A Parallel Incremental Learning Algorithm for Neural Networks with Fault Tolerance

Authors:
Jacques M. Bahi;Sylvain Contassot-Vivier;Marc Sauget;Aurélien Vasseur
Affiliations:
LIFC, University of Franche-Comté, Belfort, France;LORIA, University Henri Poincaré, Nancy, France;LIFC, University of Franche-Comté, Belfort, France and Femto-St, University of Franche-Comté, Montbéliard, France;Femto-St, University of Franche-Comté, Montbéliard, France
Venue:
High Performance Computing for Computational Science - VECPAR 2008
Year:
2008

Citing 8
Cited 1

Multilayer feedforward networks are universal approximators

Neural Networks
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations

Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations
The cascade-correlation learning architecture

Advances in neural information processing systems 2
Optimal brain damage

Advances in neural information processing systems 2
Second Order Derivatives for Network Pruning: Optimal Brain Surgeon

Advances in Neural Information Processing Systems 5, [NIPS Conference]
Output partitioning of neural networks

Neurocomputing
Neural network based algorithm for radiation dose evaluation in heterogeneous environments

ICANN'06 Proceedings of the 16th international conference on Artificial Neural Networks - Volume Part II
Learn++: an incremental learning algorithm for supervised neuralnetworks

IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews

Efficient domain decomposition for a neural network learning algorithm, used for the dose evaluation in external radiotherapy

ICANN'10 Proceedings of the 20th international conference on Artificial neural networks: Part I

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a parallel and fault tolerant version of an incremental learning algorithm for feed-forward neural networks used as function approximators. It has been shown in previous works that our incremental algorithm builds networks of reduced size while providing high quality approximations for real data sets. However, for very large sets, the use of our learning process on a single machine may be quite long and even sometimes impossible, due to memory limitations. The parallel algorithm presented in this paper is usable in any parallel system, and in particular, with large dynamical systems such as clusters and grids in which faults may occur. Finally, the quality and performances (without and with faults) of that algorithm are experimentally evaluated.