At the presentation of the new institute, Prof. Dr. Jens Hogreve, Vice President for Research at the KU said: “We see digitalization as a cross-cutting issue in order to contribute to a human-centered digital society. In this, it is central to also establish our own expertise in the field of applied mathematics, which in turn serve as a basis for the application and reflection of data science and artificial intelligence.” The City of Ingolstadt’s business consultant, Prof. Dr. Georg Rosenfeld said: “This new institute backs our city’s endeavor to sharpen its profile as a location of excellent research and teaching when it comes to user-oriented and responsible digitalization. We are pleased to see our commitment already bearing fruit and attracting renowned researchers as well as additional external funding.”
One of the research focus areas of the MIDS is the fundamentals of machine learning. Nowadays, machine learning algorithms achieve near-human performances or even occasionally manage to out-perform humans in many applications. Well-known examples are image recognition (e.g. tumor detection in medicine), speech recognition or autonomous vehicles. These advances are owed to the fact that nowadays deep neural networks can be trained successfully, thanks to our enormous computing power and available data volumes. “Despite this immense practical success, we still don’t have a comprehensive theoretical understanding of why these methods work so well,” says Prof. Dr. Götz Pfander, spokesman of the Institute and holder of the Chair of Mathematics/Computational Science at the KU. In addition, tests have repeatedly shown that trained neural networks are often not very robust and that even minimal changes to the input, which are invisible to humans, can produce incorrect output. It is therefore essential to analyze the reasons for this instability, he said. "Existing guarantees of success for machine learning are too weak to explain their practical success so far. It is of great interest - especially for critical applications such as in medicine - to develop improved guarantees on a mathematical basis," Pfander said.
Another focus of the MIDS is the processing of data to predict environmental developments - for example, for weather forecasts and climate or soil research. The advancement of the mathematical foundations of these methods is essential in this process and is highly relevant to a wide range of applications in science and industry. Besides linking physical correlations with limited data, there is another challenge in modeling such complex systems: They consist of parts that occur at different scales and times. For example, weather ranges from the formation of a single snowflake to the progression of annual mean temperatures. However, a computer simulation must always be limited to a part of the scales that exist in reality - any scales smaller or larger than that have to be modeled. That is why at MIDS, researchers are working on mathematical methods to model and simulate such multi-scale problems.
In designing such simulations, the MIDS also wants to take the aspect of sustainability into account. This is important, as complex algorithms use up immense computing power, leading to high energy consumption. Addressing this problem is a core aim of the joint project “Resource Aware Artificial Intelligence for Future Technologies”. In it, Professor Dr. Felix Voigtlaender from the Chair of Reliable Machine Learning at the KU is working cooperatively with FAU Erlangen-Nürnberg, TU München and Universität Bayreuth. This Chair is also located at the MIDS and is funded by the Bavarian Hightech Agenda. The KU won the chair in a competition of the Bavarian State for the establishment of new professorships on artificial intelligence.
The MIDS professors also form the expert core of the new Bachelor degree program „Data Science“, which is to start in the winter semester at the KU. This program aims at education highly qualified professionals for industry and research, who will contribute their expertise in a responsible and discreet manner. That is why the program teaches students the basics of machine-learning and other current methods efficiently by applying cutting-edge software technologies.