A Heuristic Relaxation Method for Nonlinear Mapping in Cluster Analysis

Abstract
A relaxation method mapping high-dimensional sample points to low-dimensional sample points is discussed. This method tries to preserve the local interdistance of sample points. Some special heuristics have been introduced to handle the difficulty arising from a large amount of data. Experimental results with handwritten character data and Iris data show that the method runs fast, converges rapidly, and requires a small amount of memory space.