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?

More Dimensionality Reduction interview questions

View all →