Visualizing the embedding of objects in Euclidean space

Michael Littman, Deborah F. Swayne, Nathaniel Dean, and Andreas Buja Bellcore Morristown, NJ 07962-1910 ABSTRACT Matrices representing dissimilarities within a set of objects are familiar in mathematics, statistics and psychology. In this paper we describe XGvis, a software system which accepts diverse input data, such as graphs and multivariate data, develops a dissimilarity matrix from the data, and then iteratively and interactively embeds objects in a Euclidean space of arbitrary dimension. Using a technique called multidimensional scaling, objects are positioned so that their pairwise distances match the target dissimilarities as well as possible. Users can interact with XGobi, a software system for visualizing highdimensional data, to browse the resulting embeddings. Mathematicians and statisticians have found XGvis to be useful for discovering and exploring structure. XGvis runs under the X Window SystemTM.