A DIFFERENTIAL ADAPTIVE LEARNING RATE METHOD FOR BACK-PROPAGATION NEURAL NETWORKS

摘要：

In this paper a high speed learning method using differential adaptive learning rate (DALRM) is proposed. Comparison of this method with other methods such as standard BP, Nguyen-Widrow weight Initialization and Optical BP shows that the networks learning speed has highly increased. Learning often takes a long time to converge and it may fall into local minimas. One way of escaping from local minima is to use a large learning rate at first and then to gradually reduce this learning rate. In this method which is used in multilayer networks using back-propagation learning algorithm, network error is reduced in a short time using differential adaptive learning rate.

关键词： standard BP optical-BP adaptive learning rate Nguyen-Widrow method activation function

作者: Saeid Iranmanesh Mohamad Amin Mahdavi

作者单位: Department of computer engineering Azad university of Qazvin, IRAN Department of computer science Emam Khomeini international university of Qazvin, IRAN

会议类型: 国际会议

会议名称: 2009年中国控制与决策会议(2009 Chinese Control and Decision Conference)

会议地点: 广西桂林

会议语种:英文

页码: 3013-3016

在线出版日期: 2009-06-17（万方平台首次上网日期，不代表论文的发表时间）