GNN Topology Design

GNN topology design focuses on optimizing the structure of graph neural networks (GNNs) to improve their performance and efficiency in various applications. Current research emphasizes developing novel architectures, such as hybrid CNN-GNN models and those incorporating neural architecture search (NAS), to overcome limitations like computational cost and the "over-smoothing" problem in deep GNNs. These advancements aim to enhance feature extraction and fusion strategies, leading to improved accuracy and reduced resource consumption in tasks ranging from image processing to graph property prediction. The resulting improvements have significant implications for advancing the capabilities of GNNs across diverse scientific and engineering domains.

GNN Topology Design

Papers

GreedyViG: Dynamic Axial Graph Construction for Efficient Vision GNNs

Graph Property Prediction on Open Graph Benchmark: A Winning Solution by Graph Neural Architecture Search

Designing the Topology of Graph Neural Networks: A Novel Feature Fusion Perspective