Approach to Parallel Training of Integration Historical Data Neural Networks

V. Turchenko (Ukraine), C. Triki (Italy), and A. Sachenko (Ukraine)

Keywords

Neural network, parallel training, historical data, high performance computing.

Abstract

There is considered the application of neural networks for accuracy improvement of sensor signal processing by sensor drift prediction. The two methods of sensor drift prediction are described. The main characteristics and training time of integrating historical data neural network are presented on the uniprocessor computer Pentium-III 600-128 Mb RAM. The two paralleling schemes of training of integrating historical data neural network are proposed. The fulfilled experimental researches of parallel programs on the high-performance computer Origin2000 are shown high efficiency of second paralleling scheme.

Important Links:



Go Back


IASTED
Rotating Call For Paper Image