摘要
结合实例,给出了递归神经网络的完整设计步骤,包括网络结构的选定,学习算法的选择和网络参数的训练过程。重点研究了学习速率的初始值选取及其调整顺序。给出的递归网络的设计方法,可以适用于多种递归神经网络。
This paper gives the whole procedures of designing recurrent neural networks combining the actual example,which include the choice of the structure and the learning algorithm of the networks,and the training process of the parameters of the networks.The key point of this paper is focusing on the method of selecting the initial learning rates and the adjusting order of the learning rates.The design method of recurrent neural networks proposed in this paper is applicable to various recurrent neural networks.
出处
《计算机工程与应用》
CSCD
北大核心
2009年第4期79-81,97,共4页
Computer Engineering and Applications
基金
国家自然科学基金No.60774098~~
关键词
递归神经网络
学习速率
参数辨识
权值动量因子
recurrent neural networks
learning rate
parameter identification
momentum factor of weight
