Reproduce “Spiking Graph Convolutional Networks” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zulun Zhu, Jiaying Peng, Jintang Li, Liang Chen, Qi Yu, Siqiang Luo

Researchers from Sun Yat-Sen University, Rochester Institute of Technology, and Nanyang Technological University developed Spiking Graph Convolutional Networks (SpikingGCN) to enable energy-efficient processing of graph-structured data. This framework achieves competitive performance in node classification and other graph tasks while demonstrating up to a 100-fold reduction in energy consumption on neuromorphic hardware compared to conventional GCNs.

Key insights

  • Spiking Neural Networks can effectively perform graph-based learning by directly integrating graph convolutions into spike encoding and decoding, achieving high performance without pre-training.
  • The event-driven, sparse computation of SNNs significantly reduces operational cost and energy consumption when applied to graph data, particularly on neuromorphic hardware.
  • Incorporating a mechanism for negative (inhibitory) spikes enhances the model's capacity to process nuanced information and can improve performance on certain graph datasets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.