摘要
针对前馈神经网络结构设计困难，传统ＢＰ算法易陷入局部极小、收敛速度慢、对初始权选取敏感等缺陷，通过修改误差函数，提出了一种融结构自适应选择和参数学习于一体的针对一般神经元激活函数的新算法，实验结果表明其高效性．
Aiming at the conventional BP algorithm that has some problems such as difficulity decision for structure, easily trap into local minimum, lower rate of convergence, noise sensitivity. This paper presents an adaptive BP algorithm, called 'structureparameter learing algorithm', that it is not only a structure learning but a parameter learning. The simulation results prove the new algorithm's superiority to traditional ones on learning speed and quality.
出处
《湘潭大学自然科学学报》
CAS
CSCD
1998年第4期10-12,共3页
Natural Science Journal of Xiangtan University
基金
国家自然科学基金
湖南省自然科学基金
