UMAP (Uniform Manifold Approximation and Projection)

Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary

Projects high-dimensional data into two or three dimensions, keeping local neighborhoods and much of the overall layout, and fast enough for millions of points.

What is UMAP (Uniform Manifold Approximation and Projection)?

UMAP is a dimensionality reduction method that projects high-dimensional data into two or three dimensions while preserving both local neighborhoods and much of the larger-scale arrangement. It is the usual first choice for plotting embeddings, since it scales to large datasets.

Unlike principal component analysis, UMAP is nonlinear, so it can unfold curved structure that a rotation of the axes cannot. Unlike t-SNE, it can also project new points into an embedding it has already built, which makes it usable inside a pipeline rather than only in a notebook.

How UMAP Works

  1. Build a neighbor graph. Each point connects to its nearest neighbors, with edge weights reflecting how confident the connection is. The neighbor count decides whether the graph describes fine detail or broad shape.
  2. Assume even coverage. UMAP treats the data as evenly spread over a curved surface and rescales each point’s distances accordingly, so dense and sparse regions get equal footing.
  3. Lay the graph out. Connected points attract, unconnected ones repel, and the 2D graph settles into something resembling the original. The minimum-distance setting controls how tightly clusters pack.

UMAP vs t-SNE

Both produce 2D scatter plots of high-dimensional data; the practical difference is that UMAP runs faster on large datasets and keeps more of the global arrangement, while t-SNE concentrates on getting each point’s immediate neighborhood right.

That changes what you may conclude from a picture. With t-SNE, the gap between two clusters means little; with UMAP, relative placement carries some information, though it remains a projection and not a measurement. Both are stochastic, so fix the random seed before anyone quotes a plot in a report.

CriterionUMAPt-SNE
Structure preservedLocal neighborhoods plus more of the global arrangementLocal neighborhoods; distances between clusters carry little meaning
SpeedFaster, and practical on millions of pointsSlow on large datasets; often run on a subsample
New data pointsCan project new points into an existing embeddingNo transform for unseen points; rerun on the whole set
Main settingsNumber of neighbors, minimum distancePerplexity

Example of UMAP

A biology lab sequences 200,000 individual cells and gets an expression level for 20,000 genes in each one. Nothing about that table is inspectable.

UMAP reduces each cell to two coordinates and the plot resolves into a dozen islands, which known marker genes identify as cell types. Two islands sit connected by a thin bridge of points, and the cells on the bridge express markers from both, consistent with a transition between states.

The lab then projects a second batch of cells onto the same embedding. New cells landing far from every island flag either a rare population or a sample-preparation problem, and both are worth checking before the analysis goes further.

Related AI terms: t-SNE · Dimensionality Reduction · Principal Component Analysis · Clustering · Anomaly Detection

Did you like the UMAP (Uniform Manifold Approximation and Projection) gist?

Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.

Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About

Read the Governor's Letter

Stay ahead with Governor's Letter, the newsletter delivering expert insights, AI updates, and curated knowledge directly to your inbox.

By subscribing to the Governor's Letter, you consent to receive emails from AI Guv.
We respect your privacy - read our Privacy Policy to learn how we protect your information.

Browse All AI Terms A–Z

Every term in the dictionary, in alphabetical order. Jump to a letter or scroll the full list.

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

U

V

W

X

Y

Z