Thesis
The parallelization of enhanced K-means clustering techniques for multiple platforms
Washington State University
Master of Science (MS), Washington State University
2015
Handle:
https://hdl.handle.net/2376/100907
Abstract
In this thesis we describe our unique parallelized approaches for two popular k-means seed selection techniques: KKZ and k-means++. We implement parallelized versions of each of these algorithms for three distinct platforms: GPU (Graphics Processing Unit), mulitple multicore processors using OpenMP, and the Cray XMT massively multithreaded architecture. We describe the implementation of each of these algorithms for each of these platforms and the unique differences between them. We test the performance of our implementations to show that our algorithms can perform significantly faster than the traditional serial approaches. We develop a linear regression model to estimate the amount of time required to perform each algorithm on each platform. We also present information comparing the performance of each platform at varying sizes of input data. We conclude with an analysis of the strengths and weaknesses of the performance of each implementation, showing that each platform has the ability to outperform the others depending on the data dimensions and number of clusters.
Metrics
36 File views/ downloads
25 Record Views
Details
- Title
- The parallelization of enhanced K-means clustering techniques for multiple platforms
- Creators
- Patrick Stark Mackey
- Contributors
- Robert R. Lewis (Degree Supervisor)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; [Pullman, Washington] :
- Identifiers
- 99900525067401842
- Language
- English
- Resource Type
- Thesis