ResearchJust now
Apple Research Unlocks UMAP’s Internal kNN Graph
Researchers at Apple found that UMAP's internal k-nearest-neighbor graph preserves high-dimensional data manifolds better than its final 2D projection. Applying PageRank and k-core decomposition to this graph identifies representative points and dense clusters. This approach reduces the distortion inherent in dimensionality reduction. Practitioners can now extract more accurate structural insights before projecting data.