A particle swarm embedding algorithm for nonlinear dimensionality reduction

Abstract

To cope with high-dimensional data dimensionality reduction has become an increasingly important problem class. In this paper we propose an iterative particle swarm embedding algorithm (PSEA) that learns embeddings of low-dimensional representations for high-dimensional input patterns. The iterative method seeks for the best latent position with a particle swarm-inspired approach. The construction can be accelerated with k-d-trees. The quality of the embedding is evaluated with the nearest neighbor data space reconstruction error, and a co-ranking matrix based measure. Experimental studies show that PSEA achieves competitive or even better embeddings like the related methods locally linear embedding, and ISOMAP. © 2012 Springer-Verlag.