Online multiple instance regression

The multiple instance regression problem has become a hot research topic recently. There are several approaches to the multiple instance regression problem, such as Salience, Citation KNN, and MI-ClusterRegress. All of these solutions work in batch mode during the training step. However, in practice...

Full description

Saved in:
Bibliographic Details
Published inChinese physics B Vol. 22; no. 9; pp. 656 - 661
Main Author 王志岗 赵增顺 张长水
Format Journal Article
LanguageEnglish
Published 01.09.2013
Subjects
KNN
Labels
Mathematical models
Online
Regression
Training
回归模型
回归法
回归问题
在线
多实例
批处理
数据集
Online AccessGet full text

Cover

More Information
Summary:The multiple instance regression problem has become a hot research topic recently. There are several approaches to the multiple instance regression problem, such as Salience, Citation KNN, and MI-ClusterRegress. All of these solutions work in batch mode during the training step. However, in practice, examples usually arrive in sequence. Therefore, the training step cannot be accomplished once. In this paper, an online multiple instance regression method ＂OnlineMIR＂ is proposed. OnlineMIR can not only predict the label of a new bag, but also update the current regression model with the latest arrived bag. The experimental results show that OnlineMIR achieves good performances on both synthetic and real data sets.
Bibliography:The multiple instance regression problem has become a hot research topic recently. There are several approaches to the multiple instance regression problem, such as Salience, Citation KNN, and MI-ClusterRegress. All of these solutions work in batch mode during the training step. However, in practice, examples usually arrive in sequence. Therefore, the training step cannot be accomplished once. In this paper, an online multiple instance regression method ＂OnlineMIR＂ is proposed. OnlineMIR can not only predict the label of a new bag, but also update the current regression model with the latest arrived bag. The experimental results show that OnlineMIR achieves good performances on both synthetic and real data sets.
mutiple instance, regression, online learning
11-5639/O4
Wang Zhi-Gang, Zhao Zeng-Shun, and Zhang Chang-Shui（ a） Department of Automation, Tsinghua University, State Key Laboratory of Intelligent Technologie and Systems, Tsinghua National Laboratory for Information Science and Technology, Beijing 100084, China b） College of Information and Electrical Engineenng, Shandong University of Science and Technology, Qingdao 266590, China c） School of Control Science and Engineering, Shandong University, Jinan 250061, China
ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:1674-1056
2058-3834
1741-4199
DOI:10.1088/1674-1056/22/9/098702