Abstract
Simulation programs for neural networks are notorious for being intensive in computations. This poster presents a toroidal lattice architecture used to simulate fully connected neural networks. An attempt is made to see the problem in its global complexity as the system with its own behaviour. First problem is defined, then the effective solution by virtual processors arranged in toroidal lattice architecture is proposed. Then decomposition, data distribution and mapping issues are explained for physical message passing architecture. Finally experimental results are presented.
References
Fujimoto, Y., Fukuda, N., Akabane, T.: Massively Parallel Architectures for Large Scale Neural Networks Simulations, IEEE Transactions on Neural Networks, vol. 3, No. 6, (Nov.1992), 876–887
Paugam-Moisy, H.: A Spy of Parallel Neural Networks Tech. rep. 90-27 Ecole Normale Superieure de Lyon, IMAG, (1990)
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1994 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Hanzalek, Z. (1994). Simulating neural networks on telmat T-node. In: Gentzsch, W., Harms, U. (eds) High-Performance Computing and Networking. HPCN-Europe 1994. Lecture Notes in Computer Science, vol 796. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0020408
Download citation
DOI: https://doi.org/10.1007/BFb0020408
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-57980-9
Online ISBN: 978-3-540-48406-6
eBook Packages: Springer Book Archive