Reproduce “SpikeGraphormer: A High-Performance Graph Transformer with Spiking Graph Attention” 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
Yundong Sun, Dongjie Zhu, Yansong Wang, Zhaoshuo Tian, Ning Cao, Gregory O'Hared
SpikeGraphormer combines a spiking attention branch with a sparse GNN branch for large-graph node classification. It converts query, key, and value signals into binary spikes, replacing dense attention multiplication with masks and sparse additions. The authors claim all-pair interactions at linear attention complexity. On OGB-Proteins, it reports ROC-AUC 79.62±0.90 versus NodeFormer’s 77.45±1.15; scaling tests show lower runtime and memory than a vanilla Transformer. Linear-in-node-count complexity depends on sparse graphs and D≪N.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.