Deep Generative Modeling for Data-Driven Simulation
Institutskolloquium
- Date: Nov 14, 2025
- Time: 10:30 AM - 12:00 PM (Local Time Germany)
- Speaker: Prof. Vlado Menkovski
- Dr. Vlado Menkovski is an Associate Professor in the Department of Mathematics and Computer Science at Eindhoven University of Technology (TU/e), where he leads the Machine Learning for Physical Science research group. His work centers on developing computational methods that accelerate scientific discovery across a range of disciplines, including materials science, mechanical engineering, and nuclear fusion. His core expertise lies in machine learning, with a particular focus on deep generative modeling and geometric deep learning. He works on advancing surrogate modeling techniques and physically consistent, data-driven simulations that bridge the gap between machine learning and the physical sciences. Dr. Menkovski earned his Ph.D. from TU/e in 2013 and holds an M.Sc. from Carnegie Mellon University (2008). Prior to rejoining academia in 2016, he worked as a Research Scientist at Philips Research and as a Data Scientist in industry. Since his return, he has played a key role in expanding TU/e’s machine learning capabilities in both research and education, especially at the intersection of AI and fundamental science.
- Location: IPP Garching
- Room: Günter-Grieger Lecture Hall (Greifswald) and Zoom
- Host: Dmitry Moseev
- Contact: dmitry.moseev@ipp.mpg.de
Machine learning surrogate modeling promises to overcome long-standing challenges of the traditional simulation approaches. This is particularly useful for complex systems where symbolic models are lacking or solving them is computationally prohibitive. These methods leverage the success of deep learning in developing efficient representations to scale simulation of dynamical systems with high-dimensional state spaces. They build on the developments in geometric deep learning to adapt to the varied data topologies that arise from the advanced discretization by representing the space as grids, meshes and graphs.
This talk discusses how deep generative modeling techniques that underpins the progress in generative AI are used to address the challenge of modeling dynamical systems sensitive to parameter changes and capture phenomena such as bifuractions and symmetry breaking. I will illustrate the successful application of such deep generative modeling approaches for simulating dynamic trajectories in diverse domains, including pedestrian dynamics, bubbly flows and nuclear fusion.