Research on some problems in the Kohonen SOM algorithm

The article analyzes the relation between initial parameters setting and the formation of a topographic map of the input patterns in which the spatial locations of the neurons in the lattice are indicative of intrinsic statistical features contained in the input patterns of a Kohonen self-organizing map (SOM) algorithm. Taking a network arranged in the form of a two-dimensional lattice and trained with a two-dimensional input vector as an example, the author puts forward an initializing method for connection weights of the neurons in the competition layer.