吉林大学学报(理学版) ›› 2021, Vol. 59 ›› Issue (5): 1212-1218.

• • 上一篇    下一篇

一种新型的自适应多核学习算法

聂逯松1, 常方圆2, 常学智3, 刘畅4, 金有为5,6,7, 刘国晟5,6,7, 付加胜8, 韩霄松5,6,7   

  1. 1. 吉林大学 共青团吉林大学委员会, 长春 130012; 2 吉林大学 护理学院, 长春 130012;
    3. 吉林大学 生物与农业工程学院, 长春 130022; 4. 长春中医药大学  医药信息学院, 长春 130117;
    5. 吉林大学 符号计算与知识工程教育部重点实验室 长春 130012; 6. 吉林大学 计算机科学与技术学院, 长春 130012;  7. 吉林大学 软件学院, 长春 130012;8. 中国石油集团工程技术研究院有限公司, 北京 102206
  • 收稿日期:2021-01-25 出版日期:2021-09-26 发布日期:2021-09-26
  • 通讯作者: 韩霄松 E-mail:hanxiaosong@jlu.edu.cn

A Novel Self-adaptive Multiple Kernel Learning Algorithm

NIE Lusong1, CHANG Fangyuan2, CHANG Xuezhi3, LIU Chang4, JIN Youwei5,6,7, LIU Guosheng5,6,7, FU Jiasheng8, HAN Xiaosong5,6,7   

  1. 1. Jilin University Communist Youth League Committee, Jilin University, Changchun 130012, China;
    2. College of Nursing, Jilin University, Changchun 130012, China;
    3. College of Biological and Agricultural Engineering, Jilin University,  Changchun 130022, China;
    4. College of Medical Information, Changchun University of Chinese Medicine,  Changchun 130117, China;
    5. Key Laboratory for Symbol Computation and Knowledge Engineering of National Education Ministry, Changchun 130012, China; 
    6. College of Computer Science and Technology, Jilin University, Changchun 130012, China;
    7. College of Software, Jilin University, Changchun 130012, China;  8. CNPC Engineering Technology R&D Company Limited, Beijing 102206, China
  • Received:2021-01-25 Online:2021-09-26 Published:2021-09-26

摘要: 针对样本基数较大、 维数较高、 特征较复杂的数据集训练问题, 将支持向量机与蚁群优化算法相融合, 提出一种自适应多核学习算法. 利用吸引子传播聚类算法自适应地发现相似特征, 并据此利用蚁群算法自适应地选择核函数参数, 从而快速选择最优核函数. 通过UCI数据集的5组数据实验表明, 该算法相比于传统的支持向量机分类准确率和F1值更高, 验证了该算法的有效性和可行性.

关键词: 多核学习, 支持向量机, 蚁群算法, 聚类算法

Abstract: Aiming at the problem of the training data set with large samples, high dimension, and complex features, a self-adaptive multiple kernel learning algorithm was proposed by integrating support vector machine with ant colony optimization algorithm. The affinity propagation clustering algorithm was used to find the similar features adaptively, and then the parameters of the kernel function were selected adaptively by ant colony algorithm, so as to select the optimal kernel function quikly. Experimental results of five groups of UCI data sets show that the proposed algorithm has higher classification accuracy and F1 value than the traditional support vector machine, which verifies the effectiveness and feasibility of the proposed algorithm.

Key words: multiple kernel learning, support vector machine, ant colony algorithm, clustering algorithm

中图分类号: 

  • TP391.1