摘要
提出了一个改进的自适应变步长ＢＰ网络学习算法，对多个布尔学习问题以及Ｆｉｓｈｅｒ收集的一个植物分类问题进行计算．结果表明，该算法不仅有相当快的收敛速度，而且在避免学习过程陷入局部极小方面也取得较好结果．
On the basis of the work (6), we present an improved adaptive BP learning algorithm with variable stepsizes in which the nonmonotone line search technique from unconstrained optimization is employed. The algorithm has been tested on several Boolean learning problems as well as a well-known plant classification problem due to Fisher, and the results suggest that the algrithm not only converges in a quite fast way, but also avoids stagnation problem to some extent.
出处
《福州大学学报（自然科学版）》
CAS
CSCD
1998年第4期19-21,共3页
Journal of Fuzhou University(Natural Science Edition)
基金
福建省自然科学基金
关键词
神经网络
搜索方法
学习算法
最优化
neural network
search method
learning algorithm
optimization
