摘要
针对典型的对角递归神经网络,推导出递归神经网络稳定条件下网络输出层、隐含层及关联层学习速率的具体取值范围。提出设计者可通过从具体系统中获得的数据确定网络各层学习速率的上、下界数值,确定自适应学习速率的初值与调节方法,并可选取最优学习速率。因而该法具有很强的可操作性和实用性。还给出一个具体数值实例,说明自适应学习速率与最佳学习速率的调整过程。
The bounds of learning rates in network's output layer, hidden layer and connected layer of diagonal recurrent neural networks are deduced in the condition of guaranteeing the stability of the networks. The contribution is given to a method that the bounds of learning rates in networks' different layers can be obtained by the data from the individual system, from which the initial value and method of adaptive learning rate can be confirmed and the optimum learning rates can be gained. The method proposed has good operability and practicability. A numerical demonstration to explain the procedure how to get the adaptive and optimum learning rates is given.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2005年第5期942-947,共6页
Systems Engineering and Electronics
基金
安徽省自然科学基金资助课题(03042301)
