Parallel execution of SVM training using graphics processing units (SVMTrGPUs)

摘要

Parallel computing is a simultaneous use of multiple compute resources, for example, processors to solve complex computational problems. It has been used in high-end computing areas such as pattern recognition, medical diagnosis, national defense, and web search engine. This paper focuses on the implementation of pattern classification technique, Support Vector Machine (SVM) using vector processor approach. We have carried out a performance analysis to benchmark the sequential SVM program against the Graphics Processing Units (GPUs) optimization. The result shows that the parallelization of SVM training duration achieves a better performance than the sequential code speedups by 6.40.