多层前传神经网的广义误差反传训练与模式分类

GENERALIZED ERROR BACK-PROPAGATION TRAINING FOR MULTI-LAYERED FEEDFORWARD NEURAL NETS

本文以天然留兰香的组分构成与其品质的关系为例,讨论人工神经元方法用于复杂信息模式分类的问题。提出一种广义的误差反传训练策略,将网络的训练范围从联接权扩大到神经元模型。这种新的训练方法(GBP)能提高多层前传网络的学习效率,加快收敛的速率。实际运行的结果表明,所需训练时间仅为普通误差反传(BP)训练方法的1/15,并能达到较高的预报精度。

A new strategy called Generalized Error Back-Propagation(GBP)method, which applies the generalized delta rule to change not only the weights but also the parameters of processing functions of neurons,is proposed for training Multi-Layered Feedforward (MLF) Artificial Neural Nets (ANN). As comparing with the conventional Error Back-Propagation(BP) method, the new method is much faster in learning,and far better in performing pattern classification of complex chemical information. The classification of spearmint essence sample is used as an evaluation example. The results show the advantages of the proposed method clearly.