UMAP is a nonlinear dimensionality reduction technique focused on preserving both local and global structure. It often produces better visualizations than t-SNE and can scale to large datasets. UMAP is popular for exploratory data analysis and cluster identification.