We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Effective Implementation of the EM-algorithm using GPGPU.
- Authors
Kasitskyj, O. V.; Bidyuk, P. I.; Korshevnyuk, L. O.
- Abstract
The problem of decreasing of running time for the data processing algorithms is very important especially when they are used in real time. For example, in real time image processing, process control systems, speech recognition, etc. The paper considers the possibility of decreasing running time of the expectation maximization (EM) algorithm using modern computing systems. The proposed modified EM-algorithm is aimed at better parallelism for the general purpose graphical processing unit (GPGPU).The experimental results are obtained with solving of the classical problem of Gaussian random variables mixture separation. The proposed implementation of the algorithm was performed on one and two 8-core processor (CPU) setup, as well as on the general purpose graphical processing unit. The graphics processor, because of its abilities for parallel computations and due to the properties of the EM-algorithm considered, showed substantially higher effectiveness in all the computational experiments. Besides, the modified EM-algorithm showed almost two times faster performance on GPGPU than on one or two CPU using large sample sizes (from 5 million values and higher). The lower price of graphics processor is an additional advantage of the approach proposed for such parallel algorithms and GPGPU usage.
- Publication
Naukovi visti NTUU - KPI, 2013, Vol 91, Issue 5, p35
- ISSN
1810-0546
- Publication type
Article