Evaluate ensemble visualization methods across analytic contexts

Evaluate which of the five ensemble visualization methods—median projection, small multiples, confidence ellipses, movement lines, and KDE with binned points—most effectively supports interpretation across different analytic tasks and contexts.

Background

The paper develops five visualization methods for communicating variability across ensembles of UMAP projections: median projections, small multiples, confidence ellipses, movement lines, and KDE with binned points. Each method emphasizes different aspects of projection uncertainty, such as pointwise movement, structural changes, clustering variability, or consistent density patterns, and each has stated limitations.

The authors present the methods as exploratory alternatives rather than establishing which visualization is best for particular analytical purposes. They therefore identify the need for future evaluation to determine which methods most effectively support interpretation across varying analytic tasks and contexts.