At this year's ECML-PKDD (European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases), Andreas Roth and Thomas Liebig (both SFB 876 - B4) received the "Best Paper Award". In their paper "Transforming PageRank into an Infinite-Depth Graph Neural Network" they addressed a weakness of graph neural networks (GNNs). In GNNs, graph convolutions are used to determine appropriate representations for nodes that are supposed to link node features to context within a graph. If graph convolutions are performed multiple times in succession, the individual nodes within the graph lose information instead of benefiting from increased complexity. Since PageRank itself exhibits a similar problem, a long-established variant of PageRank is transformed into a Graph Neural Network. The intuitive derivation brings both theoretical and empirical advantages over several variants that have been widely used so far.

https://2022.ecmlpkdd.org/index.php/2022/09/12/best-paper-awards/