In order to be able to predict severe weather events or the melting of ice in the Arctic, information in the form of heterogeneous data must be linked with numerical models of dynamic systems. This is done through data assimilation, which makes it possible to better investigate processes and predict their further development. In the field of data assimilation, the professorship is concerned with the further development of data science algorithms by incorporating physical conservation laws, and solving correspondingly large optimisation problems in the environmental sciences. Quantifying the uncertainties of predictions, numerical models and observations also plays a central role here.
Prof. Oliver works on modeling and simulation of complex multi-scale systems, in particular in climate science, but also in the bio and material sciences. In such systems, the range of relevant scales is so vast that we cannot expect that the exponentially growing capacity of modern supercomputers alone will suffice to make scientific progress. We also need to improve our algorithms, develop parametrizations - simplified models for small scale processes - and learn from data and observations. The research of the group includes structure preserving and energy consistent algorithms, mathematical approaches for model reduction, and dynamic, stochastic, and/or data driven models to represent processes at scales below the computational grid.
Research interests include compressive sensing, sampling theory for signals and operators, applied harmonic analysis, and information theory. A recurring theme in the work of the chair is the measurement, analysis and representation of continuous signals by finite sequences.
The focus of the professorship is geomatics, geomathematics or a related mathematical field related to the geosciences. This includes expertise in inverse problems, modeling in the geosciences, geomonitoring based on mathematical-statistical methods, and the analysis and visualization of spatial data and their integration into geographic information systems. The latter is the basis for long-term planning in transport, infrastructure and natural hazard prevention.
Chair of Business Administration and Business Informatics
Head of the Analytics Working Group of the Society for Operations Research (Gesellschaft für Operations Research) Program Speaker: Digital and Data-Driven Business (B.Sc.)
In our research we focus on business analytics and designing modern information systems and analytical processes for data-based planning and optimization procedures as well as digital business models and systems. One major focus is on models and methods to analyze, intelligently reduce, transform and utilize complex data sets in decision support systems. The generation of dense and useful features from heterogeneous data sets and the simplification of operational problems by integrating geometric projection and mathematical optimization play an essential role here. The statistical correction of structural biases in decisions and predictions as well as the combination of heterogeneous data, opinions and forecasts are further key areas of our research.
Prof. Stöger's research interests lie in the development and analysis of mathematical methods in machine learning and signal processing. A particular area of focus for his research is the analysis of overparameterized models in machine learning and algorithms for Compressed Sensing and Low-Rank Matrix Recovery. It combines tools from a broad range of mathematical disciplines ranging from high-dimensional probability theory to optimization. He has been collaborating with scientists from different application areas such as wireless communications and image recognition.
Funded by the Hightech Agenda Bavaria as part of the consortium "Resource Aware Artificial Intelligence for Future Technologies" of the KU Eichstätt-Ingolstadt, the FAU Erlangen-Nürnberg, the TU Munich and the University of Bayreuth.
The research group "Reliable Machine Learning" mathematically studies the properties of machine learning algorithms. Particular attention is paid to the question of stability or robustness of neural networks. Although trained neural networks often perform extremely well on real data, empirical studies have shown time and again that they are vulnerable to so-called "adversarial examples": A minimal perturbation of the input, invisible to a human, can cause the network to produce an incorrect output. The research group "Reliable Machine Learning" mathematically studies the underlying mechanisms leading to the existence of such adversarial examples, with the aim of developing novel training methods that produce provably robust neural networks. More information on research and teaching of the group can be found here and here.
Here you will learn everything about privacy
More on this
These cookies help us to see how visitors use our sites. This information is collected anonymously.
More on this
These necessary cookies provide basic functions of our website. Without these cookies you cannot use e.g. shop functions or logins. The website will therefore not function properly without these cookies.