GPUMLib: A New Library to Combine Machine Learning Algorithms with Graphics Processing Units

Authors

Abstract

The Graphics Processing Unit (GPU) is a highly parallel, multi-threaded, many-core device with enormous computational power, especially well-suited to address Machine Learning (ML) problems that can be expressed as data-parallel computations. As problems become increasingly demanding, parallel implementations of ML algorithms become critical for developing hybrid intelligent real-world applications.
The relative low cost of GPUs combined with the unprecedent computational power they offer, makes them particularly well positioned to fulfill the need to automatically analyze and capture relevant information from large amounts of data.
Although in ML field there are countless powerful learning algorithms suitable for a wide range of applications, the true potential of these methods is underused, because many implementations are not openly shared. In the GPU arena the panorama is even worse, because few algorithms have yet been implemented. In order to mitigate this problem we propose the creation of an open source GPU Machine Learning Library (GPUMLib) that aims to provide the basis and the building blocks for the scientific community to develop GPU ML algorithms.
Experimental results on benchmark datasets demonstrate that the GPUMLib algorithms already implemented achieve significant savings over the counterpart CPU implementations. Future work is foreseen towards extending the GPUMLib and its validation in complex hybrid systems.

Keywords

GPU Computing, Machine Learning

Subject

GPU Computing, Machine Learning

Conference

10th International Conference on Hybrid Intelligent Systems (IEEE), August 2010

DOI