Resolution enhancement for the ViSOM

The topology perserving self-organising maps (SOM) has become a powerful and useful tool for visualising high dimensional data. However it requires a colouring scheme to imprint the interneuron distances on the map. The recently proposed Visualisation induced SOM (ViSOM) is able to directly preserve the distance information, as well as topology, on the map. It has been proved to be a discrete principal curve/ surface. The resolution of the map is inversely proportional to the size of the grid. Large maps require more computation thus long training time. This paper proposes two simple methods for enhancing the resolution of the ViSOM so avoiding using large maps. The first interpolates a trained map, while the second incorporates local linear projections into the projection step when mapping the data on a trained map. Experiments and results are given to demonstrate the advantages.

[1]  Helge J. Ritter,et al.  Neural computation and self-organizing maps - an introduction , 1992, Computation and neural systems series.

[2]  Nei I Doherty.  Neural Computation and Self-Organising Maps: An Introduction , 1994 .

[3]  Hujun Yin,et al.  Nonlinear Multidimensional Data Projection and Visualisation , 2003, IDEAL.

[4]  Timo Honkela,et al.  WEBSOM - Self-organizing maps of document collections , 1998, Neurocomputing.

[5]  A. Ultsch,et al.  Self-Organizing Neural Networks for Visualisation and Classification , 1993 .

[6]  Hujun Yin.  Visualisation Induced SOM (ViSOM) , 2001, WSOM.

[7]  T. Kohonen,et al.  Exploratory Data Analysis by the Self-Organizing Map: Structures of Welfare and Poverty in the World , 1996 .

[8]  Hujun Yin,et al.  Interpolating self-organising map (iSOM) , 1999 .

[9]  Juha Karhunen,et al.  Generalizations of principal component analysis, optimization problems, and neural networks , 1995, Neural Networks.

[10]  John W. Sammon,et al.  A Nonlinear Mapping for Data Structure Analysis , 1969, IEEE Transactions on Computers.

[11]  Bernhard Schölkopf,et al.  Nonlinear Component Analysis as a Kernel Eigenvalue Problem , 1998, Neural Computation.

[12]  Hujun Yin,et al.  Data visualisation and manifold mapping using the ViSOM , 2002, Neural Networks.

[13]  R. Tibshirani,et al.  Adaptive Principal Surfaces , 1994 .

[14]  M. Kramer.  Nonlinear principal component analysis using autoassociative neural networks , 1991 .

[15]  Hujun Yin,et al.  ViSOM - a novel method for multivariate data projection and structure visualization , 2002, IEEE Trans. Neural Networks.