Reproduce “Dimensionality Reduction Meets Network Science: Sensemaking on UMAP's kNN Graph” 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

Duen Horng Chau, Donghao Ren, Fred Hohman, Dominik Moritz

The authors analyze UMAP’s directed nearest-neighbor graph as a resource beyond its 2D scatter plot. They use PageRank to select representative examples, in-degree k-core decomposition to identify hierarchy, and clustering coefficient to examine local neighborhoods. On MNIST and Fashion MNIST, PageRank selections have lower class-distribution divergence than k-medoids for budgets of at least 200 points, but not every selection objective favors PageRank.

Summary by alphaXiv.

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