计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (20): 43-45.

• 学术探讨 • 上一篇    下一篇

基于郭涛算法的演化神经网络

郭 艳1,康立山1,2,刘福江3,4   

  1. 1.中国地质大学(武汉) 计算机学院,武汉 430074
    2.武汉大学 软件工程国家重点实验室 武汉 430072
    3.中国地质大学(北京) 地球科学与资源学院,北京 100083
    4.中国地质大学(武汉) 信息工程学院,武汉 430074
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-07-11 发布日期:2007-07-11
  • 通讯作者: 郭 艳

Evolutionary neural networks based on GT algorithm

GUO Yan1,KANG Li-shan1,2,LIU Fu-jiang3,4   

  1. 1.School of Computer,China University of Geosciences(Wuhan),Wuhan 430074,China
    2.Key Lab of Software Engineering,Wuhan University,Wuhan 430072,China
    3.School of Earth Science and Resourses,China University of Geosciences(Beijing),Beijing 100083,China
    4.School of Information Engineering,China University of Geosciences(Wuhan),Wuhan 430074,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-07-11 Published:2007-07-11
  • Contact: GUO Yan

摘要: 提出了一种新的演化神经网络算法GTEANN,该算法基于高效的郭涛算法,同时完成在网络结构空间和权值空间的搜索,以实现前馈神经网络的自动化设计。本方法采用的编码方案直观有效,基于该编码表示,神经网络的学习过程是一个复杂的混合整实数非线性规划问题,例如杂交操作包括网络的同构和规整处理。初步实验结果表明该方法收敛,能够达到根据训练样本自动优化设计多层前馈神经网络的目的。

关键词: 演化神经网络, 郭涛算法, 网络同构和规整

Abstract: A new method of evolutionary neural networks,called evolutionary neural networks based GT algorithm(GTEANN),is proposed in this study.In this method,GT algorithm is used to simultaneousely search the satisfied structure and weights for feedforward neural networks.A straightforward effective encoding scheme for feedforward neural networks is adopted and only crossover operators are used with special topological isomorphism and regularization process.The learing process of network is a complex mixed-integer nonlinear optimization problem.The initial results of experiments indicate that GTEANN can automatically design and optimaize neural networks using the training sets.

Key words: Evolutionary Neural Networks, GT algorithm, topological isomorphism and regularization