| 激励函数可调的前馈神经网络在线学习算法 |
| |
| 引用本文： | 宋轶民,熊熊.激励函数可调的前馈神经网络在线学习算法[J].天津大学学报(自然科学与工程技术版),1999,32(2):201-203. |
| |
| 作者姓名： | 宋轶民 熊熊 |
| |
| 作者单位： | 天津大学机械工程学院 |
| |
| 摘 要： | 针对标准ＢＰ算法的不足，提出了一种激励函数可调的前馈神经网络在线学习算法。该方法去除了样本的预处理过程，以在线方式训练前馈神经网络，同时修正网络的权重，阈值与激励函数。，上于避免了神经元的输出饱和现象，算法的收敛速度明显提高，并在一定程度上防止了局部极小解的出现。仿真结果证明了这种算法的有效性。
|
| 关 键 词： | 神经网络 学习算法 激励函数 前馈神经网络 |
A FAST LEARNING ALGORITHM FOR FEEDFORWARD NEURAL NETWORKS WITH VARIABLE SATURATED FUNCTIONS |
| |
| Song,Yimin,Xiong,Xiong,Ma,Wengui,Zhang,Ce.A FAST LEARNING ALGORITHM FOR FEEDFORWARD NEURAL NETWORKS WITH VARIABLE SATURATED FUNCTIONS[J].Journal of Tianjin University(Science and Technology),1999,32(2):201-203. |
| |
| Authors: | Song Yimin Xiong Xiong Ma Wengui Zhang Ce |
| |
| Abstract: | A fast learning algorithm is proposed for feedforward neural networks (FNN) with variable saturated functions.All the parameters,including the connecting weights,the offsets and the saturated functions,are modified by means of on line learning.It prevents the neurons from saturating and therefore makes the prescaling procedure unnecessary.The nonlinear mapping capability of FNN is improved significantly.Simulation results prove the validity of the proposed algorithm. |
| |
| Keywords: | neural networks learning algorithm saturated function on line |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
