Parallel clustering algorithm by deterministic annealing
Shi Shuming · Journal of Tsinghua University(Science and Technology) · 2003
Partition clustering and hierarchical clustering are two fundamental clustering methods. Partition clustering is often implemented as an optimization problem, but traditional partition clustering algorithms have difficulty achieving global optimization. This paper describes a parallel partition clustering algorithm that uses deterministic annealing to avoid the disadvantages of traditional methods and to improve performance. The algorithm was then implemented in parallel on cluster of workstations (COW). The optimal processor number and the speedup ratio were evaluated. Theoretical analysis and the simulation results show that COW is a good choice for the parallel clustering algorithm with deterministic annealing. High speedup ratios are achieved for clustering problems with large clusters with relatively low communication to computation ratios.