[1]孔双双,王开军,林 崧.外类入侵度初始化参数的极限学习机[J].南京师范大学学报(工程技术版),2019,19(03):053.[doi:10.3969/j.issn.1672-1292.2019.03.008]
　Kong Shuangshuang,Wang Kaijun,Lin Song.Extreme Learning Machine with Initialized ParametersBased on External Class Invasion Degree[J].Journal of Nanjing Normal University(Engineering and Technology),2019,19(03):053.[doi:10.3969/j.issn.1672-1292.2019.03.008]
点击复制

外类入侵度初始化参数的极限学习机
分享到：

南京师范大学学报（工程技术版）[ISSN:1006-6977/CN:61-1281/TN]

卷:
19卷
期数:
2019年03期
页码:
053
栏目:
计算机工程
出版日期:
2019-09-30

文章信息/Info

Title:
Extreme Learning Machine with Initialized ParametersBased on External Class Invasion Degree
文章编号:
1672-1292(2019)03-0053-06
作者:
孔双双1; 2; 王开军1; 2; 林 崧1; 2
(1.福建师范大学数学与信息学院,福建 福州 350117)(2.福建师范大学数字福建环境监测物联网实验室,福建 福州 350117)
Author(s):
Kong Shuangshuang1; 2; Wang Kaijun1; 2; Lin Song1; 2
(1.School of Mathematics and Information,Fujian Normal University,Fuzhou 350117,China)(2.Digital Fujian Environmental Monitoring Internet of Things Laboratory,Fujian Normal University,Fuzhou 350117,China)
关键词:
极限学习机; 重叠区域; 类入侵度; 权值修正
Keywords:
extreme learning machine; overlapping area; external class invasion degree; weight correction
分类号:
TP183
DOI:
10.3969/j.issn.1672-1292.2019.03.008
文献标志码:
A
摘要:
针对经典极限学习机中输入权值随机初始化容易导致输出稳定性不够好进而影响分类性能的问题,提出外类入侵度初始化参数的方法,对极限学习机随机初始化的输入权值用样本的属性特征信息进行修正. 该方法对包含两个类别样本的数据集,将其中一个类作为本类,另一个类作为外类. 对于每个特征,统计本类和外类样本重叠的区域占本类取值范围的比例,也统计重叠区域中外类样本数目占重叠区域总样本数目的比例. 然后依据这两种占比值计算每个特征的外类入侵度. 再根据入侵度大小调整极限学习机模型中隐含层的输入权值. 在10个UCI数据集上进行的分类实验结果表明,新方法的准确率比经典极限学习机提高了1%～23%,且泛化性能更稳定; 与另外两方法相比,新方法的准确率稍高.
Abstract:
Aiming at the problem that the random initialization of the input weights in the classical extreme learning machine which is easy to cause the output stability is not good enough and then affects the classification performance,an extreme learning machine with initialized parameters based on external class invasion degree is proposed by adjusting the weights initialized randomly of the extreme learning machine through the attribute characteristic information of samples. For the data set containing samples of two different categories,the method takes one class as this class and the other class as outer class. For each feature,the proportion of the overlapping areas between the samples of this class and outer class in the value range of this class is calculated,and the proportion of outer class samples in the total number of samples in the overlapping areas is also calculated. Then the invasion degree of each feature according to the two proportions is calculated. Finally,the input weights of the hidden layer of the extreme learning machine model is adjusted according to the degree of invasion. Experimental results on ten UCI data sets show that the accuracy of the new method is 1%～23% higher than that of the classical extreme learning machine,and the generalization performance is more stable; and that compared with other two methods,the accuracy of the new method is slightly higher.

参考文献/References:

[1] HUANG G B,ZHU Q Y,SIEW C K. Extreme learning machine:theory and applications[J]. Neurocomputing,2006,70(1):489-501.
[2]HUANG G B,ZHOU H M,DING X J,et al. Extreme learning machine for regression and multiclass classification[J]. IEEE transactions on systems,man and cybernetics,2011,42(2):513-529,2012.
[3]LIU B,YAN S,YOU H L,et al. Road surface temperature prediction based on gradient extreme learning machine boosting[J]. Coputers in industry,2018,99:294-302.
[4]裘日辉,刘康玲,梁军. 基于极限学习机的分类算法及在故障识别中的应用[J]. 浙江大学学报(工学版),2016,50(10):1965-1972.
QIU R H,LIU K L,LIANG J. Classification algorithm based on extreme learning machine and its application in fault identification of Tennessee Eastman process[J]. Engineering of journal of Zhejiang university,2016,50(10):1965-1972.(in Chinese)
[5]HUANG G B,LEI C,SIEW C K. Unversal approximation using incremental constructuve feedforward networks with random hidden nodes[J]. IEEE transactions on neural networks,2006,17(4):879-892.
[6]WANG R,SAN K,WANG X Z. A study on random weights between input and hidden layers in extreme learning machine[J]. Soft computing,2012,16(9):1465-1475.
[7]ZHANG X F,LIN X L,ASHFAQ. Impact of different random initializations on generalization performance of extreme learning machine[J]. Journal of computers,2018,13(7):805-821.
[8]CAO J W,LIN Z P,HUANG G B,et al. Voting based extreme learning machine[J]. Information sciences,2012,185(1):66-77.
[9]PACHECO A G C,KROHLING R A,SILVA C A S,et al. Restricted Boltzmann machine to determine the input weights for extreme learning machines[J]. Expert system with application,2018,96:77-85.
[10]WANG W H,LIU X Y. The selection of input weights of extreme leaning machine:a sample structure preserving point of view[J]. Neurocomputing,2017,261:28-36.
[11]WANG Y G,CAO F L,YUAN Y B. A study on effectiveness of extreme learning machine[J]. Neurocomputing,2011,74(16):1483-1490.
[12]RAKHA,MEDHAT A. On the Moore-Penrose generalized inverse matrix[J]. Applied mathematics and computation,2004,158(1):185-200.
[13]BARTLETT,PETER L. The sample complexity of pattern classification with neural networks:the size of the weights is more important than the size of the network[J]. IEEE transactions on information theory,1998,44(2):525-536.
[14]BLAKE C L,MERZ C J. UCI Repository of machine learning databases[EB/OL]. [2019-3-20]:https://archive.ics.uci.edu/ml/index.php.
[15]HUANG G B. ELM source codes[EB/OL]. [2019-3-20] http://www.ntu.edu.sg/home/egbhuang/.
[16]PAHCECO A G C. RBM-ELM source codes[EB/OL]. [2019-3-20] http://github.com/paaatcha/RBM-ELM.
[17]HAN J W,MICHELINE K,JIAN P,等. 数据挖掘:概念与技术[M]. 北京:机械工业出版社,2012.HAN J W,MICHELINE K,JIAN P,et al. Data mining:concepts and techniques[M]. Beijing:China Machine Press,2012.(in Chinese)

备注/Memo

备注/Memo:
收稿日期:2019-07-05.
基金项目:福建省自然科学基金(2018J01778)、国家自然科学基金(61772134)、博士后基金(2016M600494).
通讯联系人:王开军,博士,副教授,研究方向:机器学习、数据挖掘等. E-mail:wkjwang@qq.com
更新日期/Last Update: 2019-09-30