Un problème d'ordonnancement en milieu hétérogène appliqué aux réseaux de neurones

Abstract : Neural networks are well known as universal approximators. But the performance of neural network depends on several parameters. Finding the best set of parameters has no theoretical answer. Neural network users are forced to try different sets of parameters, and then, choose the best among them. In this paper, we present a parallel solution, based on a master-slave software architecture. Tests are reported on two platforms (workstations network and Cray T3D), using the PVM environment. The load-balancing problem is also addressed. Results are given, for the approximation and the speed-up of the parallel execution.
Document type :
Reports
Complete list of metadatas

Cited literature [7 references]  Display  Hide  Download

https://hal-lara.archives-ouvertes.fr/hal-02102077
Contributor : Colette Orange <>
Submitted on : Wednesday, April 17, 2019 - 9:13:41 AM
Last modification on : Wednesday, May 1, 2019 - 1:27:34 AM

File

RR1996-15.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02102077, version 1

Collections

Citation

Richard Baron, Xavier-Francois Vigouroux. Un problème d'ordonnancement en milieu hétérogène appliqué aux réseaux de neurones. [Research Report] LIP RR-1997-15, Laboratoire de l'informatique du parallélisme. 1996, 2+13p. ⟨hal-02102077⟩

Share

Metrics

Record views

5

Files downloads

8