Granular multiple birth support vector machine based on weighted linear loss

Recently proposed multiple birth support vector machine (MBSVM), which is extended from TWSVM, is an efficient algorithm for multi-class classification. MBSVM keeps the advantage of TWSVM. However, the solution of MBSVM classifier follows solving quadratic programming. Solving quadratic programming...

Full description

Saved in:
Bibliographic Details
Main Authors: Shifei Ding, Xiekai Zhang
Format: Conference Proceeding
Language:English
Subjects:
Computational modeling
Computer science
Fasteners
granular computing
Information processing
multi-category classification
Quadratic programming
support vector machine
Support vector machines
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently proposed multiple birth support vector machine (MBSVM), which is extended from TWSVM, is an efficient algorithm for multi-class classification. MBSVM keeps the advantage of TWSVM. However, the solution of MBSVM classifier follows solving quadratic programming. Solving quadratic programming requires long time. This paper presents a granular multiple birth support vector machine based on weighted linear loss (WLGMBSVM) to enhance the performance of MBSVM classifier. WLGMBSVM uses the strategy of "all-versus-one" as MBSVM does. By introducing the weighted linear loss, the proposed algorithm only needs to solve simple linear equations. Inspired by granular support vector machine, WLGMBSVM handles classification in granules. The experiments also show the efficiency of WLGMBSVM.
ISSN:2161-4407
DOI:10.1109/IJCNN.2016.7727504