13.08.2025
From Physics Dreams to Algorithm Discovery - With Niki Kilbertus
Research Film
As a kid, Niki Kilbertus dreamed of becoming a theoretical physicist and discovering a fundamental law of nature. But when reality proved more complex, he found a new path through computer science.
Now a professor at TUM and PI at the Helmholtz Center as well as the MCML, Kilbertus works at the intersection of AI and causal inference. His mission: build algorithms that don’t just detect patterns, but help uncover cause and effect.
In medicine, for example, data might suggest a non-invasive kidney stone treatment works better. But if it’s mostly given to patients with smaller stones, that’s correlation, not causation. To truly compare treatments, randomized trials are needed—removing hidden biases and revealing real effects.
The research of Kilbertus helps close this gap. His algorithms support more reliable scientific decisions and accelerate discovery in fields like biology, chemistry, and healthcare.
What began as a quest for physical laws has become a drive to make science itself smarter.
©MCML
The film was produced and edited by Nicole Huminski and Nikolai Huber.
Related
11.12.2025
From Sitting Dog to Standing: A New Way to Morph 3D Shapes
ICLR 2025 work by Lu Sang and Daniel Cremers in collaboration with U Bonn enables smooth, physics-aware 3D shape deformation from point clouds.
©BAdW / Karl Neunert
08.12.2025
Tom Sterkenburg Wins Karl-Heinz Hoffmann Prize of the Bavarian Academy of Sciences
MCML JRG Leader Tom Sterkenburg receives the Karl-Heinz Hoffmann Prize of the BAdW for his interdisciplinary research.
©Juli Eberle / TUM / ediundsepp Gestaltungsgesellschaft
04.12.2025
World’s First Complete 3D Model of All Buildings Released
Xiaoxiang Zhu’s team releases GlobalBuildingAtlas, a high-res 3D map of 2.75B buildings for advanced urban and climate analysis.
04.12.2025
When to Say "I’m Not Sure": Making Language Models More Self-Aware
ICLR 2025 research by the groups of David Rügamer, and Bernd Bischl introduces methods to make LLMs more reliable by expressing uncertainty.