Design of neural networks with the hidden-layer control part and memory part

Author

Lim, Chunhwan ; Kim, Jaimin ; Han, Seungjo ; Park, Jongan

Author_Institution

Chosun Univ., Kwangju, South Korea

Volume

2

fYear

1996

fDate

5-10 Aug 1996

Firstpage

893

Abstract

In this paper, a neural network model, which easily controls weights and unit offsets, is proposed. It consists of multilayer neural networks with a control part and an address memory part. The control part controls weights between the input layer and the hidden layer and unit offsets of the hidden layer, and sends the output data to the address memory part. The address memory part memorizes the output pattern of the hidden layer, which is compared with the input pattern, and sends the learning data to the output layer after learning. Simulation results show that the weights control and unit offsets control between layers are easy, convergence speed is fast, and it does not fall into the local minima during learning

Keywords

learning (artificial intelligence); multilayer perceptrons; address memory part; convergence speed; hidden-layer control part; input layer; memory part; multilayer neural networks; multilayer perceptron adaptation; multilayer perceptron learning; neural networks design; output data; output pattern memorisation; Backpropagation algorithms; Electronic mail; Iterative algorithms; Iterative methods; Learning systems; Multi-layer neural network; Multilayer perceptrons; Neural networks; Neurons; Weight control;

fLanguage

English

Publisher

ieee

Conference_Titel

Industrial Electronics, Control, and Instrumentation, 1996., Proceedings of the 1996 IEEE IECON 22nd International Conference on

Conference_Location

Taipei

Print_ISBN

0-7803-2775-6

Type

conf

DOI

10.1109/IECON.1996.565996

Filename

565996