Relative Transformation with CamNN Applied to Isometric Embedding

Neighborhood selection is one of the most important link in low-dimensional representations of high-dimensional data sets. Also, a good distance measure among the data points is where the shoe pinches. In this paper, we use the cam weighted distance to find a more flexible neighborhood of a data point in a newly-created space of r-isomap algorithm. It is a major advantage of r-isomap to optimize the process of intrinsic structure of the local information in a data set. Short-circuit edges are reduced in a certain extent because of the relative transformation space which is constructed in r-isomap. Furthermore, we can get a well performance on both orientation and scale adaptive side, because we utilize the cam weighted distance to search the neighborhood of a data point. It has been proved that this distance measure is more efficient than the Euclidean distance. Experiments demonstrated that the proposed method can give better results on dimension reduction than r-isomap, Weighted Locally Linear Embedding (WLLE) and some other approaches on the data sets which have obvious classifications. Especially robust to data sets with noise.

[1]  Sungzoon Cho,et al.  Locally linear reconstruction for instance-based learning , 2008, Pattern Recognit..

[2]  Lijun Jiang,et al.  Local relative transformation with application to isometric embedding , 2009, Pattern Recognit. Lett..

[3]  Michiel H. M. Smid,et al.  Linear-Space Algorithms for Distance Preserving Embedding , 2007, CCCG.

[4]  D. Donoho,et al.  Hessian Eigenmaps : new locally linear embedding techniques for high-dimensional data , 2003 .

[5]  Yousef Saad,et al.  Enhanced graph-based dimensionality reduction with repulsion Laplaceans , 2009, Pattern Recognit..

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

[7]  Peter E. Hart,et al.  Nearest neighbor pattern classification , 1967, IEEE Trans. Inf. Theory.

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

[9]  Yan Qiu Chen,et al.  Improving nearest neighbor classification with cam weighted distance , 2006, Pattern Recognit..

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

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

[12]  Hongyuan Zha,et al.  Principal Manifolds and Nonlinear Dimension Reduction via Local Tangent Space Alignment , 2002, ArXiv.

[13]  Keinosuke Fukunaga,et al.  The optimal distance measure for nearest neighbor classification , 1981, IEEE Trans. Inf. Theory.

[14]  Keinosuke Fukunaga,et al.  An Optimal Global Nearest Neighbor Metric , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[15]  Lijun Jiang,et al.  Globalizing Local Neighborhood for Locally Linear Embedding , 2006, 2006 IEEE International Conference on Systems, Man and Cybernetics.

[16]  Abdullah Al Mamun,et al.  Weighted locally linear embedding for dimension reduction , 2009, Pattern Recognit..

[17]  Paul L. Rosin,et al.  Selection of the optimal parameter value for the Isomap algorithm , 2006, Pattern Recognit. Lett..

[18]  Shuzhi Sam Ge,et al.  Neighborhood linear embedding for intrinsic structure discovery , 2008, Machine Vision and Applications.