“…PCA cannot embrace nonlinear structures, consisting of arbitrarily shaped clusters or curved manifolds, since it describes the data in terms of a linear subspace. Therefore, several methods have been proposed for reproducing nonlinear higher-dimensional structures on a lower-dimensional display: multidimensional scaling and its modifications [6], [20], [21], [22], Isomap [23], locally linear embedding [24], etc. Various neural network approaches are used for this aims as well (see e.g.…”