摘要
本文介绍了引入信赖域优化理论解决神经网络中学习问题的新算法,提出了计算有效信赖域步方法,以保证信赖域算法的正确性,采用变系数方法避免了信赖域半径自适应调整过程中不稳定和低效的问题。实验表明,信赖域学习算法优于变尺度算法。
A new learning algorithm based on the trust region optimization theory is introduced for training neural networks.The method of obtaining valid steps ensures the correctness of the trust region algorithm. A self adjusting method with variable coefficients is proposed to resolve the problem of oscillatory behaviors and low efficiency in the progress of trust region radius adjusting. Experimental results show that it is more efficient than self scaling algorithms.
出处
《计算机工程与科学》
CSCD
2005年第6期75-77,共3页
Computer Engineering & Science
关键词
神经网络
信赖域算法
变尺度算法
学习算法
neural networks
trust region algorithm
self scaling algorithm
learning algorithm
