On modifications of Kohonen's feature map algorithm for an efficient parallel implementation

Two new variants of Kohonen's self-organizing feature maps based on batch processing are presented in this work. The purpose is to make available a finer grain of parallelism to be used in massively parallel systems. Ordering and convergence to asymptotic values for 1-D maps and 1-D continuous input and weight spaces are proved for both variants. Simulation on uniform 2-D data using 1-D and 2-D maps as well as simulations on 12-D speech data using 2-D maps are also presented to back the theoretical results.