geometry-aware-spiking-gnn

What problem does it solve?

Graph learning with energy efficiency and non-Euclidean representation by combining spike-based computation with geometry-aware representations, enabling efficient modeling of hierarchical and cyclical structures.

Core Features & Use Cases

• Riemannian embedding layer: projects nodes into constant-curvature manifolds to capture hierarchies and cycles
• Manifold spiking layer: models membrane potential dynamics in curved spaces with geometry-consistent message passing
• Joint objective and training: supports node classification and link prediction with Riemannian SGD on manifolds

Quick Start

Prepare a graph dataset and run the GSG training pipeline to begin learning on your graphs.