The field of graph neural networks (GNNs) is moving towards incorporating physical constraints and topology-awareness to improve performance in real-world scenarios. Recent developments have focused on preserving high-frequency components of nodal signals, addressing out-of-distribution generalization challenges, and adapting attention mechanisms to satisfy physical laws. These innovations enable GNNs to better model flow dynamics, capture detailed topological differences, and improve generalizability. Notable papers include:

• Graph Neural Networks for Automatic Addition of Optimizing Components in Printed Circuit Board Schematics, which demonstrates the potential of GNNs in automating design optimizations.
• Flow-Attentional Graph Neural Networks, which proposes a novel attention mechanism that satisfies Kirchhoff's first law, leading to enhanced performance on graph-level classification and regression tasks.