A method for visualization of multivariate data in a lower dimension

摘要

This paper describes a method for visualization of multivariate data in a lower dimension, primarily in 2D. The method, called Distance Conservation with Filtering (DCF), creates a parameterized mapping of a data set in a lower dimension. Special functions, called filters, extract the most important pairs of points with distances between them to be preserved. A particular construction of a filter and the corresponding algorithm for learning the mapping parameters are described in detail. The DCF is favorably compared with other visualization methods on a number of data sets.