机构地区: 上海交通大学电子信息与电气工程学院计算机科学与工程系
出 处: 《高技术通讯》 2000年第5期36-38,26,共4页
摘 要: 提出了两种用于前向神经网络的进化学习算法 ,实验证明它们能有效地在网络权值空间中寻找全局最优解。在比较实验的基础上 ,得出了在神经网络的进化学习过程中变异是起主导作用的遗传算子的结论 ,并以此为指导配置算法的各个关键参数。通过对XOR问题和IRIS模式分类问题的学习证明 ,这两种算法均能获得远高于传统BP算法的性能。 Two evolutionary learning algorithms for feed forward neural network are presented. They are demonstrated to have strong ability of finding global optimization It is concluded that mutation is the dominating operator during the evolutionary procedure of neural network based on the experimentation carried According to this conclusion, two evolutionary learning algorithms are configured and they can all get far superior performance over traditional BP algorithm
领 域: [自动化与计算机技术] [自动化与计算机技术]