Reproduce “GVE-Leiden: Fast Leiden Algorithm for Community Detection in Shared Memory Setting” 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

Subhajit Sahu

GVE-Leiden introduces a highly optimized parallel implementation of the Leiden algorithm for shared-memory multicore CPUs, achieving a processing rate of 403 million edges/second and running up to 436 times faster than sequential Leiden, while consistently producing well-connected communities and outperforming GPU-based solutions on several large graphs.

Key insights

  • Careful optimization of all three phases of the Leiden algorithm (local-moving, refinement, aggregation) is crucial for achieving high performance in parallel implementations.
  • Optimized CPU-based solutions can be highly competitive, and even surpass GPU-accelerated methods on certain massive graphs, especially when GPU memory limits are a factor.
  • A greedy refinement strategy can lead to improved performance and community quality compared to the original randomized approach in the Leiden algorithm.

Summary by alphaXiv.

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