Title

BP神經網絡中定性信息的使用

Translated Titles

The Using of Qualitative Information in BP Neural Networks

Authors

楊華波(Hua-Bo Yang);張士峰(Shi-Feng Zhang);蔡洪(Hong Cai)

Key Words

神經網絡 ; 定性信息 ; 虛擬樣本 ; 非線性函數 ; neural network ; qualitative information ; fictitious sample ; nonlinear function

PublicationName

計算技術與自動化

Volume or Term/Year and Month of Publication

24卷3期(2005 / 09 / 01)

Page #

15 - 17

Content Language

簡體中文

Chinese Abstract

利用BP神經網絡方法模擬各種非線形函數,通常的訓練方法需要大量的訓練樣本。本文通過引入虛擬樣本,提出了使用非線性函數定性信息(如一階偏導數信息)的網絡訓練方法,利用均分佈的隨機數對網絡參數進行再調整。算例說明該方法對於定性信息的利用非常成功,在不降低網絡精度的要求下,有效減少了訓練樣本數量。

English Abstract

The Back-Propagation Neural Networks (BPNN) usually need a great deal of training populations when it approxi-mats a nonlinear function. The paper puts forward a training method using qualitative information (such as one-order partial derivative) by importing fictitious sample. The simulation result shows the usage of qualitative information is rational, the method can reduce effectively quantity of populations with the same network mapping precision.

Topic Category 基礎與應用科學 > 資訊科學
工程學 > 電機工程