2 EFFICIENT CONNECTIONIST SCHEMES FOR STRUCTURE PRESERVING DIMENSIONALITY REDUCTION

We propose two neural-net-based methods for structure preserving dimen sionality reduction. Method 1 selects a small representative sample an d applies Sammon's method to project it. This projected data set is th en used to train an MLP, Method 2 uses Kohonen's self-organizing featu re map (SOFM) to generate a small set of prototypes which is then proj ected by Sammon's method. This projected data set is then used to trai n an MLP. Both schemes are quite effective in terms of computation tim e and quality of output, and both outperform methods of Jain and Mao o n the data sets tried.