Title :
Parallel subgroup discovery on computing clusters — First results
Author :
Trabold, Daniel ; Grosskreutz, Henrik
Abstract :
Data mining tasks often have very high computational costs. In this paper, we present a parallel computation approach for the local pattern mining task of subgroup discovery. Unlike earlier related approaches, we do not distribute the data to be analyzed, but instead distribute portions of the overall search space to be considered on different computing nodes. Our approach has low communication costs, only submitting messages when new exceedingly good patterns are visited. While the paper describes work-in-progress, we already present first experiments, witnessing a speedup factor of about 34 on 64 computing units.
Keywords :
data analysis; data mining; parallel processing; workstation clusters; computing clusters; local pattern mining task; parallel computation ap- proach; parallel subgroup discovery; subgroup discovery; Algorithm design and analysis; Clustering algorithms; Computational modeling; Context; Data mining; Databases; Heuristic algorithms;
Conference_Titel :
Big Data, 2013 IEEE International Conference on
Conference_Location :
Silicon Valley, CA
DOI :
10.1109/BigData.2013.6691625