[1]黄华娟,韦修喜,周永权.光滑孪生参数化不敏感支持向量回归机[J].郑州大学学报(工学版),2022,43(02):28-34.[doi:10.13705/j.issn.1671-6833.2022.02.005]
 Huang Huajuan,Wei Xiuxi,Zhou Yongquan,et al.Smooth Twin Parametric Insensitive Support Vector Regression[J].Journal of Zhengzhou University (Engineering Science),2022,43(02):28-34.[doi:10.13705/j.issn.1671-6833.2022.02.005]
点击复制

光滑孪生参数化不敏感支持向量回归机()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
43卷
期数:
2022年02期
页码:
28-34
栏目:
出版日期:
2022-02-27

文章信息/Info

Title:
Smooth Twin Parametric Insensitive Support Vector Regression
作者:
黄华娟韦修喜周永权
广西民族大学人工智能学院;广西民族大学广西混杂计算与集成电路设计分析重点实验室;

Author(s):
Huang Huajuan; Wei Xiuxi; Zhou Yongquan;
Guangxi University for Artificial Intelligence Institute; Guangxi University of Ethnological University Guangxi Mixed Computing and Integrated Circuit Design Analysis Laboratory;

关键词:
Keywords:
DOI:
10.13705/j.issn.1671-6833.2022.02.005
文献标志码:
A
摘要:
作为机器学习方法之一的孪生参数化不敏感支持向量回归机(Twin parametric insensitive support vector regression, TPISVR)有着简洁的数学模型,良好的学习性能,特别适合于求解带有结构异方差噪声的数据回归问题。然而,TPISVR的训练速度较低,训练效率有待提高。在本文中,引入光滑函数和正则项,将TPISVR的数学模型转化为两个无约束的极小化问题,从而可以通过具有快速求解能力的Newton法进行求解,提出光滑孪生参数化不敏感支持向量回归机(Smooth twin parametric insensitive support vector regression, STPISVR)。在人工数据集和UCI数据集上的实验结果表明,和其他机器学习方法相比,STPISVR在保证精度不下降的情况下,获得了更高的训练效率。
Abstract:
Twin parametric insensitive support vector regression (TPISVR) is a recently proposed machine learning method. Compared with other machine learning algorithms, it has unique advantages in dealing with heteroscedastic noise. However, the mathematical model of TPISVR is to solve two quadratic programming problems with multiple inequality constraints, which costs more training time. In order to improve the training speed of TPISVR, smooth twin parametric insensitive support vector regression (STPISVR) is proposed by introducing smooth technology. In STPISVR, two constrained quadratic programming problems of TPISVR are transformed into two unconstrained minimization problems, which can be solved by Newton method with fast solving ability, thus improving the training speed of the algorithm. It is proved theoretically that STPISVR has strict convexity and can satisfy the performance of any order smoothness. Experimental results on artificial data set and UCI data set show that compared with TPISVR, STPISVR can achieve higher training efficiency without reducing the accuracy
更新日期/Last Update: 2022-02-25