Reproduce “Revisiting Over-smoothing and Over-squashing Using Ollivier-Ricci Curvature” 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

Khang Nguyen, Hieu Nong, Vinh Nguyen, Nhat Ho, Stanley Osher, Tan Nguyen

Researchers developed a unified theoretical framework using Ollivier-Ricci curvature to explain how positive curvature causes over-smoothing and negative curvature leads to over-squashing in Graph Neural Networks. Building on this, they introduced the Batch Ollivier-Ricci Flow algorithm, which improves GNN performance on node and graph classification tasks by adaptively rewiring graphs to mitigate both issues.

Key insights

  • Ollivier-Ricci curvature provides a unified geometric lens to understand both over-smoothing and over-squashing in Graph Neural Networks.
  • Positive Ollivier-Ricci curvature correlates with shared neighborhoods, leading to feature homogenization and over-smoothing.
  • Negative Ollivier-Ricci curvature indicates structural bottlenecks that impede information flow and cause over-squashing.

Summary by alphaXiv.

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