A Distance Adaptive Embedding Method in Dimension Reduction

Abstract

The distribution preservation is a challenge inthe dimension reduction methods. This paper proposes a distance adaptive embedding method (DAE). The DAE method includes the cosine similarity technology and a new distance transformation function. It has the characteristics of easy handling and strong similarity distinction. The DAE method can make small loss value and good cluster discrimination by using the new distance transformation function in the embedding.The experiment results show that the DAE method has a good performance in distribution preservation, better than the performance of the multidimensional scaling method. © Springer-Verlag Berlin Heidelberg 2013.