ScaleMAP: Preserving Local Density and Neighborhood Structure in Low-Dimensional Embeddings
Researchers have developed ScaleMAP, a novel method for low-dimensional embeddings that preserves both local density and neighborhood structure. Unlike previous techniques that normalize distances and lose scale information, ScaleMAP divides embedding displacements by the geometric mean of original-space local radii. This approach successfully recovers sparse structures and density variations in datasets from transcriptomics and hyperspectral imaging, outperforming existing methods like UMAP and DensMAP in preserving critical data characteristics. AI
IMPACT Improves data visualization and analysis for complex datasets, potentially aiding AI model interpretability.