CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Performance analysis of distributed asynchronous multilayer perceptron model for Erlang/OTP platform
 |
Read | Plotnikova Natalya P., Fedosin Sergey A., Kazakov Vitaliy G. Performance analysis of distributed asynchronous multilayer perceptron model for Erlang/OTP platform // Caspian journal : management and high technologies. — 2013. — №4. — pp. 86-94. |
Plotnikova Natalya P. - Junior Researcher, Ogarev Mordovia State University, 39 B. Khmelnitskiy St., Saransk, 430005, Russian Federation, linsierra@yandex.ru
Fedosin Sergey A. - Ph.D. (Engineering), Professor, Ogarev Mordovia State University, 39 B. Khmelnitskiy St., Saransk, 430005, Russian Federation, fedosinsa@mrsu.ru
Kazakov Vitaliy G. - Researcher, Ogarev Mordovia State University, 39 B. Khmelnitskiy St., Saransk, 430005, Russian Federation, vgkazakov@gmail.com
This article presents research results for two multilayer perceptron asynchronous distributed modeling approaches: model with neurons groups dispatching and model based on “one neuron – one process” principle without dispatching. Implementation of these approaches uses entities and mechanisms of Erlang/OTP platform. Multilayer network models in this research use RPROP training algorithm. Research work includes training process modeling, average training speed measuring and comparison of discussed approaches for different neural networks. The article analyzes advantages and disadvantages of each model type for distributed cluster experimental systems with high speed interconnection bus. Main result of this research is proof of systems without dispatching efficiency.
Key words: multilayer perceptron,neural network training,asynchronous distributed model,model with dispatching,model without dispatching