首页 | 本学科首页   官方微博 | 高级检索  
     

前馈神经网络混合训练算法及其应用
引用本文:车四方,戴侃侃,曹飞龙.前馈神经网络混合训练算法及其应用[J].中国计量学院学报,2014(4):424-431.
作者姓名:车四方  戴侃侃  曹飞龙
作者单位:中国计量学院理学院;
基金项目:国家自然科学基金资助项目(No.61272023)
摘    要:介绍了单隐层前馈神经网络的混合训练算法(HFM)和正则化混合训练算法(RHFM),然后将该算法应用于UCI数据库上的实际回归例子中,并将其与BP、NNRW以及FM算法进行了比较.仿真实验表明,HFM算法的收敛速度优于其它几种算法,RHFM算法有较好的泛化性能,而NNRW算法在训练时间上占优,尽管如此,HFM算法在时间上还是大大优于BP算法.说明,混合训练算法是一种有效的算法.

关 键 词:前馈神经网络  混合训练算法  正则化

The hybrid training algorithm for feedforward neural networks and its applications
CHE Sifang,DAI Kankan,CAO Feilong.The hybrid training algorithm for feedforward neural networks and its applications[J].Journal of China Jiliang University,2014(4):424-431.
Authors:CHE Sifang  DAI Kankan  CAO Feilong
Affiliation:(College of Sciences, China Jiliang University, Hangzhou 310018, China)
Abstract:The hybrid training algorithm (HFM) and regularization hybrid training algorithm (RHFM) of feedforward neural networks with single hidden layer were presented. These methods were applied to regression examples from the UCI repository of machine learning databases. BP, NNRW and FM algorithms were also applied to the same problems for comparison. Simulation results show that the HFM algorithm is superior to other techniques in terms of rate of convergence, and the better generalisation performance is RHFM algorithm, while NNRW dominates in the training time. Nonetheless, HFM is much better than BP in the training time. Consequently, the hybrid training algorithm is an efficient procedure.
Keywords:feedforward neural networks  hybrid training algorithm  regularization
本文献已被 CNKI 维普 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号