Research on the Solution of BP Neural Network Training Problem

DOI

10.2991/icaita-18.2018.11How to use a DOI?

Keywords

neural network; gradient diffusion; activation function; convergence speed; classification accuracy

Abstract

BP neural network is a common kind of neural network for recognition and classification, but it is obvious that some difficulties are find during the training process. In this paper, we introduced the basic principle of BP algorithm and propose a parameter adjustment procedure for activation function to speed up convergence and avoid gradient diffusion, and the feasibility of this method is proved by a basic BP neural network with simple structure. We analyzed this parameter's available value through a group of experiments based on simple three-layer BP full connected network. In addition, this activation function parameter adjustment procedure is also used in is suitable for Multilayer BP neural networks.

Copyright

© 2018, 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  - Junyang Tan
AU  - Dan Xia
AU  - Shiyun Dong
AU  - Binshi Xu
AU  - Ye Li
PY  - 2018/03
DA  - 2018/03
TI  - Research on the Solution of BP Neural Network Training Problem
BT  - Proceedings of the 2018 2nd International Conference on Artificial Intelligence: Technologies and Applications (ICAITA 2018)
PB  - Atlantis Press
SP  - 41
EP  - 45
SN  - 1951-6851
UR  - https://doi.org/10.2991/icaita-18.2018.11
DO  - 10.2991/icaita-18.2018.11
ID  - Tan2018/03
ER  -