摘要
极限学习机(ELM)在训练过程中无须调整隐层节点参数,因其高效的训练方式被广泛应用于分类和回归,然而极限学习机也面临着结构选择与过拟合等严重问题。为了解决此问题,针对隐层节点增量数目对收敛速度以及训练时间的影响进行了研究,提出一种利用网络输出误差的变化率控制网络增长速度的变长增量型极限学习机算法(VI-ELM)。通过对多个数据集进行回归和分类问题分析实验,结果表明,提出的方法能够以更高效的训练方式获得良好的泛化性能。
The extreme learning machine adjusted the hidden layer nodes to obtain a certain network output error and greatly raised the training speed, because it didn＇t need to adjust the hidden layer node parameters in.the training process. However, structure selection and over fitting limited the development of extreme learning machine. To solve the problem ,this paper ana- lyzed the effect of the number of hidden layer nodes on the convergence speed and training time, and derived a new network construction mode named variable length incremental extreme learning machine （VI-ELM） , its output error rate controlled the network incremental speed. The experimental results show that the proposed method can obtain good generalization performance in a more efficient way of training in regression and classification data sets.
作者
王诗琦
赵书敏
耿江东
杨非
蒋忠进
Wang Shiqi Zhao Shumin Geng Jiangdong Yang Fei Jiang Zhongjin(State Xey Laboratory of Millimeter Waves, Southeast University, Nanjing 210000, China Dept. of Prevision Instrument, Tsinghua University, Beijing 100084, China China Airborne Missile Academy, Luoyang Henan 471009, China)
出处
《计算机应用研究》
CSCD
北大核心
2016年第12期3696-3699,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(11303004
11573007)
航空基金资助项目(20140169001)
江苏省自然科学基金资助项目(K20130637)
关键词
极限学习机
增量学习
泛化性能
增量型极限学习机
变长增量型极限学习机
extreme learning machine （ELM）
incremental study
generalization performance
incremental extreme learning machine （I-ELM）
variable length incremental extreme learning machine （VI-ELM）
