Reproduce “DRGCN: Dynamic Evolving Initial Residual for Deep 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
Lei Zhang, Xiaodong Yan, Jianshan He, Ruopeng Li, Wei Chu
DRGCN addresses over-smoothing in deep graph convolutional networks by adapting how much each node retains its initial representation. A dynamic block estimates node-specific residual weights, and an evolving block updates them across layers. On semi-supervised Cora node classification, 64-layer DRGCN reports 87.5 ± 0.2% accuracy versus 87.1 ± 0.3% for GCNII; this benchmark accuracy does not directly measure representation distinguishability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.