generalizing, and predicting complex business process behavior across domains. The research builds on and extends the work of the Process and Data Science (PADS) Group at RWTH Aachen University, which has laid the theoretical and methodological groundwork in object-centric process mining. Through close collaboration with the CSG Data Science and Machine Learning in NHR4CES, the project integrates advanced deep learning and data engineering techniques to create foundation models that can drive automation, compliance, and optimization in digital enterprises. The resulting models are expected to serve as a universal representation layer for operational processes, enabling novel applications in predictive analytics, process simulation, and generative process design.

Working group and link(s) to working group:

• For further information, please have a look at the Process and Data Science (PADS) Group at RWTH Aachen University and the Cross-Sectional Group Data Science and Machine Learning

Professional prerequisites for applicants:

• Strong data science and machine learning skills
• Being able to develop complex software systems in an HPC setting
• Interest in improving operational processes, as is demonstrated by basic knowledge of process mining and process modelling

Working language/required language skills:

• English
• German

Description of supervisor and scientific institution:

• Prof. Dr. Wil van der Aalst, Chair of Process and Data Science, RWTH Aachen University.

Metastable coatings of thin films prepared through physical vapor deposition (PVD)
techniques, such as magnetron sputtering, are widely used in cutting and forming
applications. Understanding the conditions for metastable phase formation, e.g.,
metastable solid solubility, is crucial to the target for determining the material’s
properties. The limitations of purely thermodynamic energy-based approaches
become apparent, as they cannot account for the full extent of metastable solid
solubility measured experimentally.
Describing the metastable phase formation during thin film growth is based on both
thermodynamics and kinetics. The proposed project is designed such that the
metastable phase formation of the proposed materials will be studied by ab initio
calculations and machine learning. Additionally, the results will be validated through
experiments.

Working group and link(s) to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Experience in ab initio calculations and materials simulations
• Experience in coupling modelling to PVD synthesis techniques
• Strong background in Materials Sciences
• Programming skills in Python, working experience in the Linux interface

Working language/required language skills:

• English (mandatory)
• German (optional)

Description of supervisor and scientific institution:

• Prof. Dr. Jochen Schneider, RWTH Aachen University, Chairman of the Board of StrucMatLab

The neural network of the human brain is not hardwired. Even in the mature brain,
new connections between neurons are formed and existing ones are deleted, which
is called structural plasticity.

This process is key to understanding how learning, memory, and
healing after lesions such as stroke work. In this PhD project, the candidate will
design scalable algorithms to simulate structural plasticity and apply them to study
memory and learning models.

Working group and link(s) to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Master’s degree in computer science or related discipline
• Experience in algorithm design
• Knowledge of graph theory
• Proficient in C/C++
• A background in neuroscience and/or parallel programming is an advantage

Working language/required language skills:

• Very good command of the English language

Description of supervisor and scientific institution:

• Prof. Dr. Felix Wolf, Professor for Parallel Programming at TU Darmstadt,

Medical imaging, especially magnetic resonance imaging (MRI), is essential for the early detection of cancer. However, the high cost of these procedures limits the frequency of examinations and the scope of preventive screening. We therefore advocate combining affordable low-cost MRI hardware with powerful highperformance computers (HPC). By offloading image accuracy and reconstruction intelligence to computers, diagnostic quality is maintained while costs are drastically reduced. The research project involves developing comprehensive simulation models for MRI systems that integrate all components – from system engineering and signal encoding to image reconstruction and post-processing – into a single framework. The core idea is to substitute costly physical parts with advanced simulations and machine learning models, utilizing regular access to HPC resources even after optimization. This innovative solution requires tightly coupled multiscale calculations that generate large amounts of data and require GPU clusters in HPC environments for efficient processing. When traditional physical modelling reaches its limits, neural networks with residual physics help close gaps in dynamic approximations, while reinforcement learning agents optimize pulse sequences under low signal-to-noise ratio conditions. With reconstruction and system modelling outsourced to HPC, this new class of imaging systems can deliver high diagnostic quality, enabling cost-effective, simulation-based imaging for cost-efficient diagnostics.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Master’s degree in physics, electrical engineering, or computer science
• Solid background in Physics and numerical simulation (Maxwell/Bloch/FEM)
• Experience with machine learning (PyTorch/JAX), reinforcement learning, and high‑performance computing
• Strong programming skills in C++ and Python; familiarity with CUDA or GPU‑accelerated frameworks is a plus

Working language/required language skills:

• English (mandatory)
• German (optional)

Description of supervisor and scientific institution:

• Prof. Dr. Volkmar Schulz, RWTH Aachen, Director of Chair for Imaging and Computer Vision, Faculty of Electrical Engineering, Physics and Medicine
• Prof. Dr. Julia Kowalski, RWTH Aachen, Director of Chair of Model-based Development (MBD), Faculty of Mechanical Engineering

Positron emission tomography (PET) – the most sensitive method for examining metabolism is on the verge of transitioning from static snapshots to continuous 3D films of the entire body, providing terabytes of raw data for each patient. To convert these large amounts of data into real-time maps of tracer kinetics, a fast simulation and reconstruction platform is required that can model, predict and instantly update the flow of the radiotracer through each organ by accessing HPC. This project aims to develop an integrated computational framework that models the entire PET imaging chain – from detector physics and annihilation events to particle transport and tracer kinetics – within a unified simulation and optimization platform. Highfidelity Monte Carlo simulations will model the physical interactions, while organspecific tracer uptake will be described by coupled differential equations. These equations will be solved in real time using GPU-accelerated computing. To enhance efficiency, neural-network-based surrogate models will approximate computationally intensive components. Additionally, reinforcement learning will be employed to dynamically optimize scan parameters for different tracers and anatomies, with the goal of maximizing image quality and quantitative accuracy at minimal radiation dose. By offloading computationally intensive tasks to scalable computing platforms, we enable total-body PET scans with detailed kinetic modelling, paving the way for a new level of diagnostic precision.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Master’s degree in physics, electrical engineering, or computer Science
• Sound knowledge of physics, numerical solutions of PDEs and Monte Carlo simulations
• Experience with machine learning (PyTorch/JAX) and high‑performance computing
• Strong programming skills in C++ and Python; familiarity with CUDA or
• GPU‑accelerated frameworks is a plus

Working language/required language skills:

• English (mandatory)
• German (optional)

Description of supervisor and scientific institution:

• Prof. Dr. Volkmar Schulz, RWTH Aachen, Director of Chair for Imaging and Computer Vision, Faculty of Electrical Engineering, Physics and Medicine
• Prof. Dr. Felix Mottaghy, Director of the Clinic for Nuclear Medicine, University Clinic RWTH Aachen.

Due to the considerable heterogeneity of patients, especially in intensive care medicine, optimal individualized treatment strategy requires continuous monitoring of all relevant patient specific parameters. However, despite substantial academic and industrial research efforts with AI-based or mechanistic equation-based models, Currently, there are no models available that fulfil the clinical requirements for reliability, predictive quality, individual adaptability, explainability and response time in a clinical IT environment. Furthermore, the disease process develops dynamically in interaction with the respective therapeutic measures. Thus, the concept of a digital twin is the technology of choice. For the simulation of patient-specific ventilation strategies in intensive care medicine, our group uses the ICSM simulator, which has around 1500 algebraic and dynamic equations that can be adapted to a single patient with the help of up to approx. 600 parameters. As part of our research, we are developing a novel mathematical strategy in which transfer learning based on very large, generic patient data sets and a hybrid approach of machine learning and mechanistic modelling is used. We aim to develop a dynamic concept that allows reliable quantification of uncertainty for patient-specific settings.

Link to working group:

• For further information, visit the Website from our CSG Data Management

Professional prerequisites for applicants:

• Strong programming skills (e.g. Python, MATLAB)
• Experience in ML tools. – Knowledge in numerical and optimization methods is desirable

Working language/required language skills:

• English (mandatory)
• German (optional)

Description of supervisor and scientific institution:

• Prof. Dr. Andreas Schuppert, RWTH Aachen University