基于样本重构的神经网络集成学习方法  

Learning approach for neural network ensemble based on specimen reconstruct

作　　者：巩文科[1] 李长河[1] 石争浩[1] 赵洁[1] 

机构地区：[1]西安理工大学计算机科学与工程学院,陕西西安710048

出　　处：《计算机应用》2006年第6期1428-1430,共3页journal of Computer Applications

摘　　要：在回顾以往神经网络集成的研究成果基础上,提出一种新的负相关学习方法,该方法易于执行,计算量小,有效的消除了学习中的复合线性问题,减小了集成误差,最后用测试用例对该方法进行了考察,证明该方法可以有效的降低集成预测误差,得到较为理想的集成效果。Based the review of the research of neural network ensemble, a new negative,, correlation learning method was proposed, which was both easy to implement and had the less computation, the co-linearity problem was eliminated, the error of the neural network ensemble was reduced. In the end, gives an test to this method, which ）prove that the method can reduee the error of the neural network ensemble effectively and get an expected result.

关 键 词：神经网络集成 负相关学习 样本重构 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]