Partition Around Medoids Clustering on the Intel Xeon Phi Many-Core Coprocessor
Timofey Rechkalov · Electronic scientific archive of UrFU (Ural Federal University) · 2015
Abstract. The paper touches upon the problem of implementation Partition Around Medoids (PAM) clustering algorithm for the Intel Many Integrated Core architecture. PAM is a form of well-known k-Medoids clustering algorithm and is applied in various subject domains, e.g. bioinformatics, text analysis, intelligent transportation systems, etc. An optimized version of PAM for the Intel Xeon Phi coprocessor is introduced where OpenMP parallelizing technology, loop vectorization, tiling technique and efficient distance matrix computation for Euclidean metric are used. Experimental results for different data sets confirm the efficiency of the proposed algorithm.