Two-sheet hyperboloid.

Publication award for more predictive power in GCN

June 15, 2023 /

IPVS researchers are improving the predictive power of graph neural networks using methods developed in-house. The approaches are described in the paper "Peudo-Riemannian Graph Convolutional Networks". This paper has now been awarded the Publication Prize 2022 of Faculty 5: Computer Science, Electrical Engineering and Information Technology.
[Picture: Bo Xiong et al., CC BY 4.0,]

Success for researchers from the Institute for Parallel and Distributed Systems (IPVS) at the Department of Computer Science: The paper "Peudo-Riemannian Graph Convolutional Networks" presented last year at the NeurIPS 2022 conference was honored with a publication award from Faculty 5. In addition to Bo Xiong and Prof. Steffen Staab of the University of Stuttgart, the paper was contributed by: Shichao Zhu, Ph.D. (Chinese Academy of Sciences, CAS), Dr. Nico Potyka (Imperial College London), Prof. Shirui Pan (Griffith University) and Associate Professor Chuan Zhou (Chinese Academy of Sciences).

Group picture from the Research Day on June 7, 2023.
Prof. Steffen Staab (right) was one of the award winners.

The paper was nominated because of the method developed in it, which allows Graph Neural Networks to be embedded in pseudo-Riemannian manifolds. These manifolds combine properties of hyperbolic, spherical and Euclidean geometries–and are thus particularly suitable for geometrically embedding different graph structures and improving the predictive power of Graph Neural Networks.

The publication prize was awarded during the Day of Research on June 7, 2023. At the ceremony in lecture hall V 57.02, Prof. Staab presented the work in an unconventional style and used the soccer game to explain: nodes stood for players, edges for a pass.

Outstanding publications honored on Research Day

To the top of the page