外类入侵度初始化参数的极限学习机  

Extreme Learning Machine with Initialized Parameters Based on External Class Invasion Degree

作　　者：孔双双 王开军 林崧[1,2] Kong Shuangshuang;Wang Kaijun;Lin Song(School of Mathematics and Information,Fujian Normal University,Fuzhou 350117,China;Digital Fujian Environmental Monitoring Internet of Things Laboratory,Fujian Normal University,Fuzhou 350117,China)

机构地区：[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

出　　处：《南京师范大学学报（工程技术版）》2019年第3期53-58,共6页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：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)

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.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.

关 键 词：极限学习机 重叠区域 类入侵度 权值修正 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]