Abstract
Generative Topographic Mapping (GTM) is a data visualization technique that uses a nonlinear topographically preserving mapping from latent to data space. The conventional GTM model can be interpreted as a probabilistic model with Gaussian process prior, and the choice of covariance function in the Gaussian process prior has an important effect on its properties. However the conventional GTM approach uses a covariance function with a constant length-scale for whole latent space, and therefore fails to adapt to variable smoothness of the nonlinear topographically preserving mapping. In this paper, we propose the GTM model which can control the smoothness in local areas of the latent space individually.
