Parallel Clustering on a Unidirectional Ring

In this paper a parallel version of the squared error clus tering method is proposed which groups N data vectors of dimension M into K clusters such that the data vectors within each cluster are as homogeneous as possible whereas the clusters are as heterogeneous as possible compared to each other The algorithm requires O NMK P time on P processors assuming that P K Speed up measurements up to processors are given It is sketched how this algorithm could be applied to parallel global optimization over continuous variables