The choice of a method for feature space decomposition for non-linear dimensionality reduction

This paper considers two approaches to the hierarchical decomposition of the feature space to improve the efficiency of the non-linear dimensionality reduction method. The first approach suggested by the author of the paper is based on the decomposition of the original feature space using hierarchical clustering. The second original approach is based on a hierarchical decomposition of the target space by using a KD-Tree. The approaches analyzed are evaluated in terms of the efficiency of the non-linear dimensionality reduction method.