Minimization of the mapping error using coordinate descent

Abstract

Visualization harnesses the perceptual capabilities of humans to provide the visual insight into data. Structure preserving projection methods can be used for multidimensional data visualization. The goal of this paper is to suggest and examine the projection error minimization strategies that would allow getting a better and less distorted projection. The classic algorithm for Sammon’s projection and two new its modifications are examined. All the algorithms are oriented to minimize the projection error because even a slight reduction in the projection error changes the distribution of points on a plane essentially. The conclusions are made on the results of experiments on artificial and real data sets.