Taking all positive eigenvectors is suboptimal in classical multidimensional scaling.

Jeffrey Tsang Rajesh Pereira

Published in: CoRR (2014)

Keyphrases

multidimensional scaling
principal component analysis
low dimensional
dimensionality reduction
latent structure
kernel pca
euclidean distance
cluster analysis
covariance matrix
vector space
spectral clustering
machine learning
image processing
nearest neighbor
web documents
geodesic distance