GENOUD-BP: A novel training algorithm for artificial neural networks

DOI

10.2991/icimm-15.2015.166How to use a DOI?

Keywords

neural networks; optimization; evolutionary algorithms; GENOUD; BP

Abstract

BP Algorithm, as a well-known training method for artificial neural network, has been widely used in all the main fields of science and engineering. However, owing to the overwhelming dependency on gradient of loss function, BP-ANN still suffers several drawbacks, for example, the training process is prone to stuck at local optima, cause early convergence while the whole process is lack of generalization performance. Aimed at improving the current BP training Algorithm, a new algorithm called GENOUD-BP is proposed in this paper by introducing the GENOUD algorithm which combines the global searching power of genetic algorithms and convergence speed of traditional gradient based optimization algorithms. Two UCI datasets are employed to carry out benchmark experiments, the result of which shows that the GENOUD-BP significantly outperforms traditional BP algorithms.

Copyright

© 2015, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Zhiyong Yang
AU  - Shiyuan Zhang
AU  - Taohong Zhang
PY  - 2015/07
DA  - 2015/07
TI  - GENOUD-BP: A novel training algorithm for artificial neural networks
BT  - Proceedings of the 5th International Conference on Information Engineering for Mechanics and Materials
PB  - Atlantis Press
SP  - 908
EP  - 912
SN  - 2352-5401
UR  - https://doi.org/10.2991/icimm-15.2015.166
DO  - 10.2991/icimm-15.2015.166
ID  - Yang2015/07
ER  -