Next MosAIk seminar will take place on Tuesday, September 8th at 2pm in the Amphitheater.

Barbara Hammer, full Professor for Machine Learning at the CITEC Cluster at Bielefeld University, Germany, will give a presentation entitled Machine Learning in critical infrastructure: Physics-informed machine learning surrogates within optimization and control tasks in water distribution systems.

Summary

Critical infrastructure such as water distribution systems is subject to an increasing amount of stress in the context of global warming. Typical tasks such as monitoring, control, and planning are challenging given the size of the systems and high degrees of uncertainties due to unclear demands. In this context, recent advances in deep learning carry the promise to substitute computationally costly simulations or only partially observable dynamics by deep surrogate models which are trained on example data. Thereby, physics-informed training enables the integration of physical laws which ensure the generalization ability beyond the observed training signals provided the regions are covered by the integrated physical principles. Since the resulting deep models are fast to evaluate and they are given in explicit analytic function, deep surrogates carry diverse promises: They allow for a fast approximation of complex dynamic behavior; they enable real-world state inference given partial information; and they support efficient system optimization based on gradient information.

In the talk, I will focus on opportunities and challenges of surrogate models for a technical application, namely hydraulic simulation of water distribution systems. I will demonstrate the power of graph neural networks to learn simulations of the dynamics based on hydraulic principles such that inference based on limited information becomes possible. I will have a short glimpse at requirements as posed by the EU’s AI-act such as fairness and robustness. Further, I will address exemplary downstream tasks, which become possible based on the surrogate, specifically sensor placement and network optimization.

Biography:

Barbara Hammer is a prominent German computer scientist and professor renowned for her research in machine learning and artificial intelligence. She earned her diploma in mathematics in 1995, her Ph.D. in computer science in 1999, and her habilitation in 2003, all from the University of Osnabrück. After serving as a professor of theoretical computer science at Clausthal University of Technology from 2004 to 2010, she joined Bielefeld University in 2010, where she leads the Machine Learning Group within the Center of Cognitive Interaction Technology (CITEC). Her core research focuses on trustworthy and explainable AI, lifelong learning, neural networks, and the analysis of complex structured data, supported by extensive international academic collaborations and visiting professorships worldwide.