Generalization of topology preserving maps: a graph approach

The work presents a novel algorithm, which is based on the self-organizing map (SOM) method. The combination of an undirected acyclic graph with the Kohonen learning rule results the efficient self-organizing neuron graph (SONG) algorithm. It has two modi: one is based on the adjacency information of the neuron graph, the other integrates an all-pair shortest path function, which permanently updates a generalized distance matrix. The newly developed SONG techniques were involved in pattern recognition tasks, where they proved their efficiency and flexibility.