J4 ›› 2010, Vol. 28 ›› Issue (02): 147-.

• 论文 • 上一篇    下一篇

回归分析人工神经网络

林和平,张秉正,乔幸娟   

  1. 东北师范大学 计算机学院,长春 |130117
  • 出版日期:2010-03-25 发布日期:2010-06-10
  • 通讯作者: 林和平(1956— ),男,长春人,东北师范大学教授,硕士生导师,主要从事人工智能和系统开发方法论研究,(Tel)86-13500810142 E-mail:linhp@nenu.edu.cn
  • 作者简介:林和平(1956— ),男,长春人,东北师范大学教授,硕士生导师,主要从事人工智能和系统开发方法论研究,(Tel)86-13500810142(E-mail)linhp@nenu.edu.cn;张秉正(1984— ),男,长春人,东北师范大学硕士研究生,主要从事人工智能和系统开发方法论研究|(Tel)86-13514407129(E-mail)zhangbz095@nenu.edu.cn
  • 基金资助:

    国家自然科学基金资助项目(60473042);(60573067)

Regression Analysis Artificial Neural Network

LIN He-ping| ZHANG Bing-zheng|QIAO Xing-juan   

  1. College of Computer Science,Northeast Normal University,Changchun 130117,China
  • Online:2010-03-25 Published:2010-06-10

摘要:

为避免每次训练都必须随机生成样本序列的问题,提出网络动态拓扑的概念,对各种前向式网络进行统一表述;提出正、反序训练方法,并给出解的唯一性证明,同时,网络连接权在初始化时不再需要随机生成。回归分析人工神经网络有效解决了两次随机过程对训练结果造成的不利影响,在稳定性和可信性上对人工神经网络的应用提供了理论依据和技术支持。

关键词: 人工智能, 人工神经网络, 回归分析

Abstract:

The concept of network dynamic topology is proposed, for a variety of pre-integrated presentation of the network to the ceremony; make positive and negative sequence training methods, and gives proof of uniqueness of solution, avoiding each training sample must be randomly generated sequence of problems, the network connection right to randomly generated initialization is no longer needed. Regression analysis of artificial neural networks an effective solution to two stochastic processes on the training result, the adverse effects on the stability and credibility of the application of artificial neural networks provides a theoretical basis and technical support.

Key words: artificial intelligence(AI), artificial neural network(ANN), regression analysis(RA)

中图分类号: 

  • TP183