Training Acceleration of Multi-Layer Perceptron using Multicore CPU and GPU under MATLAB Environment

Shefa A. Dawwd / Assist. Prof. Noor M. AL Layla shefa.dawwd.2014@ieee.org aneng.noor@gmail.com Computer Engineering Department College of Engineering, University of Mosul, Mosul-Iraq Abstract Training of Artificial Neural Networks (ANNs) for large data sets is a time consuming mission. In this paper, accelerating the training of artificial neural network is achieved by a parallel training using either Multicore Central Processing Unit (CPU) or General Purpose Graphics Processing Unit (GPGPU). The training is implemented using five datasets with diverse amounts of patterns and with different neural network parameters in Multilayer Perceptron (MLP). The results show a significant increase in computation speed, which is increased nearly linear with the number of cores in multicore processor for problems with medium and large training datasets. Also, a considerable speed up is achieved when the GPU is used to train the MLP with the large training datasets. While a single core processor is a better choice when the data set size is small. The optimal number of cores or the type of the parallel platform should be employed according to the load of computation.