Research

Dr. Neuer and his team use physics-informed learning in a simple form of artificial intelligence for their solutions. This gives learning models direct access to mathematical and physical knowledge. These models learn in a supervised, unsupervised or reinforcing manner and are coordinated via multi-agent systems. By integrating existing knowledge such as mathematical equations, laws of nature or expert knowledge, these algorithms are able to learn quickly and on little data. He and his team used ontologies and generative transformers to digitize complex content of human knowledge.

The core element of all the methods used by the team is the stringent inclusion of uncertainty and chance. The various learning paradigms are supplemented by stochastic elements. Variational autoencoders and mixture-density networks are examples of this approach. With the help of sampling methods from the field of perturbation theory, the behavior of the learning model can be analyzed after the training process and often mapped to purely mathematical relationships. The team uses these approaches to achieve explainability of a model prediction and thus increase confidence in the model.

As a result of his many years of work, Dr. Neuer also deals with the philosophical and ethical aspects of artificial intelligence and its impact on society.