The Use of Interpolation Methods for Nonlinear Mapping

摘要

In this paper we consider the possibility of using several multivariate interpolation methods as a supplement to existing dimensionality reduction techniques. Analyzed methods including nearest neighbor interpolation, inverse distance weighting, radial basis functions, and data mapping error minimization are evaluated using well-known datasets. Conducted experiments showed that radial basis functions and interpolation by the data mapping error minimization outperformed other considered methods in terms of the data mapping error yielding slightly worse quality then using stochastic gradient descent method for the whole data sets without interpolation.