<ul><li>Neighbor embedding methods, such as t-SNE and UMAP, are widely used for visualizing high-dimensional data.</li><li>A lack of data-independent notions of embedding maps in these methods can introduce misleading visual artifacts.</li><li>Researchers have introduced LOO-map, a framework that extends embedding maps to the entire input space, aiming to improve reliability.</li><li>Two types of diagnostic scores have been developed to detect unreliable embedding points and improve hyperparameter selection.</li></ul>

Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective

Discover more