一种用于MIMD平台的并行遗传／神经网络混合学习算法  

A Parallel Genetic/Neural Network Hybrid Learning Algorithm for an MIMD Platform

作　　者：孙亚军[1] 曹磊[1] 虞厥邦[1] 

机构地区：[1]电子科技大学

出　　处：《重庆邮电学院学报（自然科学版）》1996年第2期42-52,共11页Journal of Chongqing University of Posts and Telecommunications(Natural Sciences Edition)

摘　　要：本文提出了一种用于多层前馈神经网络训练的新算法，它把遗传算法与自适应变形梯度学习算法集成起来。这种并行混合学习算法已经在ＭＩＭＤ平台实现了。通过把该算法用于二个图像识别领域例子，对它的性能进行了评估。本文还显示了所提出的并行混合神经网络学习算法的良好的收敛性。A new algorithm is presented for training multilayer feedforward neural networks by integrating a genetic algorithm with an adaptive conjugate gradient neural network learning algorithm. This parallel hybrid algorithm hasbeen implemented on an MIMD platform. The performance of the algorithm has been evaluated by applying it to two examples in the domain of image recognition. This parallel hybrid neural network learning algorithm presented in this paper has superior convergente properties.

关 键 词：神经网络 遗传算法 混合算法 MIMD 学习算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]