基于可缩放hinge损失的支持向量数据描述  

Support Vector Data Description Based on Rescaled Hinge Loss

作　　者：王余波 胡文军[1,2,3] 王士同 WANG Yubo;HU Wenjun;WANG Shitong(School of Information Engineering,Huzhou University,Huzhou 313000,China;Zhejiang Province Key Laboratory of Smart Management&Application of Modern Agricultural Resources,Huzhou 313000,China;Huzhou Key Laboratory of Waters Robotics Technology,Huzhou University,Huzhou 313000,China;School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214122,China)

机构地区：[1]湖州师范学院信息工程学院,浙江湖州313000 [2]浙江省现代农业资源智慧管理与应用研究重点实验室,浙江湖州313000 [3]湖州市水域机器人技术重点实验室,浙江湖州313000 [4]江南大学人工智能与计算机学院,江苏无锡214122

出　　处：《湖州师范学院学报》2024年第8期36-46,共11页Journal of Huzhou University

基　　金：国家自然科学基金项目(U20A20228,61772198)。

摘　　要：支持向量数据描述(SVDD)极易受到异常值的影响,导致其鲁棒性不佳.利用可缩放的hinge损失函数,提出一种新的支持向量数据描述方法(RH-SVDD).将有界可缩放的hinge损失作为松弛变量构建超球模型;通过共轭函数理论将超球模型转化为凸优化问题;利用半二次优化技术迭代求解凸优化问题,并在迭代过程中实现权重更新,从而削弱异常值的影响,提升鲁棒性.实验结果表明,提出的RH-SVDD在分类任务中具有更好的性能优势.Support Vector Data Description(SVDD)is highly susceptible to outliers,resulting in poor robustness.To this end,a new support vector data description method(RH-SVDD)is proposed using the rescaled hinge loss function.First,the bounded rescaled hinge loss is used as slack variables to construct a hypersphere model;secondly,the hypersphere model is transformed into a convex optimization problem through the conjugate function theory;finally,the convex optimization problem is solved iteratively using half-quadratic optimization technology,and the weights are updated during the iteration process,thereby weakening the impact of outliers and improving robustness.Experimental results show that the proposed RH-SVDD has better performance advantages in classification tasks.

关 键 词：支持向量数据描述 可缩放hinge损失 半二次优化 鲁棒性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]