AI for Physical Systems

The AI for Physical Systems Research Department investigates the integration of machine learning methods – in particular, neural networks – into the modeling, simulation, and control of real-world physical systems. 

The goal is to combine data-driven methods with physical knowledge to develop powerful, robust, and interpretable models. This involves combining general AI models, such as Large Language Models (LLMs), with specialized architectures tailored to physical laws, boundary conditions, and domain-specific data structures. 

A central approach is to use LLMs as high-level orchestrators in goal-oriented multi-agent systems. These systems break down complex problems into subtasks, devise solution strategies, and coordinate the interaction of heterogeneous components. The actual processing of domain-specific tasks, however, is performed by specialized neural networks optimized for their respective application domains.