BROWSE
Detail
Conference Paper
Energy-efficient Training of SVM on Multi core Platforms
Cited - time in 
Share 
Share
- Authors
- Heegon Kim, Sungju Lee, Yongwha Chung, Daihee Park, Hansung Lee
- Issue Date
- 2012-02
- Citation
- International Conference on Applied and Theoretical Information Systems Research (ATISR) 2012, pp.1-8
- Language
- English
- Type
- Conference Paper
- Abstract
- Recent developments in multicore processors have enabled high performance implementations of machine learning algorithms such as Support Vector Machine (SVM). In this paper, we parallelize the training phase of a SVM in order to reduce the energy consumption. After analyzing the computational characteristics of the SVM and the machine characteristics, we distribute the most time consuming steps to multiple cores while satisfying the data dependencies. Our experimental results show that the 4-core based parallel implementation can reduce the energy consumption by a factor of 1.5 with the exactly same quality as the sequential SVM training.
- KSP Keywords
- Data Dependencies, High performance, Machine Learning Algorithms, Machine characteristics, Parallel implementation, Recent developments, SVM training, Support VectorMachine(SVM), efficient training, energy consumption, energy-efficient