MBPGPU: A Supervised Pattern Classifier for Graphical Processing Units

Authors

Abstract

Neural networks have proven to be suitable for pattern recognition and classification tasks. This is validated by an on growing number of successfully applications on a wide variety of areas. However, as the interest in pattern recognition is shifting to more challenging and computationally demanding problems, the long training times required become a serious drawback. In this paper we show that it is possible to overcome this problem by designing a neural network based supervised pattern classifier whose GPU parallel implementation is successfully tested on well-known benchmarks and on a real problem.

Keywords

Multiple Back-Propagation, CUDA, GPU Computing, Parallel Programming

Subject

GPU Computing, Neural networks

Conference

15th edition of the Portuguese Conference on Pattern Recognition - RECPAD 2009, October 2009