A Parallel, Structural Comparison Scheme of Time-Series Implemented on a PC Cluster

This paper presents a scheme for pallarel computation of similarity matrix using a PC cluster. Pairwise comparison of N sequences essentially requires N(N + 1)/2 computational tasks; therefore, a simple and efficient comparison method such as Euclidean distance or Dynamic Time Warping is often used. Focusing on the fact that the pair-wise comparison can be processed independently for different pairs of sequences, we employ a palallel computation scheme with a PC cluster and attempt to include more complex, structure-based pairwise comparison method. Experimental results on the cylinder- bell-funnel data set demonstrate that the combination of parallel computation and structural comparison may be used for handling a difficult clustering task of time series.