neuro-inspired Enhancing Spiking Graph Convolutional Networks

We introduce Enhanced GC-SNN, the first integration of Leaky Integrate-and-Fire (LIF) neurons with graph convolutional networks for link prediction. Our architecture incorporates temporal spike dynamics and novel Spatial-Temporal Feature Normalization (STFN) to enable event-driven graph processing. Unlike static GNN activations, LIF neurons provide membrane-based temporal integration, while STFN stabilizes spiking activity through homeostatic regulation.Evaluation on citation networks reveals mixed performance characteristics: Enhanced GC-SNN achieves superior results on large, sparse graphs (PubMed: 0.937 ROC-AUC vs. 0.917 GCN) but underperforms on smaller, denser networks (Cora: 0.884 vs. 0.932 GCN; Citeseer: 0.877 vs. 0.931 GCN). The temporal processing requires 5.86× more operations than static baselines due to 40-timestep recurrence, though 65.3% neuron sparsity suggests potential for neuromorphic acceleration.Our primary contribution establishes a technical framework bridging spiking neural computation with graph learning, introducing STFN normalization and spike-based pairwise decoding. While computational overhead limits immediate practical deployment, the architecture provides a foundation for future neuromorphic graph processing and offers research directions toward hardware-efficient, temporally-aware graph neural networks. The work demonstrates both the potential and current limitations of biologically-inspired approaches in graph machine learning<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>.