E. N. Lychkin, V. A. Ryzhkov

The Self-Organizing Map (SOM) is one of the popular Artificial Neural Networks used for clustering and visualizing of a high dimensional data complex. Conventional SOMs [1] are based on a two-dimensional grid structure, which usually represents topological mapping of a high dimensional data complex. However, there is a typical problem coonected with conventional SOMs. This is a so-called «border effect». The intend of this paper is to show several models of spherical grids of SOM. It also includes the comparison work results of spherical SOMs and conventional SOMs. The research proves that spherical SOMs approximate input data for fewer number of training epochs.