摘要
分析了具有遗忘特性及信息锁存能力的状态回归神经网络的计算方法。针对多输入多输出时序样本,提出了更能反映网络短时记忆能力以及时序样本数据物理特性的同时刻反馈控制和计算方法。实验结果显示,该文提出的方法对时序样本的学习和记忆不但具有更高的准确性,而且不增加计算的复杂性。
An analysis is made in the paper about the computing method of the recurrent neural networks that has the characteristic of oblivion and the ability of information latching.To the time-sequence sample which has both multiple inputs and multiple outputs,a new computing method called the same time point feedback control computing method is presented that can better embodies the short time memory ability of the network and the future of the time-sequence sample data.Experiment results show that the proposed method is not only more accurate for learning and remembering of time-sequence sample,but also the complexity is not increased in its calculation.
出处
《计算机工程与应用》
CSCD
北大核心
2002年第21期32-33,87,共3页
Computer Engineering and Applications
基金
国家自然科学基金资助项目(编号:69783008)
广东省自然科学基金资助项目(编号:970461)
