paper.md

title	UMAP: Uniform Manifold Approximation and Projection
tags	manifold learning dimension reduction unsupervised learning
authors	name orcid affiliation Leland McInnes 0000-0003-2143-6834 1 name affiliation John Healy 1 name affiliation Nathaniel Saul 2 name affiliation Lukas Großberger 3, 4
affiliations	name index Tutte Institute for Mathematics and Computing 1 name index Department of Mathematics and Statistics, Washington State University 2 name index Ernst Strüngmann Institute for Neuroscience in cooperation with Max Planck Society 3 name index Donders Institute for Brain, Cognition and Behaviour, Radboud Universiteit 4
date	26 July 2018
bibliography	paper.bib

Summary

Uniform Manifold Approximation and Projection (UMAP) is a dimension reduction technique that can be used for visualisation similarly to t-SNE, but also for general non-linear dimension reduction. UMAP has a rigorous mathematical foundation, but is simple to use, with a scikit-learn compatible API. UMAP is among the fastest manifold learning implementations available -- significantly faster than most t-SNE implementations.

UMAP supports a number of useful features, including the ability to use labels (or partial labels) for supervised (or semi-supervised) dimension reduction, and the ability to transform new unseen data into a pretrained embedding space.

For details of the mathematical underpinnings see [@umap_arxiv]. The implementation can be found at [@umap_repo].

-

References