Self-Organizing Maps Neural Networks on Parallel Cluster

Parallel design and realization of artificial neutral networks on clusters can fully employ the advantage of ANN parallel processing, shorten the training time and reduce the algorithm complexity. As the parallel techniques become more and more developed, it´s increasingly important to design artificial neutral networks on clusters via a combined software and hardware method. The support of parallel cluster technique on artificial neutral networks is discussed in several aspects on software and hardware platforms. A design method and basis frame of SOM neutral networks on parallel clusters is presented, and problems on the training efficiency of neutral networks on parallel clusters are further discussed. This approach is widely applicable to the computer realization on many neutral network models.