This research introduces a novel quantum-inspired method improving long-range dependency modeling in graphs, suggesting a new pathway for advanced graph learning.
Key Points
The aim is to enhance Graph Neural Network performance by mitigating long-range dependency issues through a novel entropy-based approach.
Introduced Quantum-Inspired Graph Neural Network (QGNN) with Quantum Entanglement Loss (QEL) function.
Minimized von Neumann entropy of the node embedding correlation matrix.
Evaluated on benchmark datasets including Cora, Citeseer, PPI, and Long Range Graph Benchmark (LRGB).
QGNN achieved 37.6% relative MAE reduction on Peptides-struct compared to GCN.
4.0% improvement over Graph Transformers (GraphGPS) on LRGB datasets.
97% better performance than GCN on node pairs separated by 7+ hops.