Relaxed maximum-variance unfolding

Nonlinear dimensionality reduction (recently called manifold learning) is crucial in many research fields. Maximum-variance unfolding (MVU) is one of the most important approaches to it. Unfortunately, due to too strict local constraints, MVU cannot unfold the manifold when short-circuit edges appear or the embedded mapping is conformal but not isometric. A relaxed version, relaxed MVU (RMVU), is proposed. Neighbors are adaptively assigned when short-circuit edges appear, and local distance is rescaled when the manifold is assumed to be angle-preserving. RMVU can effectively solve the preceding problems with MVU. More importantly, RMVU also performs better than MVU in general cases, and hence it has huge potential in many fields. Additionally, the proposed two strategies can also be used in other manifold learning algorithms. Experiments, accompanied with numerical comparisons, were performed on both synthetic and real data sets to demonstrate the effectiveness of RMVU.

[1]  Kilian Q. Weinberger,et al.  Graph Laplacian Regularization for Large-Scale Semidefinite Programming , 2006, NIPS.

[2]  Michael H. Bowling,et al.  Subjective Localization with Action Respecting Embedding , 2005, ISRR.

[3]  Chenping Hou,et al.  Learning High Dimensional Correspondences Based on Maximum Variance Unfolding , 2007, 2007 International Conference on Mechatronics and Automation.

[4]  S T Roweis,et al.  Nonlinear dimensionality reduction by locally linear embedding. , 2000, Science.

[5]  B. Borchers.  CSDP, A C library for semidefinite programming , 1999 .

[6]  Chenping Hou,et al.  Novel semisupervised high-dimensional correspondences learning method , 2008 .

[7]  Mikhail Belkin,et al.  Laplacian Eigenmaps for Dimensionality Reduction and Data Representation , 2003, Neural Computation.

[8]  J. Tenenbaum,et al.  A global geometric framework for nonlinear dimensionality reduction. , 2000, Science.

[9]  Lawrence K. Saul,et al.  Analysis and extension of spectral methods for nonlinear dimensionality reduction , 2005, ICML.

[10]  Kilian Q. Weinberger,et al.  Nonlinear Dimensionality Reduction by Semidefinite Programming and Kernel Matrix Factorization , 2005, AISTATS.

[11]  H. Zha,et al.  Principal manifolds and nonlinear dimensionality reduction via tangent space alignment , 2004, SIAM J. Sci. Comput..

[12]  Michael Pepper,et al.  Electrically Driven Single-Photon Source , 2001, Science.

[13]  Joshua B. Tenenbaum,et al.  Global Versus Local Methods in Nonlinear Dimensionality Reduction , 2002, NIPS.

[14]  Lawrence K. Saul,et al.  Think Globally, Fit Locally: Unsupervised Learning of Low Dimensional Manifold , 2003, J. Mach. Learn. Res..

[15]  Kilian Q. Weinberger,et al.  Unsupervised learning of image manifolds by semidefinite programming , 2004, CVPR 2004.

[16]  Mukund Balasubramanian,et al.  The Isomap Algorithm and Topological Stability , 2002, Science.

[17]  Kilian Q. Weinberger,et al.  Learning a kernel matrix for nonlinear dimensionality reduction , 2004, ICML.