Apple AI Updates: July 31, 2026

1. Apple Applies Network Science To UMAP’s Internal kNN Graph

Apple researchers published work showing that the k-nearest-neighbor graph UMAP builds internally, before it projects data down to two dimensions, is a valuable analytical resource on its own. The authors demonstrate that standard network science algorithms applied to this graph can enhance data exploration: PageRank identifies representative samples, k-core decomposition separates dense central regions from sparse outer regions, and the clustering coefficient reveals tightly connected neighborhoods of similar points. Testing on MNIST and Fashion-MNIST, the team reports these graph-based techniques are competitive with or complementary to purpose-built methods such as k-medoids and HDBSCAN. Source