The Study Of Parallel RSOM Algorithm

Research of the clustering algorithm for large data sets , to which the existing clustering algorithms can not work effectively, is a hot point. In this case, a RSOM tree method is proposed in paper[7]. A RSOM tree is of hierarchical SOM network structure, and can be distributedly trained and stored. The efficiency of this clustering method can be improved, benefiting from the theory of parallel computing and the development of parallel computers.Researches in this paper are based on the algorithm of RSOM tree. By using multi-thread theory, which is based on the SMP system, a parallel cluster algorithm and a parallel recognition algorithm are proposed Experiments show that the parallel algorithm based on multi-thread is more efficient than the one-thread algorithm on the large database clustering.By means of parallel distributed computing method, two parallel RSOM tree algorithms are proposed. For the first one, called center-controlled RSOM parallel algorithm, all computing tasks are controlled and distributed by only one center computer, while for the second one, all computing tasks might be controlled and distributed by many computers, called distributedly-controlled RSOM algorithm. Comparing the two parallel algorithms, the latter is much more efficient than the former.