Maximization of Expectation-Parallelism Algorithm Using OpenCL

Nowadays both organizations and companies store big volumes of data to achieve their purposes. One of the variants to obtain valuable information consists on the employment of Data Mining. Inside Data Mining, different tasks exist and one of them is clustering. In this task the data group according to their likenesses among them differ with elements of other groups. One of the algorithms that carry out these clusters is Expectation-Maximization, which presents high times of execution in their data. This article discusses about the parallelization of the mentioned algorithm, using techniques of parallel programming. The design of the proposed algorithm is based on the use of the graphic process unit, GPU. OpenCL, language used for the programming in hybrid architectures, allows to take advantage of the available hardware architectures, which it is possible to diminish the time of execution of the sequential implementation. The reason to improve this time is due to the quantity of parallel processes that can rush in threads of independent prosecutions. For the achievement of the described results, knowledge of the field of Data Mining and Parallel and Distributed Computation are integrated. As part of this investigation, an implementation of the algorithm using the libraries of OpenCL was carried out to diminish the time of execution. The implementation is able to diminish the sequential implementation in 82%, this means that the parallel algorithm is executed 5,5 times quicker that its sequential corresponding implementation.