What is UMAP in dimensionality reduction?
Updated May 16, 2026
Short answer
UMAP is a manifold learning method that preserves both local and global structure.
Deep explanation
UMAP constructs a fuzzy topological representation of data and optimizes a low-dimensional embedding preserving structure better than t-SNE in some cases.
Real-world example
Used in genomics for visualizing cell clusters.
Common mistakes
- Assuming UMAP is deterministic by default.
Follow-up questions
- How is UMAP different from t-SNE?
- Does UMAP scale well?