Internships may earn an offer for a PhD position.

AI/ML

Using LLMs to infer ontological properties of system design documents

Goal: Ontological reasoning has been a topic of particular interest in the engineering of complex heterogeneous systems. However, externalizing ontological properties (i.e., modeling them) is an issue that limits the adoption of ontology-based methods. The goal here is to use large language models (LLMs) to build ontologies.

Simulation

Simulator inference by machine learning

Goal: Simulation helps understand the behavior of complex systems using sound mathematical foundations. However, the simulator of a complex system will be complex itself as well. To alleviate this complexity, simulator inference methods (such as this, this, and this) are in high demand. This project will build on previous that developed reinforcement learning based methods to infer DEVS simulators.

Approximate computing for simulation

Goal: Simulation is a powerful technique to investigate properties of a system. However, too complex simulations tend to run for a long time and consume more energy. In some situations, approximate simulations are sufficient to assess properties of the system. This project will develop approximate simulation mechanisms, mostly influenced by the general domain of approximate computing.

Forest fire and flood simulation

Goal: As a direct effect of drastic climate change, forest fires and floods are now pressing issues in numerous countries. This project will develop simulation methods for the containment and prevention of forest fires and floods.

Digital twins

Interoperable digital twins

Goal: Digital twins are real-time, virtual representations of real, physical systems. However, realistic systems are often hierarchical. To accommodate this hierarchy of subsystems, digital twins are now becoming hierarchical themselves. This raises problems of interoperability, i.e., the ability of digital twins to effective communicate and collaborate. This project will extend the state of the art on digital twin interoperability with a special focus on the (co-)simulation functionality of digital twins.

Architecting self-inferring digital twins

Goal: Digital twins are real-time, virtual representations of real, physical systems. Since real systems tend be very complex, engineering digital twins, that are meant to be faithful representations of real systems, is really hard. To alleviate this complexity, digital twin inference methods (such as this, this, and this) are in high demand. This project will have a closer look at the architectural concerns of digital twin inference.

Collaborative modeling

Garbage collection in conflict free replicated data types

Goal: Collaborative modeling is an important technique to tackle the complexity of engineered systems. In collaborative modeling, multiple experts work together on engineering models. Real-time collaboration, often implemented by conflict-free replicated data types (CRDTs) is the latest big trend in the area. However, garbage collection is a severe performance bottleneck in such setups. This project will develop semantic garbage collection methods to handle the performance bottlenecks of CRDTs.