Interview with Prof. Dr. Gitta Kutyniok, Keynote Speaker at the Kaiserslautern Applied and Industrial Mathematics Days KLAIM 2023

»The Best of Both Worlds: Synthesis of Model-Based Approaches and Data-Driven Techniques«

Interview / September 27, 2023

KLAIM provides a platform for mathematicians from academia, research labs and industry to exchange ideas and present recent results. The focus of the second KLAIM conference is on the synthesis of models and data, the current research focus of Prof. Dr. Gitta Kutyniok. She holds the Chair of Mathematical Foundations of Artificial Intelligence Understanding at Ludwig-Maximilians-Universität Munich and gave the opening lecture at this year's KLAIM.

Prof. Dr. Kutyniok, what made you decide to come to Kaiserslautern for the Applied and Industrial Mathematics Days?

The Fraunhofer ITWM has the research philosophy that, on the one hand, model-based approaches are pursued, and, on the other hand, this whole new world of data-based approaches and Artificial Intelligence has emerged, which is now being brought together. In a way, combining the best of both worlds. The mathematical view is important here because it helps us to ensure the reliability of the methods that we develop. This is, of course, exactly my direction and the philosophy of my research: to combine model-based and data-based approaches.

Do you have any application examples here?

Oh yes, we deal with various applications, especially in the field of medical image processing or in the field of robotics and communication.

On the other hand, we are looking at reliable AI. »Making it reliable« in the sense of understanding in depth how it works, for example, how the training process works or what architecture is best to take, for example, how many layers to use. Another point is »explainability«: So can I understand the results of AI or also such model-based approaches, can I understand that? In my opinion, this is a direction that is very underestimated and little pursued.

Are you targeting the limits of Artificial Intelligence here?

Yes, because it is not a panacea. We found out that you can't solve certain problems with AI methods at all. If you work on digital hardware – CPU and GPU – on which you normally train neural networks, you can prove that some problems are not computable at all. This is a fundamental limit that is given to us by the hardware that we currently use, because we break everything down to zeros and ones.

Further Information

about the Conference on our event website:

About the Person

Gitta Kutyniok studied mathematics and computer science at the University of Paderborn. In 2006, she habilitated at the Justus Liebig University of Giessen. She then conducted research as a Heisenberg Fellow of the German Research Foundation at Princeton University, Stanford University and Yale University. She was then Professor of Applied Analysis at the University of Osnabrück before accepting an appointment at the Technical University of Berlin in 2011, where she has since served as Einstein Professor of Mathematics and held the Chair of Applied Functional Analysis.

Since 2020, Gitta Kutyniok has been teaching and researching at Ludwig-Maximilians-Universität München on mathematical foundations for understanding Artificial Intelligence.

Website LMU München

Will the quantum computer help us in the future?

Exactly. What helps us in particular is to move away from the digital and to use innovative new hardware that also has analog components: i.e. quantum computing, neuromorphic computing or bio computing. There are many interesting directions; the idea is to augment the digital hardware with this modern, innovative new hardware, so that you really get absolute reliability and can explain results in a provable way, as required by law, especially by the EU AI Act. That is also one of our central research directions now.

© Fraunhofer ITWM
Lecture by Prof. Dr. Gitta Kutyniok at KLAIM 2023 digitally transmitted from the LMU. At the podium: Prof. Dr. Bernd Simeon.

© Fraunhofer ITWM
The title of Prof. Dr. Gitta Kutyniok's talk at KLAIM 2023 was »Reliable AI: Successes, Challenges, and Limitations«.

Are there any touches with the application orientation of our institute?

There are various ones. You already mentioned Quantum Computing, a focus of your institute. And there are certainly overlaps at Next Generation Computing and technical image processing in the field of medicine and health, for example computed tomography or magnetic resonance imaging. Machine Learning is of course also a great connecting point, as is sustainability. With current hardware, you have huge energy problems, which goes against sustainability. We hope that with neuromorphic computing, for example, and innovative new hardware in general, we can get a better grip on this problem and solve it in some way.

© Fraunhofer ITWM
Conference Group Picture KLAIM 2023

Press Release