Graph Rewiring

Graph rewiring modifies the structure of a graph to improve the performance of graph neural networks (GNNs) and related graph-based machine learning tasks. Current research focuses on developing rewiring algorithms that address issues like over-squashing (inefficient information flow) and under-reaching (limited receptive fields), often employing spectral methods, curvature-based approaches, or probabilistic techniques to strategically add or remove edges. These advancements enhance GNN accuracy and efficiency, particularly for large-scale graphs and those exhibiting heterophily (nodes with dissimilar labels are connected), impacting diverse applications such as digital twin forecasting and 3D geometry processing.