均衡竞争神经元获胜机会的方法  

Method for balancing winning chances of competitive neurons

作　　者：冯乃勤[1] 董亚杰[1] 南书坡[1] 郭战杰[1] 

机构地区：[1]河南师范大学计算机与信息技术学院,河南新乡453007

出　　处：《计算机工程与设计》2009年第4期971-973,共3页Computer Engineering and Design

摘　　要：竞争型神经网络存在"死点"问题,某些神经元在竞争中可能始终未能获胜而成为"死神经元",不仅造成神经元的浪费,而且造成训练误差偏大,无法达到训练误差的精度要求,不能很好完成它所担负的聚类或分类任务。针对该问题,研究竞争型神经网络的切入点,深入探讨了LVQ神经网络并且通过引入阈值学习规则,均衡神经元获胜的机会,较好地解决了该类网络在遇到"死"点时训练误差偏大的问题,仿真实验结果表明了该方法的有效性。Nowadays competitive neural network has been widely used in artificial intelligence and other aspects. But there is a problem exists in the competition neural networks which is called blind spot： Some neurons have never won in competition so become dead neurons. This problem only causes a waste of neurons, but also results in the training error too great to meet the request of training precision and hardly completes the task of classification or clustering. Aiming at this problem, the learning vector quantization neural networks is probed into, settles the problem of large training error properly when it comes to bind spot in such networks by inducting threshold learning rules and balancing the neurons＇ winning chances. Finally emulational experiment proves the validity of this method.

关 键 词：竞争 学习矢量量化 死点 训练误差 阈值 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]