Distributed Unsupervised Learning Using the MULTISOFT Machine

Giuseppe Patanè and Marco Russo

Abstract

Unsupervised learning using K-means techniques is successfully employed in several application fields. When the training set and the number of reference vectors increases, the computational effort can become prohibitive for monoprocessor computers. This paper illustrates the parallelization of two clustering techniques using the MULTISOFT machine, a commodity supercomputer, built at the University of Messina. The particular management policy of the MULTISOFT machine and the implementation techniques have shown very interesting results: the speedup increases together with the complexity of the problem to be solved.

Keywords: Unsupervised Learning, Vector Quantization, Clustering, Parallel, Multicomputers.

[MyParClustering] G. Patanè and M. Russo, Distributed Unsupervised Learning Using the MULTISOFT Machine. Information Sciences,  vol. 43 no.1-4, 2002.

Download: [ps.gz] [pdf]