V. Sobetskyy and S. Grzegorski (Poland)
Neural networks, parallelization algorithms
The problems of artificial neural networks learning for recognition and tracking tasks are taken up in this article considered. Finding an optimal neural network model and processing the neural network training in view of the computing complexity demands application of multiprocessing systems, that's the basic difficulties of research. Maintenance of the maximal productivity clusters systems during work and training artificial neural networks depends on a degree algorithm's parallelization and parallelization approach. Tha article shows that the parallel versions of the backpropagation and Levenberg-Marquardt's algorithms are accuracy and effective for training feedforward neural networks.
Important Links:
Go Back
