Six ERC Starting Grants For Researchers At TUM

Particle physics explores the fundamental questions of the universe: What is the world made of? What holds it together at its core? What is dark matter - and can we create it in the lab? To solve such mysteries, particles are collided at nearly the speed of light, 40 million times per second, and measured with around 100 million individual sensors. AI-powered algorithms are key to handling this data deluge - but they require vast amounts of simulated training data. The simulation calculations, which are currently computationally extremely intensive and slow, thus become a bottleneck for new discoveries. This is exactly where Prof. Lukas Heinrich's LEGO project comes in: He aims to significantly accelerate the process and extract much more information from each individual simulation. To do this, he uses a novel method that works similarly to AI language models: it directly predicts the next most likely particle. This could enable not only more precise analysis of existing data, but also the development of far more effective particle detectors in the future - raising the chances for groundbreaking discoveries in physics.

Lukas Heinrich is Professor of Data Science in Physics at the TUM School of Natural Sciences and a core member of TUM's Munich Data Science Institute (MDSI) .

Many global challenges, from climate change to healthcare and pandemic preparedness, involve systems where small changes can have far-reaching effects. Understanding how interventions influence outcomes in such complex dynamics requires reliable "if-then" reasoning. Traditional mathematical dynamical models often oversimplify these systems, while purely data-driven machine learning models, though powerful, can be difficult to interpret and may not generalize well to new situations. The DYNAMICAUS project, led by Niki Kilbertus, addresses this gap by combining machine learning methods with rigorous mechanistic modeling and methods from causal inference.

Niki Kilbertus is Professor of AI for Scientific Modeling at the TUM School of Computation, Information and Technology and Helmholtz AI group leader at Helmholtz Munich. He is also Principal Investigator at the Munich Center for Machine Learning .

Radiation therapy is used in approximately half of all cancer patients. Precisely defining the tissue to be irradiated - the target volume - has a major impact on treatment success: every part of the tumor must be treated while healthy tissue is spared. In his project AI-PIONEER, Dr Jan Peeken is developing an AI solution that combines cutting-edge 3D segmentation algorithms with the knowledge-processing capabilities of large language models. The system will automatically merge scientific evidence, individual patient data, and high-resolution 3D imaging, translating them into personalised target volumes that treating physicians can adjust interactively. At the same time, it allows for interactive adaptation by the treating physicians. This could help ensure consistently high treatment quality, even as cancer cases continue to rise.

Adjunct Teaching Professor Dr. Jan Peeken is Managing Senior Physician in the Department of Radiation Oncology at TUM University Hospital.

Trees and forests play a key role in the global water and carbon cycles. Through their ability to store CO₂ in wood, they are a crucial factor in climate models and in international strategies to combat climate change. But how exactly do trees regulate their water use during drought? What does this mean for wood formation? And how much CO₂ can they realistically store in the face of increasingly frequent dry spells? So far, these questions have been addressed via leaf-level processes, not considering the water needs of wood formation, which makes it uncertain how effective current climate policies will be. Richard L. Peters wants to change that. In earlier studies, he has shown that trees absorb less CO₂ from the atmosphere during drought than previously assumed. With the STEMCELL project, he will take this research further. The goal is to provide a framework for integrating the findings into growth and climate models, thereby offering a robust empirical foundation for more reliable projections to guide science-based policy decisions.

Richard L. Peters is Professor of Tree Growth and Wood Physiology at the T UM School of Life Sciences .

One of the greatest challenges in climate protection is reducing the concentration of the greenhouse gas CO₂ in the atmosphere. A particularly promising approach would be to convert CO₂ emissions from industry or transportation into useful substances through chemical processes - for example, directly into synthetic fuels or environmentally friendly base materials for plastics production. Such transformations require suitable catalysts, meaning substances that are essential for accelerating these chemical reactions. Today, the search for promising new catalysts largely relies on advanced computer simulations. However, current methods often represent real conditions only in a highly simplified way, which frequently leads to costly failed attempts. In the HeliECat project, Prof. Christopher J. Stein aims to make these simulations significantly more realistic by combining quantum chemical models with artificial intelligence. This should make it possible to predict the performance of catalysts far more reliably - an important step toward more effective climate protection.

Christopher J. Stein is Professor of Theoretical Chemistry at the TUM School of Natural Sciences .

Brain cells of patients with Alzheimer's disease display characteristic activity patterns - often decades before classic symptoms such as memory problems emerge. At this early stage, some neurons are, so to speak, hyperactive. The mechanisms behind this phenomenon are already quite well understood. Later on, in contrast, an increasing number of neurons become less active and eventually fall completely silent. This is the focus of Dr. Benedikt Zott's MONSil-AD research project. In animal models, he aims to investigate whether the silencing of neurons is connected to the loss of their links with other neurons. He is also interested in the role played by the tau protein deposits that are typical of Alzheimer's disease in the functioning of individual brain cells. In addition, Dr. Zott wants to examine whether the protein sTREM2 could be partly responsible for shutting down nerve cells. A better understanding of these processes could help pave the way for new approaches to Alzheimer's drugs.

Dr. Benedikt Zott conducts his research at the Center for Neuroradiology at TUM University Hospital and at TUM's Institute of Neuroscience .