Dagstuhl 2025

Photo album about the event.

Most of the network participants were invited to attend the Dagstuhl Seminar 25291 on (Actual) Neurosymbolic AI: Combining Deep Learning and Knowledge Graphs. Pascal Hitzler, a member of the Network, was one of the co-organisers.

Over the past decade, deep learning (DL) and knowledge graphs (KGs) have achieved major breakthroughs. DL has advanced through large-scale data, transformer models, and modern hardware, while KGs have matured thanks to web-scale integration, standardization, and improved reasoning methods. These parallel developments have sparked growing interest in how the two can complement each other.

Although DL models—especially large language models (LLMs)—excel across many tasks, they often struggle with factual accuracy. In contrast, KGs provide structured, curated knowledge that supports transparency and reasoning, though their creation remains resource-intensive.

This Dagstuhl Seminar explored how to bridge DL and KGs through neurosymbolic AI. A dedicated working group examined real-world applications, identifying (bio)medicine as a particularly relevant and impactful domain for neurosymbolic AI.

A report and white paper is currently being finalised.