AdaGraph: A Graph-Native Clustering Algorithm That Overcomes the Curse of Dimensionality and Enables Scientific Discovery

In three linesAdaGraph is a graph-native clustering algorithm that overcomes the curse of dimensionality by operating on kNN topology rather than Euclidean metrics. Without specifying k a priori, it identifies gene modules in genomics (GSE14520, 10k genes), achieves ARI=0.751 on text clustering (20NG-6cat vs HDBSCAN 0.464), and outperforms Silhouette/Davies-Bouldin on 10 benchmarks up to d=5000.

Summary generated by Claude — human-verified