The problem of detecting the modes of the multivariate continuous distribution generating the data is of central interest in various areas of modern statistical analysis. The popular self-organizing map (SOM) structure provides a rough estimate of that underlying density and can therefore be brought to bear with this problem. In this paper we consider the recently proposed, mixture-based generative topographic mapping (GTM) algorithm for SOM training. Our long-term goal is to develop, from a map appropriately trained via GTM, a fast, integrated and reliable, strategy involving just a few key statistics. Preliminary simulations with Gaussian data highlight various interesting aspects of our working strategy.
展开▼
