Graph Neural Networks (GNN) are used in many important applications in computer vision, natural language processing, and biology. For example, DeepMind's AlphaFold is a graph neural network that can accurately predict 3D models of protein structures and is accelerating research in nearly every field of biology. Understanding the predictions of graph neural networks is crucial for users of the models. However, conventional approaches for explainable artificial intelligence (XAI) are applicable to tables and images, but not to graph data because a node can have an arbitrary number of neighbors and the order of neighbors does not matter. Therefore, special XAI approaches for graphs are needed. The goal of this project group is to design novel algorithms for explaining the predictions of graph neural networks.

Related Work

Ying, Zhitao, Dylan Bourgeois, Jiaxuan You, Marinka Zitnik, and Jure Leskovec. "Gnnexplainer: Generating explanations for graph neural networks." Advances in neural information processing systems 32 (2019).


Dominik Köhler, Stefan Heindorf