Initialization of Recurrent Networks Using Fourier Analysis

摘要：

Time-Delayed recurrent neural network models preserve information through lime and are more powerful than the static feedforward networks, especially in dynamic problems. At. present, recurrent networks are mostly formulated as nonlinear autoregresaion models 4 when applied to time series prediction problem In this paper, we use a novel approach to interpret the recurrent networks. We build the linkage between Fourier analysis and recurrent networks. The major advantage of our method is that it provides a means to initialize the weights. This initialization significantly shortens the training time.

作者: Lai-Wan CHAN Ying-Qian ZHANG

作者单位: Department of Computer Science and Engineering The Chinese University of IIong Kong Shatin, Hong Kong

会议类型: 国际会议

会议名称: 8th International Conference on Neural Information Processing(ICONIP 2001)(第八届国际神经信息处理大会)

会议地点: 上海

会议语种:英文

页码: 800-805

在线出版日期: 2001-11-14（万方平台首次上网日期，不代表论文的发表时间）