ExStra > Clusters of Excellence > Machine Learning

Machine Learning

Picture: The interactive art exhibit — The interactive art exhibit "IN ML OUT" addresses the relationship between climate change and renewable energy and demonstrates the potential of machine learning. (© Machine Learning for Science)

Clusters of Excellence

New Perspectives for Science

Machine learning is changing science even more profoundly than we imagined only a few years ago. While it has proven useful in tackling individual, long-standing scientific problems, recent developments suggest even more far-reaching possibilities. Foundation models, trained on vast datasets, provide general purpose representations suitable for a wide range of downstream tasks, and have already revolutionized tasks involving language. Diffusion models allow the generation of samples from complex probability distributions, and modern programming frameworks make it possible to implement scientific theories as part of machine learning workflows. Thus, machine learning is creating new possibilities in nearly all stages of scientific discovery and inquiry. At the same time, machine learning methodologies have obvious shortcomings regarding reliability, robustness, and interpretability, and come with risks, challenges, and blind spots.

Against this backdrop, our cluster aims to advance machine learning to aid scientific understanding across a wide range of disciplines – from medicine and neuroscience to cognitive science, linguistics, and economics, to physics and the geosciences – and to better understand and steer the impact of machine learning on scientific practice.

To this end, the cluster will address four Research Areas: (1) We will design machine learning algorithms that reveal and discover new aspects of scientific laws from data. (2) We will develop techniques to validate complex machine learning models in science, quantify their uncertainty, and identify potential failure cases. (3) We will provide methods that allow scientists to control all stages of the machine learning workflow, including generating better representations of data, creating simplified interfaces, and developing tools to analyze and understand trained models. (4) We will investigate how these developments affect scientific practice and how scientific evidence is evaluated. We will demonstrate the power of our approach by tackling challenging exemplary problems in the sciences, such as improving climate models by integrating paleoclimatic data, understanding the structure of endangered languages, identifying causes of disease progression based on multi-modal clinical measurements, and understanding the dynamics of partially observed quantum systems. To address these questions, we have assembled an interdisciplinary team of 25 internationally renowned PIs who will be supported by state-of-the-art core facilities for high performance computing, sustainable software engineering, and data management.

By building on the success of our first funding period, this cluster will continue to grow and foster an environment for machine learning in science that is second to none in Germany, Europe, or around the world.

Involved Institutions:

African Institute for Mathematical Sciences (AIMS)
ELLIS Institute Tübingen gGmbH
Leibniz-Institut für Wissensmedien (IWM)
Max Planck Institute for Biological Cybernetics
Max Planck Institute for Intelligent Systems

Website of the Cluster of Excellence

Machine Learning

New Perspectives for Science

Further Information

University of Tübingen