Faculty of Computer Science

Research Group Theoretical Computer Science

Oberseminar: Heterogene formale Methoden

Abstract:

Graph neural networks (GNNs) provide a form for neural-symbolic integration, because there classification power corresponds to a fragment of first-order logic. This talk introduces GNNs, their correspondence to logic and to the Weisfeiler-Leman algorithm for graph isomorphism test. Moreover, applications to classification of chemical molecules are discussed. The talk will use the slides of Martin Grohe, who has presented this work recently at the workshop “When deep learning meets logic, a virtual workshop on neural-symbolic integration”.