Insights into a market with great potential were offered at the AI Symposium 2025, which was organized by the mathematics department at HFT Stuttgart on 7 May. Mathematics professors Dr. Steffen Gilg, Dr. Antje Muntzinger and Dr. Darko Katic moderated the diverse program of lectures.
In her welcoming address, Prof Dr Katja Rade, Rector of HFT Stuttgart, took a look at the market for AI applications, which is already estimated at around 3 million euros in Germany alone and is expected to rise to around 20 million dollars by 2031. "We need you so that we can work through this in the future," emphasised Rade to the many students from the Department of Mathematics among the 100 or so guests.
In his keynote lecture "Mathematics meets Generative AI", Dr Lenon Vogel (Amazon, Tübingen) explained how large language models are structured, how they are trained and how they can be scaled while maintaining the same performance. "Being a good coder is extremely helpful when starting a career in the field of AI," said the AI specialist. "Knowledge of stochastics, analysis and numerics makes it easy to read current papers." Internships in software development help to put what has been learnt into practice.
Dr Birgit Graf from Fraunhofer IPA (Stuttgart) offered an interesting overview of the state of the art in assistance robots for care - the 'care emergency' and high sickness rates in care are the drivers for their use. Logistical tasks, as well as cleaning or analytics, could also be carried out with robot support in the future. In the truest sense of the word, exo-skeletons or carrying/transferring bed aids make care work easier. Rehabilitation or therapy robots support healing. In addition to cost savings, the main aim is to enable carers to spend their time working on and with people.
Prof Dr Nicola Wolpert from HFT Stuttgart presented the results of Quendrim Schreiber's research "METNet: A Mesh Exploring Approach for Segmenting 3D Textured Urban Scenes", which were published in the ISPRS Journal. In the SUM dataset, which was created by the University of Delft and contains a database of the city of Helsinki, the METNet approach was used to classify each of the 19 million triangles in the triangular mesh. RGB values, coordinates and normal vectors of the neighbouring triangles were used for the structured input into a neural network. The success: compared to previous methods, a recognisably higher accuracy was achieved.
Dr Michael Rottmaier, who works for Airbus (Ulm), presented the possibilities offered by the use of AI in the defence sector for faster data-based decision-making.
Dr Larissa Triess from Mercedes-Benz (Sindelfingen) explained the use of AI in automated vehicles at Mercedes-Benz, which differ both in terms of the degree of autonomy (autonomous driving levels 1-5) and modularity. What is important here is a verifiable, end-to-end trainable approach and the use of comprehensive sensor technology. End-to-end trainable approaches for autonomous driving tasks are the subject of current research. Safe driving, even in critical scenarios for which there is generally little data, is made possible, for example, by timely collision detection or the recognition of as yet undefined objects in the environment and early countermeasures.
Benjamin Alt from the Institute for Artificial Intelligence at the University of Bremen spoke on the topic of 'World models 2.0: From stochastic parrots to intelligent digital twins'. Robots should be able to consider the consequences of their actions on the environment. A human does this intuitively. With the help of a digital twin of a robot, these consequences can be 'tested' in advance.
Interesting discussions took place in the Q&A sessions following the presentations and during the breaks. The subsequent get-together also offered participants the opportunity to network.