北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2002, Vol. 25 ›› Issue (2): 14-17.

• 学术论文 • 上一篇    下一篇

并行协作模块化神经网络体系结构

凌卫新1,郑启伦2,陈琼2,吕翠英1   

  1. 1.华南理工大学应用数学系, 广州 510640;2.华南理工大学计算机科学与工程学院, 广州 510640
  • 收稿日期:2001-07-16 出版日期:2002-03-10
  • 作者简介: 凌卫新(1966—),女,华南理工大学讲师,在读博士
  • 基金资助:
    国家自然科学基金资助项目(69783008); 国家博士点基金资助项目(98056117); 广东省自然科学基金资助项目(990582); 华南理工大学自然科基金资助项目

A Modular Neural Network Architecture #br# Based on Parallel Cooperation

LING Wei-xin1, ZHENG Qi-lun2, CHEN Qiong2, LU Cui-ying1   

  1. 1. Dept. of Applied Mathematics, South China Univ. of Tech., Guangzhou510640, China;
    2. Dept. of Computer Engineering and Science, South China Univ. of Tech., Guangzhou 510640, China)
  • Received:2001-07-16 Online:2002-03-10
  • Supported by:
     

摘要: 提出了一种并行协作模块化神经网络的体系结构和学习算法,它可实现复杂任务的自动分解判定和模块化训练策略。实验表明,本文提出的体系结构和算法与非模块化神经网络技术相比,提高了训练速度、改善了网络性能,它具有高效并行运行效率、容易实现新增样本学习等特点。

关键词: 神经网络, 模块化结构, 学习算法, 分解判定

Abstract: The architecture and algorithm of Parallel Cooperative Modular Neural Network (PCMNN) are proposed in this paper. It has theadvantage of automatic decomposition of given task and its modular training of network. Experiments proposed in this paper is practicable and effective and is superior to non-modular method due to its high efficiency for a parallel network system structure, easy implementation of additional learning and modular structure for implementation by hardware.

Key words: neural network, modular structure, learning algorithm, decomposition decision

中图分类号: