Developing Multilayer Neural Network Applications on a Distributed Memory Parallel Machine

• Eugenio Suárez Caner UFRJ
• Jose Manoel de Seixas UFRJ

DOI: https://doi.org/10.5753/sbac-pad.1998.22691

Resumo

This paper concern the development of multilayer artificial neural network applications in a transputer (T9000) based parallel machine with distributed memory. For defining the partition of processing tasks within the machine, the natural parallelism exhibited by neural network is explored. Both training and production phases can be implemented and the designer defines the set of parameters required by the backpropagation training method through an user interface. Neural Preprocessing methods based on topological mapping and principal component analysis are also available to be integrated into a neural network based hybrid system design. As a case study, the principal component analysis for a particle discriminator in experimental physics is developed.

Referências

E. Littmann: Trends in Neural Network Research and an Application to Computer Vision. New Computing Techiniques in Physics Research III, World Scientific, pp 253-268, 1994.

S. Haykin: Neural Networks, a Comprehensive Foundation. Macmillan, 1994.

Telmat Multinode. TN-310 System Manual and Training Set. France 1995.

The SGS Thomson. T9000 Transputer Hardware Reference Manual. France, 1993.

Edited by M.D. May, P.W. Thompson, P.H. Welch. Networks, Routers & Transputers: Function, Performance and Application. IOS Press, Netherlands, 1993.

J. Henz, A. Krogh, R.G. Palmer: Introduction to the Theory of Neural Computation. Addison-Wesley, 1991.

T. Nordström. B. Svensson: Using And Design Massively Parallel Computers for Artificial Neural Networks, Journal Of Parallel And Distributed Computing 14, pp. 260-285, 1992.

M. Besh, H.W. Pohl: PROMOTER- Application Study. How to Simulate Artificial Neural Network on Large Scale Parallel Computers Exploiting Data Parallelism and Object-Orientation, Technical Report TR-94020. Nov. 1994.

L. Coetzee: Parallel Approaches to Training Feedforward Neural Nets. PHD Thesis. Univ. of Pretoria. Feb. 1996.

Rabello dos Santos. Projeto Final No. 38. Sistema de Clasificação de uma Maquina com Processamento Distribuído. DEL. UFRJ. I997.

J.M. Seixas, L.P. Caloba, B. Kastrup: A Neural Second-Level Trigger System Based on Calorimetry and Principal Component Analysis. New Computing Techniques in Physics Research IV, pp 545-550, World Scientific, 1995.

Telmat Multinode. DSP HTRAM Software. User Manual, France, 1995.