摘要
针对神经网络中具有大量的相似解,提出了一种规范化方法,大大地减少了神经网络中解的数目,然后,给出了多种不同的算子,通过对简单网络问题的实验,比较了它们的优劣,得出了一个新型的遗传算法NGA。实验表明,NGA比用传统的GA训练神经网络效果要优。
In this paper, we propose a standardized method to reduce the large number of similar solutions. With experiments in simple network problem, several operators are compared and a new genetic algorithm named NGA is presented. The result demonstrates that NGA is better than the traditional GA in training neural networks.
出处
《北方交通大学学报》
CSCD
北大核心
1995年第4期528-532,共5页
Journal of Northern Jiaotong University
关键词
神经网络
遗传算法
相似解
分层前馈
ss: neural networks
genetic algorithm
operator/standardized method
similar solution,feed forward
multi-layer training