Probability densities which are described by hyperellipsoids are trans
formed into hyperspheres. The hypersphere is topologically equivalent
to the hyperellipsoid, hence there is a transformation. The consequenc
es for the design of neural networks is discussed.