UniMap

UniMap is a new way to visualize high-dimensional data, in development at the Visual Inference Lab at Columbia. Standard methods such as t-SNE, UMAP, and MDS draw data as a scatterplot, where the area a cluster covers says little about how many points it contains.

How it works

The high-level idea is a cross between a word cloud and PCA. As in a word cloud, the space something takes up on the map reflects how much of the data it accounts for, and as in PCA, similar data points are placed close together.

Status

A paper is in preparation, and more details will follow once it is submitted.