On Thursday, 5 February 2026, the HNI-Forum 2026 took place under the topic "Neurosymbolic Artificial Intelligence". More than 80 people attended the lecture series and listened to two exciting presentations.
In his lecture "Representation Learning to Act and Plan", guest speaker Prof Hector Geffner (RWTH Aachen University) spoke about the limits of modern AI approaches such as deep learning, deep reinforcement learning and large language models. Although these methods achieve impressive results, they require large amounts of data, are often not very transparent and have difficulties with structural reuse and generalisation. Prof Geffner argued that these challenges can be addressed by learning suitable symbolic representations. Using examples from the field of actions and planning, he showed how general world models and general planning strategies can be learnt in order to make AI systems more robust and explainable.
The presentation was complemented by the talk "Explainable Neurosymbolic Machine Learning on Knowledge Graphs" by Prof Axel-Cyrille Ngonga Ngomo, head of our Data Science workgroup, who emphasised the importance of knowledge graphs for applications in safety-critical and medical fields. Although knowledge graphs provide a natural basis for explainable AI, they have long been neglected due to their high complexity. In his talk, Prof Ngonga presented novel, scalable learning methods for knowledge graphs. The focus was on neurosymbolic approaches that combine symbolic representations with neural learning methods and thus strive for a fast and targeted solution.
We would like to thank all participants for this informative event!
Both presentations can be found on our YouTube channel at the following links:
https://youtu.be/_NI3VQJkTzI (Prof. Hector Geffner)
https://youtu.be/Bd2hostc00M (Prof. Axel-Cyrille Ngonga Ngomo)
