Fabian Jogl presented the paper “Expressivity-Preserving GNN Simulation” at the 37th Conference on Neural Information Processing Systems (NeurIPS 2023) in New Orleans. The work systematically investigates graph transformations that enable standard message

passing graph neural networks to simulate state-of-the-art graph neural networks (GNNs) without loss of expressivity. By simulating non-standard GNNs, many advanced GNNs can be implemented using standard library operations, reducing implementation complexity.

Experiments on molecular benchmark datasets show competitive and in several cases superior predictive performance.