摘要
传统的神经网络学习算法(如BP算法)需要调整大量的网络参数,例如输入权值以及隐层单元的偏置,而极速学习机只需要设置网络的隐层节点个数,在算法执行过程中不需要调整网络的输入权值,便可以产生唯一的最优解,因此它具有学习速度快且泛化性能好的优点.随着极速学习机的研究发展,核极速学习机的相关理论被提出.核极速学习机是将核函数引入到极速学习机中,可以得到最小二乘解,具有更稳定的泛化性能.本文在核极速学习机的基础上提出了一种基于Bagged聚类核的核极速学习机的分类方法,首先对已有的标记样本和所有的无标记样本采用多次k均值聚类,去构造Bagged聚类核,然后对Bagged聚类核和径向基核进行求和,最终用于核极速学习机的训练中.与传统核极速学习机相比,本文提出的方法可以使用所有的无标记样本,从而尽可能地提高分类的准确率.最后本文通过实验验证了方法的可行性.
The traditional neural network learning algorithm(BP algorithm)need to set a large amount of network training parameter,and prone to local optimal solution.Extreme learning machine(ELM)need to set the number of hidden layer nodes of networks,while execution of the algorithm does not need to adjust the network weights of the input and hidden element bias,and can produce the optimal solution,thus it has advantages of fast learning speed and good generalization capability.Extreme learning machine as a kind of machine learning method,with simple and easy to use,and effective single hidden layer feed forward neural network learning algorithm,caught the attention of more and more researchers.With the research and development of extreme learning machine,the theory of nuclear extreme learning machine has been continually raised.Nuclear ultimate learning machine is introduced to limit the kernel learning machine,with which you can get a least-squares optimization solution,a more stable,better generalization performance.We now put forward a novel extreme learning machine based on bagged kernel classification method.First of all,the existing tag samples and all unmarked samples use k-means clustering algorithm for many times to construct the bagged clustering nucleus.Then,bagged clustering nucleus and the radial basis calculate the sum,and eventually it is used in training and classification of extreme learning machine.Compared with the traditional extreme learning machine,the new algorithm can use all unmarked sample information,as much as possible to improve the classification accuracy,and further improve the running speed.Through the experimental data set,we verify the feasibility of the method.
作者
王丽娟
丁世飞
Wang Lijuan;Ding Shifei(School of Computer Science and Technology,China University of Mining and Technology,Xuzhou 221116,China;School of Information and Electrical Engineering,Xuzhou College of Industrial Technology,Xuzhou 221140,China)
出处
《南京师大学报(自然科学版)》
CAS
CSCD
北大核心
2019年第3期145-150,共6页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金项目(61672522、61379501)