Incremental Manifold Learning Algorithm Using PCA on Overlapping Local Neighborhoods for Dimensionality Reduction

Authors:
Yubin Zhan;Jianping Yin;Guomin Zhang;En Zhu
Affiliations:
School of Computer, National University of Defense Technology, Changsha, China;School of Computer, National University of Defense Technology, Changsha, China;School of Computer, National University of Defense Technology, Changsha, China;School of Computer, National University of Defense Technology, Changsha, China
Venue:
ISICA '08 Proceedings of the 3rd International Symposium on Advances in Computation and Intelligence
Year:
2008

Citing 7
Cited 1

Self-Organizing Maps

Self-Organizing Maps
Laplacian Eigenmaps for dimensionality reduction and data representation

Neural Computation
Think globally, fit locally: unsupervised learning of low dimensional manifolds

The Journal of Machine Learning Research
Principal Manifolds and Nonlinear Dimensionality Reduction via Tangent Space Alignment

SIAM Journal on Scientific Computing
Incremental Nonlinear Dimensionality Reduction by Manifold Learning

IEEE Transactions on Pattern Analysis and Machine Intelligence
Non-isometric manifold learning: analysis and an algorithm

Proceedings of the 24th international conference on Machine learning
Riemannian Manifold Learning

IEEE Transactions on Pattern Analysis and Machine Intelligence

An incremental manifold learning algorithm based on the small world model

LSMS/ICSEE'10 Proceedings of the 2010 international conference on Life system modeling and and intelligent computing, and 2010 international conference on Intelligent computing for sustainable energy and environment: Part I

Quantified Score

Hi-index	0.00

Visualization

Abstract

A novel manifold learning algorithm called LPcaML is proposed in this paper. Based on the geometric intuition that d-dimensional manifold locally lies on or close to d-dimensional linear space, LPcaML first finds an *** -TSLN of the whole high-dimensional input data set and then obtains the low-dimensional local coordinates of each neighborhood in the *** -TSLN using classical PCA technique while preserving the local geometric and topological property of each neighborhood. At last LPcaML transforms each local coordinates to a unified global low-dimensional representation by processing each neighborhood in their order appeared in *** -TSLN. And the transformation function of each neighborhood is obtained by solving a least square problem via the overlapped examples. By using the divide and conquer strategy, LPcaML can learn from incremental data and discover the underlying manifold efficiently even if the data set is large scale. Experiments on both synthetic data sets and real face data sets demonstrate the effectiveness of our LPcaML algorithm. Moreover the proposed LPcaML can discover the manifold from sparsely sampled data sets where other manifold learning algorithms can't.