Parallelization of Partitioning Around Medoids (PAM) in K-Medoids Clustering on GPU
Adhi Prahara, Dewi Pramudi Ismi, Ahmad Azhari · Knowledge Engineering and Data Science · 2020
Due to the high computational complexity, the efficiency of k-medoids clustering becomes a major concern in the k-medoids algorithm improvement.Researchers have been working on attempts to improve the performance of k-medoids clustering [11][12].In general, the efforts on improving kmedoids clustering focus on three different approaches [13] such as 1) empowering the local search and global search for medoids selection, 2) the number of data to be used for the medoids calculation: use the entire data (PAM: Partitioning Around Medoids) algorithm or just a sample of the data (CLARA: Clustering Large Application) algorithm, and 3) the computation method: serial or parallel.